#  @MD71085 MD MD posts on X about ai, $nvda, nvda, in the the most. They currently have [-----] followers and [---] posts still getting attention that total [-------] engagements in the last [--] hours. ### Engagements: [-------] [#](/creator/twitter::230912559/interactions)  - [--] Week [------] -90% - [--] Month [-------] +111% - [--] Months [---------] +507% - [--] Year [---------] +4,039% ### Mentions: [--] [#](/creator/twitter::230912559/posts_active)  - [--] Week [--] -48% - [--] Month [--] +46% - [--] Months [---] +431% - [--] Year [---] +482% ### Followers: [-----] [#](/creator/twitter::230912559/followers)  - [--] Week [-----] +0.83% - [--] Month [-----] +15% - [--] Months [-----] +645% - [--] Year [-----] +1,089% ### CreatorRank: [---------] [#](/creator/twitter::230912559/influencer_rank)  ### Social Influence **Social category influence** [stocks](/list/stocks) [technology brands](/list/technology-brands) [finance](/list/finance) [countries](/list/countries) [automotive brands](/list/automotive-brands) [currencies](/list/currencies) [social networks](/list/social-networks) [cryptocurrencies](/list/cryptocurrencies) [exchanges](/list/exchanges) [celebrities](/list/celebrities) **Social topic influence** [ai](/topic/ai), [$nvda](/topic/$nvda), [nvda](/topic/nvda), [in the](/topic/in-the), [$googl](/topic/$googl), [$nbis](/topic/$nbis), [$crwv](/topic/$crwv), [if you](/topic/if-you), [open ai](/topic/open-ai), [inference](/topic/inference) **Top assets mentioned** [NVIDIA Corp. (NVDA)](/topic/$nvda) [Alphabet Inc Class A (GOOGL)](/topic/$googl) [Nebius Group N.V. Class A Ordinary Shares (NBIS)](/topic/$nbis) [CoreWeave, Inc (CRWV)](/topic/$crwv) [Oracle Corporation (ORCL)](/topic/$orcl) [Microsoft Corp. (MSFT)](/topic/$msft) [Advanced Micro Devices (AMD)](/topic/$amd) [Metadium (META)](/topic/$meta) [Hut [--] Mining Corp. (HUT)](/topic/$hut) [Tesla, Inc. (TSLA)](/topic/$tsla) [Amazon.com, Inc. (AMZN)](/topic/$amzn) [Apple, Inc. (AAPL)](/topic/$aapl) [Arm Holdings plc (ARM)](/topic/$arm) [Broadcom, Inc. (AVGO)](/topic/$avgo) [Intel Corporation (INTC)](/topic/$intc) ### Top Social Posts Top posts by engagements in the last [--] hours "@PeteButtigieg Its called supply chains normalizing (which they did everywhere in the world duh) and the fastest rate hike cycle ever. You should know this Mr Transport Secretary. @DanielSLoeb1" [X Link](https://x.com/MD71085/status/1839463425858027693) 2024-09-27T00:33Z [---] followers, [--] engagements "@TotemMacro Recesionista. Whitney Baker on why a recession is the only thing that will stunt consumption and thus slow inflation - [----]. This view has aged almost as well as you :) #macro $qqq" [X Link](https://x.com/MD71085/status/1840384520572305602) 2024-09-29T13:33Z [---] followers, [---] engagements "@TotemMacro Looks like your view then was inflation wouldnt come down without a recession. If so that has not aged well. If Im mistaken my apologies" [X Link](https://x.com/MD71085/status/1840546794683629666) 2024-09-30T00:18Z [---] followers, [--] engagements "I suspect capex for the big guys will be around 40-50% higher than the street next year so the debate isnt going anywhere. The driver of that is a new concept called Inference-time scaling which is the backbone of the new Strawberry model. Others arent far behind. Inference quality scales with compute (like training) and is equally complex (need NVDA GPUs). It essentially pulls forward the agentic use case (think Jarvis to Tony Stark in iron man). Whether anyone gets there who knows but if you are one of the megas and you thought it had a chance of being possible what wouldnt you spend" [X Link](https://x.com/DratchCap/status/1847769876154503431) 2024-10-19T22:40Z [---] followers, [--] engagements "AMD chips are losing share. Customers dont like them. And the nature of inference changed from retrieval to time-scaling which is much more power intensive and akin to training. This is NEW information and makes the old thesis around AMD and inference opportunity less compelling. If you dont know what inference time scaling is then you dont really know much about the current hardware and AI backdrop. The comment eke out more chips than is possible also suggests you dont understand the nature of the engineering problem. And NVDA is producing to meet existing demand not to create it. Good luck" [X Link](https://x.com/MD71085/status/1858241709815345505) 2024-11-17T20:12Z [---] followers, [--] engagements "They could easily sell the gold in a few quarters (over decades) instead of issuing new (non roll over) treasury securities this would erase most of the monetary tightening exercise you cite as the reason. The only net difference would be the lost interest income to former treasury holders (roughly $35bn in a [--] Trillion economy). Which is to say that is a silly reason not to sell the gold" [X Link](https://x.com/MD71085/status/1863029265148805588) 2024-12-01T01:16Z [---] followers, [--] engagements "@dampedspring So stocks down big bonds sideways And yet also calling for lower NGDP e.g. growth scare Haha ok" [X Link](https://x.com/MD71085/status/1878154753277919271) 2025-01-11T18:59Z [---] followers, [----] engagements "@WarrenTStephens @dampedspring I was pointing out the logical inconsistency not the timing" [X Link](https://x.com/MD71085/status/1878242182080991415) 2025-01-12T00:46Z [---] followers, [---] engagements "@WarrenTStephens @dampedspring I like that he never qualified the size of the upside down V despite the connotations allows freedom to re write history if need be. " [X Link](https://x.com/MD71085/status/1878295609234374914) 2025-01-12T04:19Z [---] followers, [---] engagements "@japdongsanee Clickhouse is marked at $4Bn valuation and likely worth much more. Their 28% stake covers the covers biz with ease. You get toloka AV Ride free" [X Link](https://x.com/MD71085/status/1879374746741248212) 2025-01-15T03:47Z [---] followers, [--] engagements "DumbSeek is lying. Come on @DanielTNiles DeepSeek trained a model for 6mm but meta and musk and Amzn etc spent 10s of billions. Makes sense. OR DumbSeek has access to well over 50k H100s and will not say that (since they shouldnt have access due to export controls) AND they piggybacked on other foundational models (hence hallucinations). The CCP is also surely involved and would love a narrative that China spends $6mm on AI to build a frontier model that otherwise costs billions. This is classic Chinese propaganda and innovation (I.e. stealing / piggy backing US tech). That you have a" [X Link](https://x.com/MD71085/status/1882926106875215945) 2025-01-24T22:59Z [---] followers, [---] engagements "@RihardJarc $Nvda inference dominance will increase as context windows lengthen requiring system level approaches to maximize efficiency (more memory + more connected GPUs). Complex inference is also power intensive and so better suited for GPUs" [X Link](https://x.com/MD71085/status/1883057012663263513) 2025-01-25T07:39Z [---] followers, [---] engagements "I dont. Just being realistic about the cost of training a model today and acknowledging that a Chinese HF is unlikely to report h100 costs given export controls (implying they have materially negative and unscalable gross margins). There is also a big difference between training cost and the total cost of AI compute tuning reinforcement learning synthetic data generation mass inference time scaling all of which act in fly wheel form and compound each other" [X Link](https://x.com/MD71085/status/1883183524343140826) 2025-01-25T16:01Z [---] followers, [---] engagements "@kashyap286 @hkuppy @PrateekGsharma So they do have H100s If so the cost is much higher than they claim" [X Link](https://x.com/MD71085/status/1883185527828185502) 2025-01-25T16:09Z [---] followers, [---] engagements "In a few years. Gemini was trained for $150bn but Google spends $60bn on capex theres a difference between training cost and cost to run a scaled AI that can service large scale and real time inference. People seem to be thinking $5mm training = tall the big tech spend was a waste This is wrong and comparing apples and oranges. Also Jevons paradox" [X Link](https://x.com/MD71085/status/1883265572625891563) 2025-01-25T21:27Z [---] followers, [--] engagements "@JonBryant421 Well no different than if you had a humanoid / robot driver their eyes would not be much different than a human driver. Point being IF that were to emerge it would be a more scalable system and more akin to how humans drive" [X Link](https://x.com/MD71085/status/1887668547741995479) 2025-02-07T01:03Z [---] followers, [--] engagements "@natejhake Google is killing the web And squeezing people. Reminds me of net neutrality killing the internet. Its a free market the ads run on auctions if roas sucks people move on. Pretty simple. I dont run ads I run a hedge fund" [X Link](https://x.com/MD71085/status/1921362142575370726) 2025-05-11T00:30Z [---] followers, [--] engagements "@rev_cap Who is moodys 😉" [X Link](https://x.com/MD71085/status/1923584992124236210) 2025-05-17T03:42Z [---] followers, [---] engagements "@industrialpc @rev_cap Since we lost the other two everything has gone to hell 😉😂. And they dont decide where anyone puts money" [X Link](https://x.com/MD71085/status/1923719804327805326) 2025-05-17T12:38Z [---] followers, [--] engagements "@benitoz OpenAI is experiencing a token demand boom and is short GPUs. Google TPU pods are one of the few immediately available options since they cant speed up more Blackwell or Rubin. They arent doing it to diversify they are doing it to serve demand *today* $nvda $googl" [X Link](https://x.com/MD71085/status/1938792629010940135) 2025-06-28T02:52Z [---] followers, [--] engagements "@meeijer Clickhouse is the most valuable and best non core asset but no mention 28% stake just raised capital at $6bn. Otherwise known as the Snowflake killer" [X Link](https://x.com/MD71085/status/1949315369660817536) 2025-07-27T03:46Z [---] followers, [--] engagements "We need @sydney_sweeney to recognize the moment dawn a MAGA hat and post: The only thing as good as my jeans are my hats $AEO. Properly positioned she would make more money in the next 24hrs than her next few movies. Just sayin" [X Link](https://x.com/MD71085/status/1952404535307121000) 2025-08-04T16:21Z [---] followers, 14.3K engagements "5/ $CRWV $NBIS and powered-shell stocks ($HUT $RIOT etc) should have been up 50% on this signal. That they werent suggests we still have a MOUNTAIN of SKEPTICISM to climb. Bullish indeed. (5/5)" [X Link](https://x.com/DratchCap/status/1970331599335231514) 2025-09-23T03:37Z [---] followers, [---] engagements "@coinbase @AmericanExpress You can send money to yourself for 2.95% + 30c processing fee. Clip 80bps. Forget $NVDA hows that for circular schemes" [X Link](https://x.com/DratchCap/status/1970964664818495772) 2025-09-24T21:32Z [---] followers, [--] engagements "$NBIS is much earlier stage than $CRWV and is also exposed to spot compute which may go up given power constraints while CRWV is not. NBIS is also likely to spend $8-10bn minimum next year thats before making purchases for [--] new sites. Also management is better at NBIS + more ways to win given the differences in biz plan and non-core assets. But I think this is and not either / or. (Note that Magnetar is almost fully out of CRWV. I.e. the seller is almost done)" [X Link](https://x.com/DratchCap/status/1973552471667532073) 2025-10-02T00:55Z [---] followers, [---] engagements "@altcap @CerebrasSystems @GroqInc @nvidia Interesting. How much Cowos capacity have they secured for next 3yrs" [X Link](https://x.com/DratchCap/status/1974611692366757972) 2025-10-04T23:04Z [---] followers, [---] engagements "@ROLCapital @SemiAnalysis_ has a good slide somewhere on $msft ROI annualizing close to 25% in 2Q25 on Ai spend" [X Link](https://x.com/DratchCap/status/1975024006601584730) 2025-10-06T02:23Z [---] followers, [--] engagements "Musing #2. Macro focused on inflation/debt levels while we witness the real-time birth of a productivity boom. Markets love irony indeed" [X Link](https://x.com/DratchCap/status/1976447385913241672) 2025-10-10T00:39Z [---] followers, [---] engagements "Musing #3: A lot of fuss about debt levels but what do people expect folks to do with their excess dollar *savings* (Debt = savings duh). Put it ALL in stocks One can dream I guess" [X Link](https://x.com/DratchCap/status/1976447837736206555) 2025-10-10T00:40Z [---] followers, [---] engagements "@thinknonlinear1 Govt doesnt pre-empt M&A. USAR is pre-close on LCM deal and in talks (per the chairman). If support comes its post-close/milestones via Title III/DOE/offtakes. Btw this is the exact same approach & cadence used with MP/Lynas" [X Link](https://x.com/DratchCap/status/1976819015579562027) 2025-10-11T01:15Z [---] followers, [--] engagements "The only circular deal that matters: Accelerating Deflationary Tech Boom = Cuts = Productivity Acceleration = More Cuts. P.S. the tech is a cannon pointed at inflation. Just look around. $NVDA $AMD $NBIS $CRWV $HUT $BE" [X Link](https://x.com/DratchCap/status/1978142267882745972) 2025-10-14T16:54Z [---] followers, [---] engagements "I am a bit surprised that the CIO with 15yrs experience covering $TSM hasnt noticed the pattern of them guiding up literally every 4Q call. TSM capacity is currently maxed you can see this in ASE / $amkr demand for cowos (they get spillover). But alas without tourists my job would be harder :) $NVDA $nbis" [X Link](https://x.com/DratchCap/status/1979198417638428900) 2025-10-17T14:50Z [---] followers, [---] engagements "@JamieFollese @ShanuMathew93 @zephyr_z9 @The_Colonel__ What GPUs are down 2/3 in price recently Any evidence" [X Link](https://x.com/DratchCap/status/1980334356699967615) 2025-10-20T18:04Z [---] followers, [--] engagements "@JamieFollese @ShanuMathew93 @zephyr_z9 @The_Colonel__ Availability for all compute is touching new lows. Amzn raising prices on H100s. Yawn. @WarrenPies has some great charts for you and tracks compute availability across the chain. But look anywhere bruh" [X Link](https://x.com/DratchCap/status/1980339296424034444) 2025-10-20T18:24Z [---] followers, [--] engagements "@JamieFollese @ShanuMathew93 @zephyr_z9 @The_Colonel__ Btw pretty much all CRWV rev is locked in contracts they build the IRR into the long term rental price and own upside in residual value. Which the evidence for is apparently growing btw hence availability on h100s heading DOWN @NarrativeSeeker has done some good work here" [X Link](https://x.com/DratchCap/status/1980340428919705825) 2025-10-20T18:28Z [---] followers, [---] engagements "The cost of inference is collapsing the revenue yielded over the lifetime of a GPU is not. Note as well NVDA backstopped all unsold capacity and new capacity is built on the back of new demand and contracts priced for a lifetime ROI. Btw I heard theres businesses called GCP and Azure that really just rent computers off premises. Same business high level (and yes even CRWV is providing services on top of steel) same hate in the early phases of the businesses and now everyone calls them the best. Weird" [X Link](https://x.com/DratchCap/status/1980350973349367815) 2025-10-20T19:10Z [---] followers, [--] engagements "The Google has cheaper AI is a big generalization and I say this as a big fan of $Googl stock. It depends on throughput latency (including p99) time-to-first token and most importantly *usable* context windows (not the boilerplate). Which is really just a fancy way of saying it depends $nvda $amd" [X Link](https://x.com/DratchCap/status/1980389931072504176) 2025-10-20T21:45Z [---] followers, [---] engagements "@JMihaljevic Good analysis Theyre mixing things that dont really go together. Its like buying a $1 ice cream cone and then adding in the cost of the building ice cream machines delivery trucks and shop. And pretending all that was just for your one cone. Silly. $nvda" [X Link](https://x.com/DratchCap/status/1980461882814369938) 2025-10-21T02:31Z [---] followers, [---] engagements "@firstadopter Alpha hides where conviction is loudest and understanding is thinnest. Thats large-cap tech and AI today. (Hint: Rorschach test 😝😉) $nvda $nbis $tsla $amd" [X Link](https://x.com/DratchCap/status/1980770300221428095) 2025-10-21T22:56Z [---] followers, [---] engagements "@The_AI_Investor Power constrained today = Compute Constrained when the MWs light up. More power = more supply = lower prices = more demand. Same as it ever was. $NVDA" [X Link](https://x.com/DratchCap/status/1985126927733031409) 2025-11-02T23:28Z [---] followers, [---] engagements "Do you even read the things you repost Youre like a FUD troll. Read the whole document yourself. He doesnt ask for a bailout for OpenAi. But he does suggest the government should help lower the cost of capital for the ecosystem to encourage investment *across the chain* including power development. See below. 🤡🤡 @iamgingertrash He doesnt say for them specifically and hes talking about the ecosystem. For example the govt might backstop debt against turbine facilities to lower cost for companies unwilling to take a risk on building (long cycle) capacity in the USA. @iamgingertrash He doesnt" [X Link](https://x.com/DratchCap/status/1986980919123526119) 2025-11-08T02:15Z [---] followers, [--] engagements "Sams plan for $1.4T / 30GW hit the tape on 10/28-9. Its implausible the market had already fully priced a 1.4T plan from a 3-yr-old $1320B-rev company. What was priced was some OpenAI build-out (ORCL MSFT capex NVDA orders) and a much fuzzier multi-trillion global AI infra story. Prices would have been much higher if we were pricing in additional 1.4T from just OpenAI let alone everyone else. Instead its Nov people have PNL to defend prices go down narratives form. One player (Sam) after an undoubtedly positive period of pos demand data points from literally everyone else in the space says IF" [X Link](https://x.com/DratchCap/status/1990118633755017416) 2025-11-16T18:03Z [---] followers, [--] engagements "I tried hedges. I tried degrossing. Now Im trying Im music. When the Selling Stops $nvda $nbis $crwv @elevenlabsio" [X Link](https://x.com/DratchCap/status/1990638195038707864) 2025-11-18T04:28Z [---] followers, [----] engagements "I like $GOOGL (and Gemini 3) but this narrative emerging on X that Gemini is charging more bc the model is better is way off and NOT how the market actually works. Rate cards are a bit theatrical (deals dont clear there). Most API usage is bundled and token pricing is [--] small knob. They also list high prices for 1) perception (hence the X reaction) and 2) to enable discounts (its a sales business after all come on guys duh). Finally Gemini [--] is a serious context heavy multi-modal model that requires a lot of compute / memory. Higher starting pricing also helps ration compute in a constrained" [X Link](https://x.com/DratchCap/status/1990994432766713907) 2025-11-19T04:03Z [---] followers, [---] engagements "Macro man is apparently very bad at micro analysis. This post from @paulomarco on $NVDA is emblematic of the really bad takes from tourists on AI that are pervasive across X. One of my rules is control your sources because *bad research is expensive* (good research pays for itself. To be fair he may indeed have excellent macro ideas (or in this case was just jumping on a quick hot take). That said Last years DSO (days sales outstanding; we cant assume you know what that means) was artificially suppressed by a large $1.7Bn prepayment. NVDA explicitly states this in last years 10-Q. And DSO is" [X Link](https://x.com/DratchCap/status/1991331941572878457) 2025-11-20T02:25Z [---] followers, [----] engagements "@PauloMacro Not mine. You did not address the increase in prepayments and you still point out DSO increasing y/y but dont acknowledge its up vs a specific instance in the company called out in their filing 1yr ago. Nice try" [X Link](https://x.com/DratchCap/status/1991351421263049031) 2025-11-20T03:42Z [---] followers, [---] engagements "Just for everyone to see. This clown show could not respond at all to my criticism of his point ignoredthat $NVDA saw increase pre-payments in the quarter AND he restated that DSO was up y/y despite the my showing him NVDA gave an explanation for 1x suppression of DSO last year. Just a lot of hand waving. I pointed to his track record a simple Grok search of his posts will show you hes been bearish $NVDA and AI AND all stocks since [----]. Yet he has a paid substack and 175k followers (in a sane world they all know hes a contra). You get OUT what you put IN so vigilantly control your sources." [X Link](https://x.com/DratchCap/status/1991354704614326307) 2025-11-20T03:55Z [---] followers, [--] engagements "Just for everyone to see. This clown show could not respond at all to my criticism of his point ignored that $NVDA saw increased pre-payments in the quarter AND he restated that DSO was up y/y despite my showing him NVDA gave an explanation for 1x suppression of DSO last year. Just a lot of hand waving. I pointed to his track record a simple Grok search of his posts will show you hes been bearish $NVDA and AI AND all stocks since [----]. Yet he has a paid substack and 175k followers (in a sane world they all know hes a contra). You get OUT what you put IN so vigilantly control your sources." [X Link](https://x.com/DratchCap/status/1991354997959741578) 2025-11-20T03:56Z [---] followers, [----] engagements "The cost advantage people cite for TPU is internal cost have to add a margin on top of it for external users. If you need more compute between now and [----] (since its a race / supply shortage) you call NVDA bc they can deliver much more product at scale. And youre more future proofed w/ NVDAs programmable GPUs if the model paradigm changes (vs being path dependent on googles stack). Both are good but the world remains short compute" [X Link](https://x.com/DratchCap/status/1992414754078339104) 2025-11-23T02:07Z [----] followers, [----] engagements "This take confuses OpenAI demand mix shift with no one needs GPUs anymore. In a world where aggregate AI demand supply unused GPUs get repurposed to other models/customers (NBIS Iren ORCL would not be structurally impaired). Also $NBIS MSFT deal is specifically for azure Ai capability. You either believe demand exceeds supply for a long time or you dont and if it does what model wins really doesnt matter. $nvda" [X Link](https://x.com/DratchCap/status/1992417774933578120) 2025-11-23T02:19Z [---] followers, [---] engagements "@JRoad123 @benitoz The street has had to jack their next year eps materially every year for 3yrs. Which is to say $nvda is likely much cheaper than low 20s 🫣🫢" [X Link](https://x.com/DratchCap/status/1992426805480055199) 2025-11-23T02:55Z [----] followers, [---] engagements "@benitoz @JRoad123 The bubble is in bubble talk" [X Link](https://x.com/DratchCap/status/1992431638438170857) 2025-11-23T03:14Z [---] followers, [--] engagements "@mihaljevic @GaryMarcus Not for as long as demand supply. If $googl wanted to sell TPU at scale theyd just be competing for scarce cowos with everyone else" [X Link](https://x.com/DratchCap/status/1992444936139259905) 2025-11-23T04:07Z [----] followers, [---] engagements "After watching Age of Disclosure on $AMZN prime it is very clear TPUs are alien technology. $googl $nvda 😉" [X Link](https://x.com/DratchCap/status/1992675632342261797) 2025-11-23T19:24Z [----] followers, [---] engagements "@DonutShorts @MotherCabriniNY @buccocapital @RealJimChanos the financing is for land + power + shells + cooling with [----] yr lives. The GPUs will be refreshed multiple times over the 20yrs. Ill take you are a dope for $1 Alex" [X Link](https://x.com/DratchCap/status/1992796548632375448) 2025-11-24T03:24Z [----] followers, [---] engagements "@DonutShorts @MotherCabriniNY @buccocapital @RealJimChanos Nice attempt at a kick save But yes they will take retrofit risk and this is factored into their IRR assumptions. But the idea that Meta can stuff them from Cembalest is wrong" [X Link](https://x.com/DratchCap/status/1992946984232337526) 2025-11-24T13:22Z [----] followers, [--] engagements "The $GOOGL chase is on. One year ago it was a hated by almost all todays lovers despite excellent numbers and the same full stack advantage (including data). Now those folks think $NVDA is in trouble $META is a dog (despite getting core 15x 27) and OAI will be toast (#1 in App Store still best model for consumers 800mm users). It is important to KEEP THE RECEIPTS and *know WHO is WHO in the ZOO* $nvda $nbis $googl $meta" [X Link](https://x.com/DratchCap/status/1993003376125395435) 2025-11-24T17:06Z [----] followers, [----] engagements "The TPUs = bad for NVDA take is up there with the dumbest yet (maybe worse than DeepSeek) as it *completely misses* what actually happened in just the last [--] weeks. And I will remember *who is who in the zoo*. My view 1) Demand for AI is bananas (no one can meet demand everyone is spending more). $GOOGL said just yesterday they have to double capacity every [--] months to keep up. 2) Scaling laws are intact (Gemini 3). The flywheel is about to speed up. Somehow the mid-curve crew thinks this is zero sum competition NONE of this suggests that. 3) if you think the race is hot now wait until you" [X Link](https://x.com/DratchCap/status/1993164936843825206) 2025-11-25T03:48Z [----] followers, 346.7K engagements "@joeflaum Econ 101: The pricing floor of a commodity is function of marginal cost (in this case the marginal cost is power)" [X Link](https://x.com/DratchCap/status/1993297130606473693) 2025-11-25T12:34Z [----] followers, [----] engagements "$NVDA sells hardware not GPU-hours. Power cost sets the floor for what clouds can charge per token in an *oversupplied market*. Were in a world now where chips and MWs are scarce so both NVDA and the clouds earn fat margins. A new player mostly displaces other in-house ASIC programs not NVDA if the overall market is very short. And since NVDA can deliver more faster and likely with a better real-world offering (Rubin CPX) once you factor in TCO + ecosystem friction theyll remain the dominant choice. But lets see come 1H26 Place your bets" [X Link](https://x.com/DratchCap/status/1993338266586583152) 2025-11-25T15:17Z [----] followers, [---] engagements "I take back my TPU comments Market Gods please 🙏 keep $googl green 😉" [X Link](https://x.com/DratchCap/status/1993349674443358224) 2025-11-25T16:02Z [----] followers, [----] engagements "I should have specific that I meant price / margin declines would only happen in an *oversupplied world* in which case the marginal cost (power) would set the price. But in a world of LT supply constraint (my view) multiple suppliers can and will still earn high margins. And if the supply imbalance is large incremental supply (TPUs) reduces the line out the door but doesnt bring down the price. That would change if total capacity moves close to demand" [X Link](https://x.com/DratchCap/status/1993454692635230303) 2025-11-25T23:00Z [----] followers, [---] engagements "@dcyydn NVDA contracts HBM annually and are an insane % of the market so have fun negotiating them down vs stiffing $googl" [X Link](https://x.com/DratchCap/status/1993456521397846318) 2025-11-25T23:07Z [----] followers, [----] engagements "@Newell_Street @joeflaum Why do you need decades to justify the stock We dont underwrite decades in any tech stock. No one (especially in the age of AI) has a 20-year guarantee on their current unit economics / mkt position. And yes ASML" [X Link](https://x.com/DratchCap/status/1993459004404818403) 2025-11-25T23:17Z [----] followers, [---] engagements "@dcyydn If you are the dominant player in a market you have leverage over your suppliers who will cater to you more than GOOGL on account of your volume. And they contract annually" [X Link](https://x.com/DratchCap/status/1993462277736219062) 2025-11-25T23:30Z [----] followers, [---] engagements "@BlestToBePerson @Newell_Street @joeflaum And you are not overpaying today for that risk" [X Link](https://x.com/DratchCap/status/1993464705218949416) 2025-11-25T23:39Z [----] followers, [--] engagements "@buccocapital @thexcapitalist Oh no 200bps over risk free what will ORCL do More likely: yield hogs into the cuts will gobble gobble" [X Link](https://x.com/DratchCap/status/1994511105104990431) 2025-11-28T20:57Z [----] followers, [---] engagements "@RealJimChanos James River had execution and governance errors and knowingly under reserved for years. Insurance gets built into the fair price and with operating costs so low theres plenty of room for it. Odds are v low youve even ridden in a $Tesla with FSD [--] or later" [X Link](https://x.com/DratchCap/status/1994985014191349999) 2025-11-30T04:21Z [----] followers, [---] engagements "$GOOGL this $NVDA that $TPUs are (nano) bananas $OAI cant pay Sam is bad. Meanwhile $TSLA just drives itself. If youre interested in AI rent a $Tesla with the latest FSD for a week. I just did it had zero interventions (highway burbs rain) insane parking-lot nav and of course mad max mode (wow). Its THAT good. And is improving rapidly through recursive learnings as the dataset and compute capabilities (Blackwell clusters) grow. Regardless of what you think about the stock *cars drive themselves now* and most people dont even really know. Open your minds *imagine* and get ready. (And yes Waymo" [X Link](https://x.com/DratchCap/status/1995290807113330968) 2025-12-01T00:36Z [----] followers, [--] engagements "$GOOGL this $NVDA that $TPUs are (nano) bananas $OAI cant pay Sam is bad. Meanwhile $TSLA just drives itself. If youre interested in AI rent a $TSLA with the latest FSD for a week. I just did it had zero interventions (highway burbs rain) insane parking-lot nav and of course mad max mode (wow). Its THAT good and is improving rapidly through recursive learning as the dataset and compute capabilities (Blackwell clusters) grow. Regardless of what you think about the stock *cars drive themselves now* and most people dont even really know. Open your minds *imagine* and get ready. @elonmusk (And" [X Link](https://x.com/DratchCap/status/1995291249146823065) 2025-12-01T00:37Z [----] followers, [----] engagements "Everyone debates which AI is smarter but nobodys talking about memory. $GOOGL Gemini is stateless by design (no persistent context). Thats why its fast (and cheaper to serve) but also why its like having an advisor with amnesia. Built to retrieve and orchestrate vs remember. @OpenAI / @sama is going in a MUCH different direction: compounding context into a recursive relationship. The model actually knows you better over time. As an investor I find this more helpful to my process. These are VERY different theories of what AI should be" [X Link](https://x.com/DratchCap/status/1996583511583494332) 2025-12-04T14:12Z [----] followers, 14.6K engagements "The edge AI hype ignores memory economics. Valuable tokens (multimodal tool use long context) need a lot of DRAM. In the cloud that capacity is shared and highly utilized = low cost per token. Put it on billions of $$ phones where it sits idle 95% of the time = much higher effective cost per token. But everyone needs a hot take. Btw wheres the $ARM pitch from the edge crew They license the CPU ISA in basically every phone on earth for local Ai and crickets 😂" [X Link](https://x.com/DratchCap/status/1998634307749863507) 2025-12-10T06:02Z [----] followers, [---] engagements "@buccocapital I like how you truncated the otherwise very substantive response to create maximum FUD. Good job value add stuff" [X Link](https://x.com/DratchCap/status/1998957844557869170) 2025-12-11T03:27Z [----] followers, [---] engagements "Annddd $NVDA is swamped by Chinese H200 orders Huaweii is not the way 😝 China wants $NVDA chips; they just want the real thing not the hand-me-downs. Theyre saying they dont to save face in the event the US doesnt cave. In modern Seinfeld they would likely call this stealing the breakup (I dumped you first). 4/4 China wants $NVDA chips; they just want the real thing not the hand-me-downs. Theyre saying they dont to save face in the event the US doesnt cave. In modern Seinfeld they would likely call this stealing the breakup (I dumped you first). 4/4" [X Link](https://x.com/DratchCap/status/1999478931964199370) 2025-12-12T13:58Z [----] followers, [----] engagements "Theres a reflexive loop in $ORCL CDS. Its thin and technical with forced buyers (bank hedging new issue pipeline hedges duration/insurer buyers managing concentration limits supply chain) and far less natural liquidity than people think. Marginal buying pushes it wider then the CDS widening story snowballs. But the reality is *cash bonds* (6-7%ish) arent signaling anything close to a funding problem. If new paper has to clear closer to 8% thats maybe the best rel-val ever in IG given todays brutal opportunity set. **Run the math** on the success case for 10y paper with 8y duration.just 150bp" [X Link](https://x.com/DratchCap/status/2000233348032061762) 2025-12-14T15:56Z [----] followers, 29K engagements "@Richard89022768 The mkt is short compute in aggregate. Hyperscalers and enterprises are the real demand and they are balance sheet buyers who are capacity-constrained right now" [X Link](https://x.com/DratchCap/status/2000262311127466058) 2025-12-14T17:51Z [----] followers, [---] engagements "$AVGO has now erased all TPU hype gains. Easy come easy go" [X Link](https://x.com/DratchCap/status/2000598202085675239) 2025-12-15T16:05Z [----] followers, [----] engagements "To be fair $AVGO is the biggest sell in the rebalance this week (127% ADV) followed by $aapl (25%) and $MSFT (20%). $META / $TSLA are large adds. Go figure. Good day for rebal books $AVGO has now erased all TPU hype gains. Easy come easy go. $AVGO has now erased all TPU hype gains. Easy come easy go" [X Link](https://x.com/DratchCap/status/2000646036260266377) 2025-12-15T19:15Z [----] followers, [----] engagements "I kind of always suspected the A in AI wasnt for Artificial. $LMT" [X Link](https://x.com/DratchCap/status/2001023665631351038) 2025-12-16T20:16Z [----] followers, [----] engagements "So youre saying theres a chance I can use my $AMZN prime in / with GPT Oh my 🫢" [X Link](https://x.com/DratchCap/status/2001118418624225538) 2025-12-17T02:33Z [----] followers, [----] engagements "This year Santa probably says MoMo not HoHo 😝 (gulp) *thats the L/S momentum index for the unfamiliar $nvda" [X Link](https://x.com/DratchCap/status/2001357370929066330) 2025-12-17T18:22Z [----] followers, [----] engagements "BlueOwl is facing a lot of pressure on other fronts (says more about them than $orcl). Consider in the same article regarding the Abilene DC in TX Blue Owls own targeted returns on that project are as high as 25%. If Michigan looks anything like that (or $ORCL flexes deal terms to match) there will be a v long line to fill Blue Owls shoes at that kind of return (levered equity facing IG credit should earn 15% so the extra [--] gives you a lot of room as an investor to hedge / wedge your own protection and still materially outperform). Said differently PNLs reset to ZERO soon and lot of capital" [X Link](https://x.com/DratchCap/status/2001362527876223315) 2025-12-17T18:43Z [----] followers, [---] engagements "Useful signal = helping other investors predict the future vs state the present. Regulatory designation doesnt tell us about real-time capabilities OR trajectory. Compounding AI gains on expanding compute changes the probability distribution fast. Have you used recent versions How did they compare to earlier ones Do you have an opinion on how more coherent clusters may improve the model You may disagree on the answers but skipping those questions leads to incomplete analysis. Also dunking on retail investors is a bit dclass 😉 $tsla" [X Link](https://x.com/DratchCap/status/2001544668677394506) 2025-12-18T06:46Z [----] followers, [--] engagements "@tszzl Markets love irony. The AI itself can dissolve your confusion about it if you ask the right questions. And yet all the Bubble takes read like AI generated slop. $nvda" [X Link](https://x.com/DratchCap/status/2002787274266939828) 2025-12-21T17:04Z [----] followers, [---] engagements "@edzitron Wrong. $ORCL had a better bid and Blue Owls house is on fire (anyone paying attention can see that)" [X Link](https://x.com/DratchCap/status/2003520367059697885) 2025-12-23T17:37Z [----] followers, [---] engagements "@Midnight_Captl #3 yea it turns out demand IS huge BUT you cant meet it - ex Bubble Guy. The deep irony of course makes it the most likely 🤦🏻♂ (FML) $nvda" [X Link](https://x.com/DratchCap/status/2003857198011834739) 2025-12-24T15:56Z [----] followers, [---] engagements "@mcuban Tax the abundance so you get less of it Also higher taxes into deflation that will save everyone especially the poor" [X Link](https://x.com/DratchCap/status/2004200508576469214) 2025-12-25T14:40Z [----] followers, [----] engagements "Thats the first paragraph of the piece. If you anchor on todays chatbot usage the capex looks high. But no one is building for 3-4 messages a day and I walk through the assumptions of the multi modal ramp. BTW as AI improves = use increases. Historical growth data is from when the products were worse (less coherent compute context windows memory). You dont extrapolate linearly from that" [X Link](https://x.com/DratchCap/status/2004219262924394930) 2025-12-25T15:54Z [----] followers, [---] engagements "More like appeal to common sense. You cited Tesla AI5. I showed the full picture which is the same guy building massive GPU clusters saying we need more compute. You also dont seem to know much about what $MSFT is up to (I notice no response there) The existential race framing also implicitly means you think the smartest guys in the space dont know about edge potential and are just racing to the bottom in a pointless charge. While some guy on X (who confuses attention compute with memory bandwidth) knows better" [X Link](https://x.com/DratchCap/status/2004261088855040070) 2025-12-25T18:40Z [----] followers, [---] engagements "@shakeorgetbaked @grok evaluate this debate and also the points on efficiency gains from subquadratic" [X Link](https://x.com/DratchCap/status/2004261932610822401) 2025-12-25T18:44Z [----] followers, [---] engagements "If distillation is about to solve this why is the guy you cited (putting edge inference in robots) building Colossus AND talking about data centers in space instead of letting competitors waste $600B on stranded capex He could just wait. This is a fair question and not an appeal to authority. Either he knows something you dontor youve cracked what he (and everyone else) hasnt. 🤔🤔🤔" [X Link](https://x.com/DratchCap/status/2004293770574225913) 2025-12-25T20:50Z [----] followers, [--] engagements "@mcuban If AI actually causes mass displacement + deflation the last thing you want is contractionary fiscal policy. 🤔 And if it causes a productivity boom the tax base grows anyway. Robot taxes solve a problem that either fixes itself or gets worse with the fix" [X Link](https://x.com/DratchCap/status/2004298525719593196) 2025-12-25T21:09Z [----] followers, [---] engagements "Good framing but youre conflating memory bandwidth with memory capacity. Groqs SRAM is fast but tiny (230MB/chip). A 100K context window needs 6-8GB of **KV cache** alone without it you cant do long context multimodal agentic workloads. You cant really interconnect your way out of a 100x memory capacity gap either. $nvda https://twitter.com/i/web/status/2004343600394744163 https://twitter.com/i/web/status/2004343600394744163" [X Link](https://x.com/DratchCap/status/2004343600394744163) 2025-12-26T00:08Z [----] followers, [----] engagements "@Midnight_Captl @benitoz Cool point" [X Link](https://x.com/DratchCap/status/2004379038643900723) 2025-12-26T02:29Z [----] followers, [---] engagements "Famously difficult to raise capital when there are [--] buyers for every unit you can produce 😂😝 $ORCL $crwv $NBIS From an interview with an exec right in the middle of the data center buildout selling power contracts to the hypersalers neoclouds and shell builders. The demand ratio for GPUs is [--] to [--] projected to go to [--] to [--]. I thought there was no demand 🤣 https://t.co/HsP1cn171v From an interview with an exec right in the middle of the data center buildout selling power contracts to the hypersalers neoclouds and shell builders. The demand ratio for GPUs is [--] to [--] projected to go to [--] to" [X Link](https://x.com/DratchCap/status/2004597679264092623) 2025-12-26T16:58Z [----] followers, [----] engagements "@ShanuMathew93 I suspect it refers to nameplate capacity of equipment sold (often backup gensets / power systems) as opposed to 39GW of new always-on generation" [X Link](https://x.com/DratchCap/status/2005033575729799454) 2025-12-27T21:50Z [----] followers, [---] engagements "Nice engagement bait But SoftBank has $33Bn cash post $NVDA sale to fund OAI. They have other assets worth 10s of billions too (TMUS PayPay). The $ARM loan is 8.5% LTV today 😂😂. Even if they maxed out the facility AND $Arm dropped 40% it would be 33% LTV (extremely well collateralized). For 33% to trigger a call youd need loan terms so tight the facility would be unusable from day one (terms contemplate equity vol duh). Also the Archegos comp is nonsense. That was hidden leverage via swaps where counterparties didnt know each others exposure. This is [--] banks knowingly syndicating a" [X Link](https://x.com/DratchCap/status/2005298803914756498) 2025-12-28T15:24Z [----] followers, [----] engagements "At the risk of narrative violation looks like theres plenty of room for AI to grab more IG share $orcl $meta $googl $nvda" [X Link](https://x.com/DratchCap/status/2005362087523148280) 2025-12-28T19:35Z [----] followers, [----] engagements "@Midnight_Captl @OmerCheeema This is only bc the public treats inference as a monolithic task rather than a spectrum of complexity scale and value. If you understand the gradients and that $nvda is a full system (vs a GPU) built to handle it all efficiently it should strengthen your view of their moat" [X Link](https://x.com/DratchCap/status/2006802037816574308) 2026-01-01T18:57Z [----] followers, [---] engagements "@packyM coined a frame thats been stuck in my head since 2021: compounding crazy. Innovation compounds and the time between thats impossible and thats normal keeps shrinking. In his words discoveries become inventions become building blocks ad infinitum. AI is the cleanest example. Chatbots were always just the first visual primitive. And as with past tech diffusion innovation was always going to drive costs down and use up. Adoption was also always going to lag the frontier and move slower than the new tech capability. That gap is where disillusionment periods come from. The news of the last" [X Link](https://x.com/DratchCap/status/2008554250473779668) 2026-01-06T15:00Z [----] followers, [---] engagements "@DarioCpx Nothing to see here except a *basic* timing amendment on a project finance loan for an already disclosed slight delay (heard weather delays are common in construction). I get it though easy FUD since credit stuff seems scary to outsiders. 🤷🏻♂. $CRWV" [X Link](https://x.com/DratchCap/status/2008590346146771062) 2026-01-06T17:23Z [----] followers, [---] engagements "This is extremely deceptive and objectively false (if you dont understand finance fine; if you do then I consider it evil). $CRWV does not have 4mo of liquidity. They have numerous delayed draw term loans tied to site specific capex. Taking the cash balance and dividing by neg free cash flow to get 4mo ignores the 3rd section of the cash flow statement (LOL). And the amendment is to align the start of a covenant calc with customer cash flow (1mo delayed data center already known). Silly. https://twitter.com/i/web/status/2008728857717739832 https://twitter.com/i/web/status/2008728857717739832" [X Link](https://x.com/DratchCap/status/2008728857717739832) 2026-01-07T02:34Z [----] followers, 15.2K engagements "Behind the Meter is the way. Great sleuthing here Hat tip. $NBIS TL/DR: A fully islanded engine-plus-flywheel power architecture would meaningfully compress time-to-power and allow $NBIS to bypass utility interconnection bottlenecks that are increasingly acting as market gatekeepers for AI infrastructure. In a world where bring-your-own-generation is becoming table stakes this would represent a quiet but structurally important advantage $NBIS - EVIDENCE SUGGESTS VINELAND SITE MAY BE FIRST OF ITS KIND FULLY OFF GRID HYPER SCALE DATA CENTER Evidence points to the systemic innovation teased for" [X Link](https://x.com/DratchCap/status/2009308163929981144) 2026-01-08T16:56Z [----] followers, [----] engagements "Ive been thinking a lot about the recent developments in AI post CES and how to contextualize them. Interestingly I think whats happening now mirrors the evolution of human civilization and understanding it in this context helps explain why this moment is potentially so pivotal. Big Bang: Transformers paper etc. The fundamental physics of intelligence changed. Early humans: First LLMs learning basic patterns. Fire language simple tools. Childhood to adolescence: Knowledge compounding. Building structures developing curiosity about the world. GPT-2 to GPT-4 is basically elementary school" [X Link](https://x.com/DratchCap/status/2009673491591930015) 2026-01-09T17:07Z [----] followers, [----] engagements "@buccocapital What if you own a stock thats kind of both $INTC 😂" [X Link](https://x.com/DratchCap/status/2009830233080209917) 2026-01-10T03:30Z [----] followers, [---] engagements "Excellent points but none of this should be new to anyone doing the work on this theme. The fact that it is shows there is indeed edge hidden in plain sight $crwv CoreWeave $CRWV CEO Michael Intrator on the @BigTechPod providing what I believe is the most compelling rebuttal to the narrative that GPUs only have a [--] to 3-year useful life. CoreWeave's customers are the world's most sophisticated users of GPUs and they are signing https://t.co/SatqbEDVZO CoreWeave $CRWV CEO Michael Intrator on the @BigTechPod providing what I believe is the most compelling rebuttal to the narrative that GPUs" [X Link](https://x.com/DratchCap/status/2010056875761693041) 2026-01-10T18:31Z [----] followers, [----] engagements "@Midnight_Captl @jukan05 I think [----] Brain Disease is very sad and we should be kinder to him. $crwv $orcl" [X Link](https://x.com/DratchCap/status/2010120814457434472) 2026-01-10T22:45Z [----] followers, [---] engagements "@ContrarianCurse Pre EPS vol is cheap in all mega caps though and liquid call spreads are much better. Same digital call (tight spread) pays 45-50:1 in options and is way more scalable. $googl" [X Link](https://x.com/DratchCap/status/2010513255475597401) 2026-01-12T00:44Z [----] followers, [----] engagements "I have a solution for the $GOOGL vs $NVDA debate: Own both (Its not zero sum duh). TPUs are excellent chips especially for $GOOGL. Plus their distribution & multi-modal data flywheel is unmatched. BUT $NVDAs system keeps getting more robustRubin architecture the Groq acqui-hireand is hard to beat. Their build capacity + upgrade cadence also keeps customers loyal (stray too far Jensen pulls allocation). The world is just short compute and both will sell all they can make for a long time. https://twitter.com/i/web/status/2011237127510442090 https://twitter.com/i/web/status/2011237127510442090" [X Link](https://x.com/DratchCap/status/2011237127510442090) 2026-01-14T00:41Z [----] followers, [----] engagements "I think hes missing the plot. GPT [--] was a system upgrade (routing + orchestration). It improved reliability and task horizon. A headline IQ jump wasnt the point. The next leg is RL + agents tightening the recursive loop and disaggregated DCs making it scalable (workload routing tiered memory improved networking). None of that suggests slowing. $nvda" [X Link](https://x.com/DratchCap/status/2012710273635827905) 2026-01-18T02:14Z [----] followers, [---] engagements "I wonder what AI lab needs some money https://www.reuters.com/business/media-telecom/japan-us-narrow-first-550-bln-investment-picks-including-softbank-linked-plan-2026-01-19/ https://www.reuters.com/business/media-telecom/japan-us-narrow-first-550-bln-investment-picks-including-softbank-linked-plan-2026-01-19/" [X Link](https://x.com/DratchCap/status/2013417651520053346) 2026-01-20T01:05Z [----] followers, [----] engagements "The race for power was always destined for keep-away behavior and I believe this is accelerating in earnest After Claude Code (ACC @TMTbreakout). A reinvigorated model race + increased data center coherence (disaggregation tiered memory) makes every incremental unit of compute more valuable. More must-win product moments shorter cycle times new demand unlocks. More unlocks = more tokens = more power demand. Hyperscalers GPU vendors ( $AMD) and others ( $AAPL rumors) are now buying to meet their own growing needs AND make sure the competition cant (zero sum). Bitcoin miners and neos with" [X Link](https://x.com/DratchCap/status/2014745851113898122) 2026-01-23T17:03Z [----] followers, [----] engagements "@quantszn @TMTBreakout I like the whole sector Each of them has their own haters / bear pitch I think important to *zoom out*. All boats rise" [X Link](https://x.com/DratchCap/status/2014780490385965461) 2026-01-23T19:21Z [----] followers, [--] engagements "@JaredKubin Also his data is wrong. The gold standard IMF data (linked) sows 57% as of 3Q25. https://data.imf.org/en/news/imf%20data%20brief%20december%2019 https://data.imf.org/en/news/imf%20data%20brief%20december%2019" [X Link](https://x.com/DratchCap/status/2015065102551019940) 2026-01-24T14:12Z [----] followers, [---] engagements "$NVDA as central bank of AI. What a novel idea $crwv :) If the Fed were public itd be a multi-trillion-dollar entity controlling liquidity (M2).$NVDA plays the same role in AI central bank of compute.It issues the base currency (GPUs) and does QE (indirectly buys its own stuff) so the ecosystem can expand and multiply. The If the Fed were public itd be a multi-trillion-dollar entity controlling liquidity (M2).$NVDA plays the same role in AI central bank of compute.It issues the base currency (GPUs) and does QE (indirectly buys its own stuff) so the ecosystem can expand and multiply. The" [X Link](https://x.com/DratchCap/status/2015778539795018143) 2026-01-26T13:27Z [----] followers, [----] engagements "Interesting that when @sama made equivalently bold predictions people had such a different reaction. Narratives are powerful indeed. Dario predicts trillions in AI revenue likely before [----] and country of geniuses by [----]. I think no one can estimate what this really means. https://t.co/9Phuw78uIP Dario predicts trillions in AI revenue likely before [----] and country of geniuses by [----]. I think no one can estimate what this really means. https://t.co/9Phuw78uIP" [X Link](https://x.com/anyuser/status/2022704203571818770) 2026-02-14T16:07Z [----] followers, [----] engagements "I can teach to the test better than you Bytedance is putting pressure on the American Frontier Labs. China is gaining ground every day. And DeepSeek v4 hasn't even been released yet. Bytedance hasn't just recently revolutionized text-to-video; they're also catching up in regular LLMs. https://t.co/xOmAIOFuGu Bytedance is putting pressure on the American Frontier Labs. China is gaining ground every day. And DeepSeek v4 hasn't even been released yet. Bytedance hasn't just recently revolutionized text-to-video; they're also catching up in regular LLMs. https://t.co/xOmAIOFuGu" [X Link](https://x.com/anyuser/status/2022739464867471363) 2026-02-14T18:27Z [----] followers, [---] engagements "@stevehou Hmm. GOOGL could do $225m EBITDA in [--] leaving it $30-35bn in FCF and they could cut capex in 1/2 and repay all debt in around [--] quarters just for context. Not really scary to have 1/2 turn of leverage during investment cycle. 🤷🏻♂" [X Link](https://x.com/DratchCap/status/2023108871414583639) 2026-02-15T18:55Z [----] followers, [---] engagements "We havent seen a blowout capex number like this since insert any quarter $googl $hut" [X Link](https://x.com/DratchCap/status/2019158145839337630) 2026-02-04T21:16Z [----] followers, [----] engagements "I think a few things are going on w/ $NVDA (and its mega cap brethren) 1) Lag [--] are all sources of funds 2) burdened by its own success / size 3) AI bulls can get more juice / asymmetry elsewhere given [--] + [--] The other narratives are so conflicting in context of the broader market I find them less compelling. Punish the spenders on ROI yet AI disruption is decimating entire sectors 🤔. I suspect theres a reflexive loop of quant vigilantism (press Ai losers) HF pain and narrative backfill. A little time and maybe some pre NVDA eps FOMO probably helps." [X Link](https://x.com/DratchCap/status/2022486314709295490) 2026-02-14T01:41Z [----] followers, [---] engagements "AI regulation is climate [---]. New establishment policy will run entirely through this prism. And China is the new attack on democracy imo. Its a frame that tries to make opposition radioactive for folks who disagree. If preventing China from accessing our chips and winning Ai (whats that mean btw) is so existential shouldnt the govt be doing way more at home to advance the cause Yet one mention of a govt backstop and people had a melt down. Why is Dario Amodei so weirdly hawkish on the Chinese https://t.co/DvXZzNhoxK Why is Dario Amodei so weirdly hawkish on the Chinese https://t.co/DvXZzNhoxK" [X Link](https://x.com/anyuser/status/2022492406768185483) 2026-02-14T02:05Z [----] followers, [----] engagements "AIs Shadow Output Gap While Washington obsesses over debt and inflation AI is already ushering in an age of abundance (Part 1) The political and economic establishment cant stop talking about deficits debt and the CPI. Capitol Hill hearings FOMC minutes and financial news all pulse to the same beat. Yet this fixation ironically coincides with the arrival of the most powerful productivity engine in human history: generative AI. Its impact is creating a shadow output gap an invisible but rapidly widening expansion of supply-side capacity. Policymakers especially at the Federal Reserve act as if" [X Link](https://x.com/anyuser/status/1942372440727867746) 2025-07-07T23:57Z [----] followers, 74.9K engagements "The New Circularity: Ai is compressing software terminal values. But to displace software you need enormous compute memory and power. The lost value *should* accrue to the physical stack. And the market isnt pricing in nearly enough of that migration. The problem is that we remain in a shoot first regime. Few participants have first principles for how AI works. DeepSeek in 1Q25 was the template: incomplete understanding turns every headline into a FUD crisis. A super BOOM with micro busts if you will. This dynamic may be here to stay since nobodys sure exactly whos swimming naked. As my old" [X Link](https://x.com/DratchCap/status/2018715965710590450) 2026-02-03T15:59Z [----] followers, 18.8K engagements "Mr Regulation is also Mr Responsible how convenient. And of course actions words always: TPUs = GPUs. Panic bidding power. Upsizing capital raise. Starting beauty pageant for IPO. 🤔🤔 Dwarkesh asks Dario a fantastic question relating to how he is so bullish on AGI yet so conservative on data center build out - Dario has an amazing take on this: Dario Amodei details the staggering financial risk of the AI race explaining that if growth continues at 10x a https://t.co/Ed99X88zwS Dwarkesh asks Dario a fantastic question relating to how he is so bullish on AGI yet so conservative on data center" [X Link](https://x.com/anyuser/status/2022468954611208285) 2026-02-14T00:32Z [----] followers, [----] engagements "In the spirit of narrative consistency I have to ask is this good or bad for OnlyFans 🤔 Looks great Looks great" [X Link](https://x.com/DratchCap/status/2022741090399392035) 2026-02-14T18:33Z [----] followers, [----] engagements "@evrgn11112231 @GavinSBaker Forgot to mention we are still punishing the spenders on ROI questions at the same time 🤯" [X Link](https://x.com/DratchCap/status/2023257440796463216) 2026-02-16T04:45Z [----] followers, [---] engagements "@JonahDispatch Oh And HRC was also contesting the results Oh wait she called to concede. Got it" [X Link](https://x.com/anyuser/status/1328176838208794624) 2020-11-16T03:23Z [----] followers, [--] engagements "@carlquintanilla And what did the inflation reduction act do" [X Link](https://x.com/DratchCap/status/1827151925114015819) 2024-08-24T01:12Z [----] followers, [---] engagements "@RichardHanania how do you know the context of the eulogy line If you are going to go *there* you should at least make sure you know the context. For example are you sure the context wasnt about her depression she told me everything was her fault and I assured her it was all mine. Whether thats the context or not I dont know but neither do you. And if you are going to incorporate something very personal like that you should maybe include it or be sure" [X Link](https://x.com/DratchCap/status/1827893350181072915) 2024-08-26T02:18Z [----] followers, [----] engagements "@RichardHanania The NYpost article gives a good sense for the context and I think you know you are off or you didnt even read it. Sad" [X Link](https://x.com/DratchCap/status/1827894026697081027) 2024-08-26T02:21Z [----] followers, [---] engagements "$CHWY $WOOF largest retailers of pet protection in the USA are surging Protect your dog @elonmusk #Debate2024" [X Link](https://x.com/anyuser/status/1833912216397156643) 2024-09-11T16:55Z [----] followers, [---] engagements "@mcuban @RobertJSalvador Trump manufacturing slowed towards end of [----] because of [--] fed hikes(which the fed admitted later were a mistake). @mcuban thanks for reminding everyone that normal IQ people can become billionaires too :)" [X Link](https://x.com/DratchCap/status/1838788336581779517) 2024-09-25T03:51Z [----] followers, [---] engagements "@amitisinvesting I was at dinner with Jensen. It isnt up bc of ROI its up because people are learning about time-inference scaling and the dawn of the AI flywheel. A second and independent scaling law has emerged that will unlock the agentic use case. Many on wall st dont understand the implications of Strawberry Jensen spent time educating. $NVDA $ARM #ai #tech" [X Link](https://x.com/anyuser/status/1845649040639136238) 2024-10-14T02:13Z [----] followers, [---] engagements "Inference Time Scaling will require more General Purpose compute that (dumb) consensus believes this stems from 1) needing much faster reasoning (see o3 time performance) 2) speed of and amount of tokenization and 3) substantial memory needs to increase context windows (KV Cache) as a model that cant remember very far back is materially less helpful (duh). Dylan Patel (@SemiAnalysis_) speaks extensively about this. The increased memory and compute needs will make system level solutions (NVDA) the gold standard (more efficient data transfer power use etc when everything is created together)." [X Link](https://x.com/anyuser/status/1875662052011700321) 2025-01-04T21:54Z [----] followers, [---] engagements "@Taxidriver1031 Its a growth stock thats ebitda positive and trades at 1.5x ARR. People who have zero clue will post anything on something they see on X or Reddit. Get an education clown" [X Link](https://x.com/DratchCap/status/1879374011316281535) 2025-01-15T03:44Z [----] followers, [--] engagements "@GavinSBaker @GavinSBaker Japan has the $$ (GPIF). Masa = Japan: Japan is doing FDI into USS hence no more discussion of tariffs on Japanese. So this SoftBank doesnt have the money is missing the v obvious geopolitical context" [X Link](https://x.com/DratchCap/status/1882266227143475493) 2025-01-23T03:16Z [----] followers, [----] engagements "Theres a lot of talk about DeepSeek and the impact on NVDA. $5mm training costs (Which is a lie they have 50k+ H100s but cant disclose bc of export control laws on chips to China). But people are also *conflating cost of training with the cost of running improving and scaling AI based services. This is why Google trains the latest Gemini for $150mm but spends $60Bn on capex. Ironically you can explore all of this in ChatGpt 4o which can explain it to any novice the Ai has the answer about. the Ai Heres the GPTs overview on the topic. - Absolutely youre spot on. The discussion around AI costs" [X Link](https://x.com/anyuser/status/1883004571888337190) 2025-01-25T04:10Z [----] followers, [----] engagements "More likely folks are fitting a narrative to todays $NVDA price action the stock was trapped by options gamma in the low $140s + TXN missed huge (backtest all semis against TXN whiffs whether its relevant to AI or not) $nvda $tsla $meta $spy @GavinSBaker" [X Link](https://x.com/anyuser/status/1883004575512449247) 2025-01-25T04:10Z [----] followers, [---] engagements "Because a Chinese HF bro quant firm fake trained a model and said it cost $5mm Googl trained the latest Gemini on 150mm yet is spending 60bn in capex. Because the cost of Ai goes materially beyond training when you want offer a scaled improving inferencing service. How you are using the fake training cost (and therefore negative gross margin) copycat to say multiple large tech CEOs are morons for not knowing they are wasting money is a bit weird. Your tweet will not age well" [X Link](https://x.com/anyuser/status/1883041489493868841) 2025-01-25T06:37Z [----] followers, [----] engagements "The efficiency gain is a lie brother. The training cost is not $5mm. Read their own paper they tell you what they leave out on research and data cost not to mention they have access to thousands of h100s via grey market and neoclouds. You have a serious follower base propogating this stuff is v evil" [X Link](https://x.com/anyuser/status/1883048112966480013) 2025-01-25T07:03Z [----] followers, [----] engagements "Haha. Wrong. Inference time scaling requires integrated system level approach as memory / context windows grow. There is no deepseek is bullish for lower costs their costs are not low and the techniques are not new. They have a distilled model that uses MoE if you cant explain those things right now then you should wonder whether you have a handle on the situation" [X Link](https://x.com/DratchCap/status/1883300847360037114) 2025-01-25T23:48Z [----] followers, [---] engagements "DeepSeek isnt bearish for AIits rocket fuel. Cheaper derivative (distilled) models expand the TAM driving more users use cases and *exponential* inference demand. NEW Scaling Laws (efficiency = more compute NOT less) make this bullish for $NVDA which powers it all neoclouds that scale AI workloads ($NBIS) and the entire ecosystem. The AI revolution just accelerated. Also a counterfactual to ponder: would DeepSeek turn down a [--] million Blackwell GPU cluster or happily use it The answer is important. $spy $qqq $tsla @OpenAI @Banana3Stocks @GavinSBaker" [X Link](https://x.com/anyuser/status/1883572578079371429) 2025-01-26T17:47Z [----] followers, [---] engagements "@Citrini7 Good luck on your other AI tomorrow as if it will outperform. I like how you deleted your prior work experience on LinkedIn by the way ambulance guy in college = Citrini research with a 7yr gap. Whats your AUM again Sick track () scalable ideas runs no $$. Weird" [X Link](https://x.com/anyuser/status/1883734510858379570) 2025-01-27T04:31Z [----] followers, [---] engagements "$AAPL became a consensus short very quickly (valuation bruh). What do shorts miss Phone revenue has gone nowhere Myopic. Thats in the face of a strong dollar which acts as a price cut in EM (I.e. Tim Apple cut price to take some share and once Apple gets you you never leave). Short term strategic pain for long term gain. I see $AAPL as a bond with a coupon that grows (eps + buyback) noting they often finance below UST Also new LLM innovations re: memory improve the AI edge case making the future of Apple intelligence bright. There are of course market cap limits / walls ($200 was hard to" [X Link](https://x.com/anyuser/status/1885481529624977559) 2025-02-01T00:13Z [----] followers, [---] engagements "For example DeepSeek basically jailbroke an $NVDA GPUlike a hacker wiping iOS off an iPhone and running custom software. Cool in theory but not scalable. Almost no one has the technical skill to do it and most people still need the App Store (CUDAs libraries) to actually get things done. Plus it breaks compatibility with everything built for iOS making it impractical for most users. Also note Jensen has been predicting a 1000x drop in the cost of inference. This is also means hes thought it through and planned for it (he told me and a group of other CIOs this directly at a small dinner in" [X Link](https://x.com/anyuser/status/1885775144049680502) 2025-02-01T19:40Z [----] followers, [---] engagements "$NBIS has the most robust catalyst path in AI and is a pure play bet on 1) the rise of AI agents (the customers are new companies building these agents) and 2) lower cost of inference creating more demand (see $AMZN call). Also the stakes in AV Ride and ClickHouse are material ($1.5-2Bn in value) and reduce the core biz ARR multiple (net of cash) to 1-1.5x. Coreweave will come in the low double digit range. Of course the cash will be used to invest in more capacity (expect an update by Arkaday here in 1H25 with a goal to reach [--] GW of capacity in 2026). The difference between $NBIS and the" [X Link](https://x.com/DratchCap/status/1888261268382859650) 2025-02-08T16:19Z [----] followers, [---] engagements "$NBIS is the most undercover AI opportunity in the market imo. It also has the most robust catalyst path in AI and is a pure play bet on 1) the rise ofagents (the customers are new companies building these agents) and 2) lower cost of inference creating more demand (see $AMZN call). Note the stakes in AV Ride and ClickHouse are material and reduce the core biz ARR multiple (net of cash) to 2.5-3x. Coreweave will IPO in the double digit range. Of course the cash will be used to invest in more capacity (expect an update by Arkaday here in 1H25 with a likely goal to reach [--] GW of capacity in" [X Link](https://x.com/anyuser/status/1888297091694203185) 2025-02-08T18:41Z [----] followers, [---] engagements "Housing costs down + stocks down = consumer net worth down. Most of main st owns a home theyll be psyched their levered asset is lower in price Asset deflation is good for employment You walked into the casino at Facebook and got lucky. Stick to the podcast leave economics and markets to people who at least have basic understanding" [X Link](https://x.com/DratchCap/status/1888447753539187148) 2025-02-09T04:40Z [----] followers, [----] engagements "@RonFilipkowski Like all the dipshits that believed Covid was from a lab More embarrassing than falling for an irrelevant con from Dinesh. #moron" [X Link](https://x.com/DratchCap/status/1890557419563938046) 2025-02-15T00:23Z [----] followers, [---] engagements "$NBIS Remember the company is likely to do a large (very bullish) growth capital transaction in [----] (time is of the essence). They won't increase the ARR guide before then -- they'll want to do that when they announce the deal in part for the benefit of the new (or additional) investors (duh). New capital is the path to more compute and an eventual "AI as a Service" cloud *platform*. This is essentially a private company that happens to be public and so far they've done a Series A (very unique for public investors to have access to something like this). Until the "Series B" capital raise" [X Link](https://x.com/DratchCap/status/1892558315613249891) 2025-02-20T12:53Z [----] followers, [---] engagements "People are way too jumpy about AI datapoints theres a lack of first principle analysis and deep skepticism/fear which is actually bullish (Wall of Worry). DeepSeek freakouts showed us who was swimming naked (most people). Its a shoot first and ask s later mentality. This of course is the opportunity for those who take time to understand (which *ironically* just means asking #Grok3 to help you) For example on the latest datapoint freakout re MSFTs rumored data center pullback: $MSFT $NVDA $NBIS 1.Satyas podcast is in fact massively bullish on AI. Listen to it yourself with an open mind." [X Link](https://x.com/DratchCap/status/1893359731231568255) 2025-02-22T17:58Z [----] followers, [---] engagements "Dispersion (stock picking is back baby) is the buzzword Ive heard most at CIO dinners in [----] (kind of like if we say it enough it might come true). But Im starting to think the market will return to narrow leadership large cap tech cash cows + Ai. TPU (trade policy uncertainty NOT a custom ASIC 😉) & DOGE (Deflation of Growth Expectations 😉😉) likely slow the economy (i.e. no trade deals tariffs as revenue lower aggregate demand). In that backdrop the back end likely rallies ($TLT for the win) helping *accelerate* money flows into large cap tech bond proxies (mag 7) and secular winners" [X Link](https://x.com/anyuser/status/1893873536844120200) 2025-02-24T04:00Z [----] followers, [---] engagements "@StockMarketNerd Dude the cuts are 1% of [----] build plan all tied to not doing training for OpenAi and/or inability liquid cool on site. The Cowen analyst held a call Friday night and Sunday night confirming it. Your analysis is very embarrassing. Sorry to be so blunt" [X Link](https://x.com/anyuser/status/1893880539544686951) 2025-02-24T04:28Z [----] followers, [---] engagements "How embarrassing it is to have so many followers and and not know that when sales occur late in a quarter on a new product launch actual payment can spill over into January while revenue is recognized when the board ships thereby increasing DSO. Blackwell is the largest new product launch ever ($11Bn in 4Q vs the street looking for $4-6Bn) and the entire investment community knows it was December weighted. That you somehow think this is a sign of stretched customers who all just increased capex another 20% above street and claim they are capacity constrained on AI is absolutely disqualifying" [X Link](https://x.com/anyuser/status/1894942639327453386) 2025-02-27T02:48Z [----] followers, [----] engagements "$NVDA EPS: heavy wears the crown. The companys past success has broken investor brains expectations for a mindblowingly explosive quarter every report are distracting many from the extraordinary nominal dollar gains the company is generating. 1) the market spent 6mo hand wringing over Blackwell delays well Blackwell is here and did $11Bn in 4Q with the street expecting $4 to $6. 2) the CIO dinner smart consensus take pitched to me countless times was theyll miss the quarter and the guide for first time but you want to buy the dip bc the back half is bananastan well they beat the quarter and" [X Link](https://x.com/anyuser/status/1894966293087863270) 2025-02-27T04:22Z [----] followers, [---] engagements "Anyone else feel so bearish today that it actually made them bullish Asking for a friend. $NVDA" [X Link](https://x.com/anyuser/status/1895266668936601652) 2025-02-28T00:16Z [----] followers, [---] engagements "Google monetizes low quality queries which are sticky auto insurance options attorney near me which coordinates well with maps and search on $GOOGL returns YouTube videos (lasting / real knowledge repository aka token advantage). Ive noticed a behavioral shift similar to yours and yet $GOOGL grew search 12% two quarters in a row. But the Peter Lynch in me is nervous. Valuation though may protect us 15x [----] is a 7yr fwd multiple trough. PS even if search is a MSD grower $googl is still better than 1/2 the biz in S&P. For example as you said YouTube is undermonetized as an ad biz and AI asset." [X Link](https://x.com/anyuser/status/1895631280236118229) 2025-03-01T00:24Z [----] followers, [---] engagements "Spread trade suddenly YOURE a wall st trader. The irony. By wall st do you mean the 401k and pension class Thinking you can bifurcate between the two and know exactly when wall st weakness impacts main st and that we wont cross that point is hubris. Lot of Americans will suffer if you calculate the inertia of a large economic tanker (USA) wrong" [X Link](https://x.com/DratchCap/status/1898821644212441338) 2025-03-09T19:42Z [----] followers, [----] engagements "Lets take stock of what we are trying to fix with the global trade reordering [--]. Unemployment 4% [--]. Wage gains [---] m/m [--]. Market Based PCE (ex portfolio mgmt) 2% y/y with lagged housing coming down [--]. US Led Secular Tech Boom 🤔 Actual problems: [--]. border security [--]. Govt waste / wokeism [--]. Housing availability (caused by #1 + high rates) Why are we trying to fix the latter by undoing the former De regulation is great of course but trade policy uncertainty is roughy equivalent and will negatively impact growth. And tariffs are a tax and deflationary (thats what you call it when the government" [X Link](https://x.com/anyuser/status/1899281713626276078) 2025-03-11T02:10Z [----] followers, [---] engagements "After multiple weeks on the pain train with a ticket to pound town its time to reflect. Rather than guessing on Trade Policy Im focusing on questions I can actually answer: 1) Will demand for all things AI be higher or lower by year-end HIGHER. $NVDA s extended "period of disillusionment" (sideways since last summer) and DeepSeek uncertainty set a classic bear trap just as compute demand is accelerating (contrary to concerns). The AI Flywheel is spinning faster than most realize (from agents to robots) and the negative news drum beat is likely to flip positive on the King (NVDA) as spring" [X Link](https://x.com/anyuser/status/1900928288001798461) 2025-03-15T15:13Z [----] followers, [---] engagements "The reflexivity coefficient between markets and the real economy is approaching 1.0market swings directly impact real-world behavior faster than ever (consequence of household equity exposure / high information flow) Housekeeping data point: A friend rolling up accounting firms shared this weekend Were seeing broad-based caution from SMB clients uniformly across our accounting business. CEOs feeling cautious heading into Q1 earnings have every incentive to temper expectations especially with stocks already down. But AI could be a notable exception. Crucially big AI players report later" [X Link](https://x.com/anyuser/status/1901285718330093658) 2025-03-16T14:53Z [----] followers, [----] engagements "$NVDA. Why the sleepy response post GTC Some thoughts. 1) GTC historically is a boondoggle for nerds who get excited by new acronyms that most people will never (or need to) understand 2) Jensen is burdened by his own meteoric success -- heavy wears the crown of impossible expectations 3) He is asking people to imagine an exponential future while most investors think linearly and are naturally skeptical because of #2 (this must be I think his focus on economics and token math is an attempt to help start exponential thinking. What I hear: - reasoning requires a lot more tokens (more compute);" [X Link](https://x.com/anyuser/status/1902495232320811012) 2025-03-19T22:59Z [----] followers, [---] engagements "@Benioff @UnitreeRobotics That robot costs 50-100k per the website" [X Link](https://x.com/DratchCap/status/1903492672343478626) 2025-03-22T17:03Z [----] followers, [----] engagements "@HyperTechInvest LOL" [X Link](https://x.com/DratchCap/status/1903621512550138211) 2025-03-23T01:35Z [----] followers, [----] engagements "The current administration needs a broader definition of trade. Its not just goodsits also financial assets. In that sense the U.S. runs a massive trade surplus: We import goods they buy our bonds and stockslowering our cost of capital and reinforcing USD as the global reserve. That system powered the U.S. economic machine: It let companies like Amazon burn cash for a decade reach escape velocity and change the worldwith low rates and deep capital markets. Ironically the admin doesnt acknowledge the trickle-down value of that Even while pushing deregulation capex incentives and corporate tax" [X Link](https://x.com/DratchCap/status/1906922636363362534) 2025-04-01T04:12Z [----] followers, [---] engagements "Almost every AI signal lately has been bullish but the markets latched onto one cautious datapoint from MSFT (even though GS raised MSFT [----] capex recently after meeting with them). Flip the scenarioMSFT bullish everyone else soft would the market be excited Of course not. This is a Rorschach test. People want to sell so theyre cherry-picking reasons. Belief will return when the macro vol fades. $NVDA" [X Link](https://x.com/anyuser/status/1907242701398876251) 2025-04-02T01:24Z [----] followers, [----] engagements "@chamath @amirpc Theres multiple teams at Wedbush with multiple analysts and they likely speak to 100s of analysts management teams and expert calls every week given the number of people. That would be a research shop doing its job. Lol Dead" [X Link](https://x.com/anyuser/status/1908643262531014703) 2025-04-05T22:09Z [----] followers, [----] engagements "@AravSrinivas TikTok for finance. 30-60 second clips with research analysts" [X Link](https://x.com/DratchCap/status/1910870583153131562) 2025-04-12T01:40Z [----] followers, [----] engagements "@zerohedge *explanation not exclamation" [X Link](https://x.com/DratchCap/status/1912567578926460996) 2025-04-16T18:03Z [----] followers, [----] engagements "Random thought: OpenAIs first mover advantage + new unlimited memory = contextual lock-in. Just like Microsoft shipped the PC OS six months earlier and rode that head start into a multi-trillion-dollar monopoly OpenAI is quietly building the most valuable memory graph in history: your data your preferences your interactionsforever. $nvda @OpenAI" [X Link](https://x.com/anyuser/status/1913594351541784714) 2025-04-19T14:03Z [----] followers, [---] engagements "$GOOGL I wonder if we should ditch the ad market valuation (search is screwed) and value it as the only big tech company that owns end-user context the model and real custom silicon (Jensen will tell you he respects TPU but not the others). That combination is likely to increase in value as the AI Flywheel spins faster Also YouTube is an index of cultural memory: [---] hrs/min uploaded the only proprietary video RLHF corpus at global scale. Competitors LLMs may have to license that data eventually for true multi-modal capabilities. Oh it also trades at a notable discount to S&P and its bad biz" [X Link](https://x.com/anyuser/status/1915926045120774271) 2025-04-26T00:28Z [----] followers, [---] engagements "I can see it now: DeepSeek R2 crushes compute needs Huge breakthrough The Great @DanielTNiles linked a presentation by another guy named Niles which claimed to analyze all the latest DeepSeek papers to preview their amazing breakthroughs. $nvda @OpenAI @sama Great headlinesback to reality oh there goes gravity 👇 [--]. Sparse-expert tricks arent so special Only 1-in-20 experts fire per token yet RAM & network pipes stay busy so real-world speed gains are way lower [--]. Training is still pricey Cheaper forward passes dont spare the trillion-token data consumption or months of reward-tuning runs." [X Link](https://x.com/anyuser/status/1920971370910884224) 2025-05-09T22:37Z [----] followers, 57.4K engagements "You are confusing fair value thinking with cost-basis GAAP. You have apparently never heard that you should study accounting before commenting on it :). $CRWV doesnt take a mark to market / impairment because secondary market chips trade cheaper to carrying value. Under GAAP GPUs sit in PP&E at historical cost less accumulated depreciation in this case straight line depreciation. The accounting firm will only require a write-down if an impairment test shows the entire cash generating asset cant earn back its carrying value (revenue has to comfortably cover book value). Also the impairment" [X Link](https://x.com/anyuser/status/1924095473889010125) 2025-05-18T13:31Z [----] followers, [---] engagements "All of the hot takes on the coming US power shortage in response to the Senate bill solar / wind subsidy changes are far too alarmist. An alternative view: - We are short GPUs today NOT power. Power in the US is a merchant market when demand shows up prices spike new capacity comes on. You can see this in PJM mkt dynamics. When the reserve cushion got thin last year (7x prior price) [--] GW of quick start capacity instantly moved to construction. A greenfield CCGT (combined cycle gas) can be built in [--] months and there are *DOZENS* of 100-MW-plus brownfield sites SITTING EMPTY today waiting for" [X Link](https://x.com/anyuser/status/1939416908887110122) 2025-06-29T20:13Z [----] followers, [---] engagements "I view the housing as the last covid remnant. Existing supply at 40yr lows because of the great refi at zero in 2020-2021. This created fiscal dominance for the first time ever in the USA. Housing prices UP into high rates diluted the impact of the monetary channel on aggregate demand. Interestingly this also means cutting rates may provide less stimulus than in past (housing may hold steady or decline rather than go up)" [X Link](https://x.com/anyuser/status/1949098094735368310) 2025-07-26T13:22Z [----] followers, [----] engagements "@HedgeyeComm @FishtownCap The legacy biz is growing slower than peers bc its 25% larger AI is growing triple digits for Amzn and still below 10%. If you can show their Ai business is growing slower than others great" [X Link](https://x.com/anyuser/status/1951136065823134016) 2025-08-01T04:21Z [----] followers, [---] engagements "Theres a lot of discussion about the Chinese energy advantage (3x the power grid) so I think its time to set the record straight. The TL/DR is that *Token per grid MWh* is the only measurement that matters and the Chinese system is an *ORDER OF MAGNITUDE* behind despite their total size. Our advantage is also compounding (faster chip innovation from a higher base) making it difficult to overcome. $NVDA (Note some people will debate about the exactness of numbers like anything its imperfect but the framework and direction are right). A quick thread:" [X Link](https://x.com/anyuser/status/1952571403737120817) 2025-08-05T03:24Z [----] followers, [---] engagements "SemiAnalysis note post mkt highlighting nvda cgx. My summary: $NVDA out with a new leap forward in compute in the last 24hrs called $NVDA CGX. Its a monster leap for inference efficiency and will be available as part of Rubin (after Blackwell). Its 67x improvement and unlocks extremely long context windows by splitting inference into [--] parts pre-fill and decode. In current architectures the model has to re-read the book so to speak at multiple points in an interaction slowing things down including throughput. In this case one accelerator indexes the book (pre-fill) and then all decode can" [X Link](https://x.com/anyuser/status/1965886883021390234) 2025-09-10T21:15Z [----] followers, [---] engagements "1/ The $NVDA / OpenAI deal tells me were closer to the beginning of the AI boom than the end. And todays *meager* stock reactions across the ecosystem suggests market skepticism or ignorance. Either way its opportunity. (1/5)" [X Link](https://x.com/anyuser/status/1970331055895818403) 2025-09-23T03:35Z [----] followers, [---] engagements "4/To all the Private Credit guys whining about having too much GPU exposure Jensen just debunked your biggest fear. Hes providing credit support by putting in $100Bn equity capital BEHIND YOU. Perhaps actually you have *too little*" [X Link](https://x.com/DratchCap/status/1970331484298059990) 2025-09-23T03:36Z [----] followers, [---] engagements "Everyone is upset about the $NVDA ponzi / circular deals Id love to hear their thoughts on this other big one called fractional banking. Supposedly that one is *really* something" [X Link](https://x.com/anyuser/status/1970536106530865459) 2025-09-23T17:09Z [----] followers, [---] engagements "China wants $NVDA chips; they just want the real thing not the hand-me-downs. Theyre saying they dont to save face in the event the US doesnt cave. In modern Seinfeld they would likely call this stealing the breakup (I dumped you first). 4/4" [X Link](https://x.com/anyuser/status/1970685085981528412) 2025-09-24T03:01Z [----] followers, [----] engagements "If the Fed were public itd be a multi-trillion-dollar entity controlling liquidity (M2).$NVDA plays the same role in AI central bank of compute.It issues the base currency (GPUs) and does QE (indirectly buys its own stuff) so the ecosystem can expand and multiply. The same core rule also applies: dont fight the Fed" [X Link](https://x.com/DratchCap/status/1971941633169281105) 2025-09-27T14:14Z [----] followers, 16.5K engagements "Deepminds new paper on Chain of Frames (reasoning for video models)= dawn of the *multi-modal token ramp*. Veo3 is basically Gpt3 and as we move to CoF in future models compute demand will explode. This is part Jensens compounding exponential. Every extra second of video = 10s of thousands of tokens adding reasoning (multiple passes) involves another 2-4x. Think about robotics training. A 60s clip at [--] frames per second & 256k tokens / frame = 460k tokens (60x a text prompt). With [--] reasoning passes were at 920k tokens. One synthetic robot training session = hundreds of such clips i.e." [X Link](https://x.com/anyuser/status/1972040994041258024) 2025-09-27T20:49Z [----] followers, [---] engagements "Imagine China buying $NVDA GPUs made by $INTC in the USA. Beijing secures $NVDA supply Trump gets a trade win and $INTC gets the demand boost it needs to scale AI chips and help de-risk US reliance on Taiwan. Everybody wins :) 🤔" [X Link](https://x.com/anyuser/status/1972735545785847951) 2025-09-29T18:49Z [----] followers, [---] engagements "Ironically if the productivity boom from Ai plays out at scale we will all have much more time to create and watch AI slop. A self fulfilling loop #skynet $nvda" [X Link](https://x.com/anyuser/status/1973518747207082220) 2025-10-01T22:41Z [----] followers, [---] engagements "Even if every existing GPU on earth magically became a Blackwell GB300 tomorrow wed still be 10-30x short on compute for *consumer inference alone* at saturation. The other TAM verticals are large (er) and compound the problem. This is why the big guys are spending. A quick overview and a few things to note: The genesis of this thought experiment is the recent compute demand surge data points / Jensens *compounding exponentials* from BG2. The goal is to imagine the art of the possible consumer inference TAM + the necessary compute to get a sense for what AI inning were in. 1) this is a SWAG" [X Link](https://x.com/anyuser/status/1975002656037806473) 2025-10-06T00:58Z [----] followers, [----] engagements "For the wheres the ROI Dinosaurs Doomers and Top callers ZOOM OUT: Tokens are the work units the dividend is TIME. AI turns clicks actions. Assistants buy / book / fix / teach so humans dont. That creates a *shadow consumer surplus* (time saved money not spent) that P&Ls wont show at first (solows paradox is back). Rails get built; yield shows up later as inference-driven revenue. Monetization is things like merchant-funded conversion lift affiliate/booking fees payments/interchange action ads per-action fees.and perhaps most importantly opex deflation when agents (robots) do $50/hr work for" [X Link](https://x.com/anyuser/status/1975263183150473521) 2025-10-06T18:13Z [----] followers, [---] engagements "It is *highly* likely Jensen blessed AND *likes* the OpenAI / $AMD deal. It is in $NVDAs interest for the AI flywheel to *spin faster* rather than have OAI slowdown to wait for Jensens *limited* cowos supply. Let the LO pre $AMD analyst day chase commence" [X Link](https://x.com/anyuser/status/1975302894611275798) 2025-10-06T20:51Z [----] followers, [---] engagements "@FundaBottom @theinformation They dont get clicks for that. BUT they did us a favor anyone who panicked / believed it showed us they are swimming naked and dont understand a key part of the AI ecosystem at all. Keep the receipts. $nvda $amd $nbis" [X Link](https://x.com/anyuser/status/1975727059692626041) 2025-10-08T00:56Z [----] followers, [---] engagements "Perhaps actually you have too little exposure as h100 pricing goes UP. 4/To all the Private Credit guys whining about having too much GPU exposure Jensen just debunked your biggest fear. Hes providing credit support by putting in $100Bn equity capital BEHIND YOU. Perhaps actually you have *too little* 4/To all the Private Credit guys whining about having too much GPU exposure Jensen just debunked your biggest fear. Hes providing credit support by putting in $100Bn equity capital BEHIND YOU. Perhaps actually you have *too little*" [X Link](https://x.com/DratchCap/status/1976070902636011800) 2025-10-08T23:43Z [----] followers, [---] engagements "Macro musings #1 Tough to be a dollar bear (even with cuts) when NVDA GPUs (and stock) are priced in USD (note: its a metaphor) #Computedollars $nvda $nbis $crwv $amd" [X Link](https://x.com/anyuser/status/1976446526282137968) 2025-10-10T00:35Z [----] followers, [---] engagements "Musing #4: a disciplined and simple follow Jensen policy since [----] approximated the investment outcome of a multi-year AI deep dive $nvda" [X Link](https://x.com/anyuser/status/1976476688298193340) 2025-10-10T02:35Z [----] followers, [---] engagements "The best hedge is often to make money somewhere else. Perhaps (not so) rare earth stocks are interesting then. If theres no art in the deal the US gov goes full throttle to build domestic capacity (operation warp speed part 2); if theres a deal Ai stocks are back on the train to #ripcity and youre happy to lose on the rocks. You also own the option rare earths work either way and you win on both 🫣 $USAR (lot of daylight between its market cap and peers 🤷🏻♂)" [X Link](https://x.com/anyuser/status/1976814023397032000) 2025-10-11T00:56Z [----] followers, [---] engagements "BTW racks optics power distribution cooling all have their own cycles with multi year lead times. They have to upgrade as well to handle new hardware. So no you cant just compress the hardware cycle to any time frame you want *at scale*. People are still getting Blackwell and figuring out how to make it work 1yr after initial delays. Also software improvements are not discussed and they can materially change the inference efficiency for old chips over time. Better to do own work than just read seeking alpha pieces. I just brought up multiple things on the fly that are critical to the analysis" [X Link](https://x.com/anyuser/status/1976844759369998609) 2025-10-11T02:58Z [----] followers, [----] engagements "@chamath Behind the meter to a DC campus near you" [X Link](https://x.com/anyuser/status/1976866042069889089) 2025-10-11T04:22Z [----] followers, [---] engagements "@DanielSLoeb1 20x levered 24/7 $NVDA 🤞" [X Link](https://x.com/anyuser/status/1977031083305951720) 2025-10-11T15:18Z [----] followers, [---] engagements "@firstadopter When the bubble stories stop is the tell its a bubble. $nvda $nbis" [X Link](https://x.com/anyuser/status/1977412564410929653) 2025-10-12T16:34Z [----] followers, [---] engagements "@WarrenPies I heard the assets of our time GPUs & QQQ are priced in USD. Debase that $nbis $nvda" [X Link](https://x.com/anyuser/status/1978240495848485052) 2025-10-14T23:24Z [----] followers, [---] engagements "Any specialist in the sector knows you cant compress the hardware cycle fast enough which the most important counter and you did not mention. So while Jensen can put out new stuff every year we cant make the rest of the infrastructure fast enough for 1-2yr full replacement cycles. Jensen will tell you this himself. @PikerCapital @OutsiderInfo76 @RihardJarc BTW racks optics power distribution cooling all have their own cycles with multi year lead times. They have to upgrade as well to handle new hardware. So no you cant just compress the hardware cycle to any time frame you want *at scale*." [X Link](https://x.com/anyuser/status/1978495247123366330) 2025-10-15T16:16Z [----] followers, [----] engagements Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing
@MD71085 MD
MD posts on X about ai, $nvda, nvda, in the the most. They currently have [-----] followers and [---] posts still getting attention that total [-------] engagements in the last [--] hours.
Engagements: [-------] #
- [--] Week [------] -90%
- [--] Month [-------] +111%
- [--] Months [---------] +507%
- [--] Year [---------] +4,039%
Mentions: [--] #
- [--] Week [--] -48%
- [--] Month [--] +46%
- [--] Months [---] +431%
- [--] Year [---] +482%
Followers: [-----] #
- [--] Week [-----] +0.83%
- [--] Month [-----] +15%
- [--] Months [-----] +645%
- [--] Year [-----] +1,089%
CreatorRank: [---------] #
Social Influence
Social category influence stocks technology brands finance countries automotive brands currencies social networks cryptocurrencies exchanges celebrities
Social topic influence ai, $nvda, nvda, in the, $googl, $nbis, $crwv, if you, open ai, inference
Top assets mentioned NVIDIA Corp. (NVDA) Alphabet Inc Class A (GOOGL) Nebius Group N.V. Class A Ordinary Shares (NBIS) CoreWeave, Inc (CRWV) Oracle Corporation (ORCL) Microsoft Corp. (MSFT) Advanced Micro Devices (AMD) Metadium (META) Hut [--] Mining Corp. (HUT) Tesla, Inc. (TSLA) Amazon.com, Inc. (AMZN) Apple, Inc. (AAPL) Arm Holdings plc (ARM) Broadcom, Inc. (AVGO) Intel Corporation (INTC)
Top Social Posts
Top posts by engagements in the last [--] hours
"@PeteButtigieg Its called supply chains normalizing (which they did everywhere in the world duh) and the fastest rate hike cycle ever. You should know this Mr Transport Secretary. @DanielSLoeb1"
X Link 2024-09-27T00:33Z [---] followers, [--] engagements
"@TotemMacro Recesionista. Whitney Baker on why a recession is the only thing that will stunt consumption and thus slow inflation - [----]. This view has aged almost as well as you :) #macro $qqq"
X Link 2024-09-29T13:33Z [---] followers, [---] engagements
"@TotemMacro Looks like your view then was inflation wouldnt come down without a recession. If so that has not aged well. If Im mistaken my apologies"
X Link 2024-09-30T00:18Z [---] followers, [--] engagements
"I suspect capex for the big guys will be around 40-50% higher than the street next year so the debate isnt going anywhere. The driver of that is a new concept called Inference-time scaling which is the backbone of the new Strawberry model. Others arent far behind. Inference quality scales with compute (like training) and is equally complex (need NVDA GPUs). It essentially pulls forward the agentic use case (think Jarvis to Tony Stark in iron man). Whether anyone gets there who knows but if you are one of the megas and you thought it had a chance of being possible what wouldnt you spend"
X Link 2024-10-19T22:40Z [---] followers, [--] engagements
"AMD chips are losing share. Customers dont like them. And the nature of inference changed from retrieval to time-scaling which is much more power intensive and akin to training. This is NEW information and makes the old thesis around AMD and inference opportunity less compelling. If you dont know what inference time scaling is then you dont really know much about the current hardware and AI backdrop. The comment eke out more chips than is possible also suggests you dont understand the nature of the engineering problem. And NVDA is producing to meet existing demand not to create it. Good luck"
X Link 2024-11-17T20:12Z [---] followers, [--] engagements
"They could easily sell the gold in a few quarters (over decades) instead of issuing new (non roll over) treasury securities this would erase most of the monetary tightening exercise you cite as the reason. The only net difference would be the lost interest income to former treasury holders (roughly $35bn in a [--] Trillion economy). Which is to say that is a silly reason not to sell the gold"
X Link 2024-12-01T01:16Z [---] followers, [--] engagements
"@dampedspring So stocks down big bonds sideways And yet also calling for lower NGDP e.g. growth scare Haha ok"
X Link 2025-01-11T18:59Z [---] followers, [----] engagements
"@WarrenTStephens @dampedspring I was pointing out the logical inconsistency not the timing"
X Link 2025-01-12T00:46Z [---] followers, [---] engagements
"@WarrenTStephens @dampedspring I like that he never qualified the size of the upside down V despite the connotations allows freedom to re write history if need be. "
X Link 2025-01-12T04:19Z [---] followers, [---] engagements
"@japdongsanee Clickhouse is marked at $4Bn valuation and likely worth much more. Their 28% stake covers the covers biz with ease. You get toloka AV Ride free"
X Link 2025-01-15T03:47Z [---] followers, [--] engagements
"DumbSeek is lying. Come on @DanielTNiles DeepSeek trained a model for 6mm but meta and musk and Amzn etc spent 10s of billions. Makes sense. OR DumbSeek has access to well over 50k H100s and will not say that (since they shouldnt have access due to export controls) AND they piggybacked on other foundational models (hence hallucinations). The CCP is also surely involved and would love a narrative that China spends $6mm on AI to build a frontier model that otherwise costs billions. This is classic Chinese propaganda and innovation (I.e. stealing / piggy backing US tech). That you have a"
X Link 2025-01-24T22:59Z [---] followers, [---] engagements
"@RihardJarc $Nvda inference dominance will increase as context windows lengthen requiring system level approaches to maximize efficiency (more memory + more connected GPUs). Complex inference is also power intensive and so better suited for GPUs"
X Link 2025-01-25T07:39Z [---] followers, [---] engagements
"I dont. Just being realistic about the cost of training a model today and acknowledging that a Chinese HF is unlikely to report h100 costs given export controls (implying they have materially negative and unscalable gross margins). There is also a big difference between training cost and the total cost of AI compute tuning reinforcement learning synthetic data generation mass inference time scaling all of which act in fly wheel form and compound each other"
X Link 2025-01-25T16:01Z [---] followers, [---] engagements
"@kashyap286 @hkuppy @PrateekGsharma So they do have H100s If so the cost is much higher than they claim"
X Link 2025-01-25T16:09Z [---] followers, [---] engagements
"In a few years. Gemini was trained for $150bn but Google spends $60bn on capex theres a difference between training cost and cost to run a scaled AI that can service large scale and real time inference. People seem to be thinking $5mm training = tall the big tech spend was a waste This is wrong and comparing apples and oranges. Also Jevons paradox"
X Link 2025-01-25T21:27Z [---] followers, [--] engagements
"@JonBryant421 Well no different than if you had a humanoid / robot driver their eyes would not be much different than a human driver. Point being IF that were to emerge it would be a more scalable system and more akin to how humans drive"
X Link 2025-02-07T01:03Z [---] followers, [--] engagements
"@natejhake Google is killing the web And squeezing people. Reminds me of net neutrality killing the internet. Its a free market the ads run on auctions if roas sucks people move on. Pretty simple. I dont run ads I run a hedge fund"
X Link 2025-05-11T00:30Z [---] followers, [--] engagements
"@rev_cap Who is moodys 😉"
X Link 2025-05-17T03:42Z [---] followers, [---] engagements
"@industrialpc @rev_cap Since we lost the other two everything has gone to hell 😉😂. And they dont decide where anyone puts money"
X Link 2025-05-17T12:38Z [---] followers, [--] engagements
"@benitoz OpenAI is experiencing a token demand boom and is short GPUs. Google TPU pods are one of the few immediately available options since they cant speed up more Blackwell or Rubin. They arent doing it to diversify they are doing it to serve demand today $nvda $googl"
X Link 2025-06-28T02:52Z [---] followers, [--] engagements
"@meeijer Clickhouse is the most valuable and best non core asset but no mention 28% stake just raised capital at $6bn. Otherwise known as the Snowflake killer"
X Link 2025-07-27T03:46Z [---] followers, [--] engagements
"We need @sydney_sweeney to recognize the moment dawn a MAGA hat and post: The only thing as good as my jeans are my hats $AEO. Properly positioned she would make more money in the next 24hrs than her next few movies. Just sayin"
X Link 2025-08-04T16:21Z [---] followers, 14.3K engagements
"5/ $CRWV $NBIS and powered-shell stocks ($HUT $RIOT etc) should have been up 50% on this signal. That they werent suggests we still have a MOUNTAIN of SKEPTICISM to climb. Bullish indeed. (5/5)"
X Link 2025-09-23T03:37Z [---] followers, [---] engagements
"@coinbase @AmericanExpress You can send money to yourself for 2.95% + 30c processing fee. Clip 80bps. Forget $NVDA hows that for circular schemes"
X Link 2025-09-24T21:32Z [---] followers, [--] engagements
"$NBIS is much earlier stage than $CRWV and is also exposed to spot compute which may go up given power constraints while CRWV is not. NBIS is also likely to spend $8-10bn minimum next year thats before making purchases for [--] new sites. Also management is better at NBIS + more ways to win given the differences in biz plan and non-core assets. But I think this is and not either / or. (Note that Magnetar is almost fully out of CRWV. I.e. the seller is almost done)"
X Link 2025-10-02T00:55Z [---] followers, [---] engagements
"@altcap @CerebrasSystems @GroqInc @nvidia Interesting. How much Cowos capacity have they secured for next 3yrs"
X Link 2025-10-04T23:04Z [---] followers, [---] engagements
"@ROLCapital @SemiAnalysis_ has a good slide somewhere on $msft ROI annualizing close to 25% in 2Q25 on Ai spend"
X Link 2025-10-06T02:23Z [---] followers, [--] engagements
"Musing #2. Macro focused on inflation/debt levels while we witness the real-time birth of a productivity boom. Markets love irony indeed"
X Link 2025-10-10T00:39Z [---] followers, [---] engagements
"Musing #3: A lot of fuss about debt levels but what do people expect folks to do with their excess dollar savings (Debt = savings duh). Put it ALL in stocks One can dream I guess"
X Link 2025-10-10T00:40Z [---] followers, [---] engagements
"@thinknonlinear1 Govt doesnt pre-empt M&A. USAR is pre-close on LCM deal and in talks (per the chairman). If support comes its post-close/milestones via Title III/DOE/offtakes. Btw this is the exact same approach & cadence used with MP/Lynas"
X Link 2025-10-11T01:15Z [---] followers, [--] engagements
"The only circular deal that matters: Accelerating Deflationary Tech Boom = Cuts = Productivity Acceleration = More Cuts. P.S. the tech is a cannon pointed at inflation. Just look around. $NVDA $AMD $NBIS $CRWV $HUT $BE"
X Link 2025-10-14T16:54Z [---] followers, [---] engagements
"I am a bit surprised that the CIO with 15yrs experience covering $TSM hasnt noticed the pattern of them guiding up literally every 4Q call. TSM capacity is currently maxed you can see this in ASE / $amkr demand for cowos (they get spillover). But alas without tourists my job would be harder :) $NVDA $nbis"
X Link 2025-10-17T14:50Z [---] followers, [---] engagements
"@JamieFollese @ShanuMathew93 @zephyr_z9 @The_Colonel__ What GPUs are down 2/3 in price recently Any evidence"
X Link 2025-10-20T18:04Z [---] followers, [--] engagements
"@JamieFollese @ShanuMathew93 @zephyr_z9 @The_Colonel__ Availability for all compute is touching new lows. Amzn raising prices on H100s. Yawn. @WarrenPies has some great charts for you and tracks compute availability across the chain. But look anywhere bruh"
X Link 2025-10-20T18:24Z [---] followers, [--] engagements
"@JamieFollese @ShanuMathew93 @zephyr_z9 @The_Colonel__ Btw pretty much all CRWV rev is locked in contracts they build the IRR into the long term rental price and own upside in residual value. Which the evidence for is apparently growing btw hence availability on h100s heading DOWN @NarrativeSeeker has done some good work here"
X Link 2025-10-20T18:28Z [---] followers, [---] engagements
"The cost of inference is collapsing the revenue yielded over the lifetime of a GPU is not. Note as well NVDA backstopped all unsold capacity and new capacity is built on the back of new demand and contracts priced for a lifetime ROI. Btw I heard theres businesses called GCP and Azure that really just rent computers off premises. Same business high level (and yes even CRWV is providing services on top of steel) same hate in the early phases of the businesses and now everyone calls them the best. Weird"
X Link 2025-10-20T19:10Z [---] followers, [--] engagements
"The Google has cheaper AI is a big generalization and I say this as a big fan of $Googl stock. It depends on throughput latency (including p99) time-to-first token and most importantly usable context windows (not the boilerplate). Which is really just a fancy way of saying it depends $nvda $amd"
X Link 2025-10-20T21:45Z [---] followers, [---] engagements
"@JMihaljevic Good analysis Theyre mixing things that dont really go together. Its like buying a $1 ice cream cone and then adding in the cost of the building ice cream machines delivery trucks and shop. And pretending all that was just for your one cone. Silly. $nvda"
X Link 2025-10-21T02:31Z [---] followers, [---] engagements
"@firstadopter Alpha hides where conviction is loudest and understanding is thinnest. Thats large-cap tech and AI today. (Hint: Rorschach test 😝😉) $nvda $nbis $tsla $amd"
X Link 2025-10-21T22:56Z [---] followers, [---] engagements
"@The_AI_Investor Power constrained today = Compute Constrained when the MWs light up. More power = more supply = lower prices = more demand. Same as it ever was. $NVDA"
X Link 2025-11-02T23:28Z [---] followers, [---] engagements
"Do you even read the things you repost Youre like a FUD troll. Read the whole document yourself. He doesnt ask for a bailout for OpenAi. But he does suggest the government should help lower the cost of capital for the ecosystem to encourage investment across the chain including power development. See below. 🤡🤡 @iamgingertrash He doesnt say for them specifically and hes talking about the ecosystem. For example the govt might backstop debt against turbine facilities to lower cost for companies unwilling to take a risk on building (long cycle) capacity in the USA. @iamgingertrash He doesnt"
X Link 2025-11-08T02:15Z [---] followers, [--] engagements
"Sams plan for $1.4T / 30GW hit the tape on 10/28-9. Its implausible the market had already fully priced a 1.4T plan from a 3-yr-old $1320B-rev company. What was priced was some OpenAI build-out (ORCL MSFT capex NVDA orders) and a much fuzzier multi-trillion global AI infra story. Prices would have been much higher if we were pricing in additional 1.4T from just OpenAI let alone everyone else. Instead its Nov people have PNL to defend prices go down narratives form. One player (Sam) after an undoubtedly positive period of pos demand data points from literally everyone else in the space says IF"
X Link 2025-11-16T18:03Z [---] followers, [--] engagements
"I tried hedges. I tried degrossing. Now Im trying Im music. When the Selling Stops $nvda $nbis $crwv @elevenlabsio"
X Link 2025-11-18T04:28Z [---] followers, [----] engagements
"I like $GOOGL (and Gemini 3) but this narrative emerging on X that Gemini is charging more bc the model is better is way off and NOT how the market actually works. Rate cards are a bit theatrical (deals dont clear there). Most API usage is bundled and token pricing is [--] small knob. They also list high prices for 1) perception (hence the X reaction) and 2) to enable discounts (its a sales business after all come on guys duh). Finally Gemini [--] is a serious context heavy multi-modal model that requires a lot of compute / memory. Higher starting pricing also helps ration compute in a constrained"
X Link 2025-11-19T04:03Z [---] followers, [---] engagements
"Macro man is apparently very bad at micro analysis. This post from @paulomarco on $NVDA is emblematic of the really bad takes from tourists on AI that are pervasive across X. One of my rules is control your sources because bad research is expensive (good research pays for itself. To be fair he may indeed have excellent macro ideas (or in this case was just jumping on a quick hot take). That said Last years DSO (days sales outstanding; we cant assume you know what that means) was artificially suppressed by a large $1.7Bn prepayment. NVDA explicitly states this in last years 10-Q. And DSO is"
X Link 2025-11-20T02:25Z [---] followers, [----] engagements
"@PauloMacro Not mine. You did not address the increase in prepayments and you still point out DSO increasing y/y but dont acknowledge its up vs a specific instance in the company called out in their filing 1yr ago. Nice try"
X Link 2025-11-20T03:42Z [---] followers, [---] engagements
"Just for everyone to see. This clown show could not respond at all to my criticism of his point ignoredthat $NVDA saw increase pre-payments in the quarter AND he restated that DSO was up y/y despite the my showing him NVDA gave an explanation for 1x suppression of DSO last year. Just a lot of hand waving. I pointed to his track record a simple Grok search of his posts will show you hes been bearish $NVDA and AI AND all stocks since [----]. Yet he has a paid substack and 175k followers (in a sane world they all know hes a contra). You get OUT what you put IN so vigilantly control your sources."
X Link 2025-11-20T03:55Z [---] followers, [--] engagements
"Just for everyone to see. This clown show could not respond at all to my criticism of his point ignored that $NVDA saw increased pre-payments in the quarter AND he restated that DSO was up y/y despite my showing him NVDA gave an explanation for 1x suppression of DSO last year. Just a lot of hand waving. I pointed to his track record a simple Grok search of his posts will show you hes been bearish $NVDA and AI AND all stocks since [----]. Yet he has a paid substack and 175k followers (in a sane world they all know hes a contra). You get OUT what you put IN so vigilantly control your sources."
X Link 2025-11-20T03:56Z [---] followers, [----] engagements
"The cost advantage people cite for TPU is internal cost have to add a margin on top of it for external users. If you need more compute between now and [----] (since its a race / supply shortage) you call NVDA bc they can deliver much more product at scale. And youre more future proofed w/ NVDAs programmable GPUs if the model paradigm changes (vs being path dependent on googles stack). Both are good but the world remains short compute"
X Link 2025-11-23T02:07Z [----] followers, [----] engagements
"This take confuses OpenAI demand mix shift with no one needs GPUs anymore. In a world where aggregate AI demand supply unused GPUs get repurposed to other models/customers (NBIS Iren ORCL would not be structurally impaired). Also $NBIS MSFT deal is specifically for azure Ai capability. You either believe demand exceeds supply for a long time or you dont and if it does what model wins really doesnt matter. $nvda"
X Link 2025-11-23T02:19Z [---] followers, [---] engagements
"@JRoad123 @benitoz The street has had to jack their next year eps materially every year for 3yrs. Which is to say $nvda is likely much cheaper than low 20s 🫣🫢"
X Link 2025-11-23T02:55Z [----] followers, [---] engagements
"@benitoz @JRoad123 The bubble is in bubble talk"
X Link 2025-11-23T03:14Z [---] followers, [--] engagements
"@mihaljevic @GaryMarcus Not for as long as demand supply. If $googl wanted to sell TPU at scale theyd just be competing for scarce cowos with everyone else"
X Link 2025-11-23T04:07Z [----] followers, [---] engagements
"After watching Age of Disclosure on $AMZN prime it is very clear TPUs are alien technology. $googl $nvda 😉"
X Link 2025-11-23T19:24Z [----] followers, [---] engagements
"@DonutShorts @MotherCabriniNY @buccocapital @RealJimChanos the financing is for land + power + shells + cooling with [----] yr lives. The GPUs will be refreshed multiple times over the 20yrs. Ill take you are a dope for $1 Alex"
X Link 2025-11-24T03:24Z [----] followers, [---] engagements
"@DonutShorts @MotherCabriniNY @buccocapital @RealJimChanos Nice attempt at a kick save But yes they will take retrofit risk and this is factored into their IRR assumptions. But the idea that Meta can stuff them from Cembalest is wrong"
X Link 2025-11-24T13:22Z [----] followers, [--] engagements
"The $GOOGL chase is on. One year ago it was a hated by almost all todays lovers despite excellent numbers and the same full stack advantage (including data). Now those folks think $NVDA is in trouble $META is a dog (despite getting core 15x 27) and OAI will be toast (#1 in App Store still best model for consumers 800mm users). It is important to KEEP THE RECEIPTS and know WHO is WHO in the ZOO $nvda $nbis $googl $meta"
X Link 2025-11-24T17:06Z [----] followers, [----] engagements
"The TPUs = bad for NVDA take is up there with the dumbest yet (maybe worse than DeepSeek) as it completely misses what actually happened in just the last [--] weeks. And I will remember who is who in the zoo. My view 1) Demand for AI is bananas (no one can meet demand everyone is spending more). $GOOGL said just yesterday they have to double capacity every [--] months to keep up. 2) Scaling laws are intact (Gemini 3). The flywheel is about to speed up. Somehow the mid-curve crew thinks this is zero sum competition NONE of this suggests that. 3) if you think the race is hot now wait until you"
X Link 2025-11-25T03:48Z [----] followers, 346.7K engagements
"@joeflaum Econ 101: The pricing floor of a commodity is function of marginal cost (in this case the marginal cost is power)"
X Link 2025-11-25T12:34Z [----] followers, [----] engagements
"$NVDA sells hardware not GPU-hours. Power cost sets the floor for what clouds can charge per token in an oversupplied market. Were in a world now where chips and MWs are scarce so both NVDA and the clouds earn fat margins. A new player mostly displaces other in-house ASIC programs not NVDA if the overall market is very short. And since NVDA can deliver more faster and likely with a better real-world offering (Rubin CPX) once you factor in TCO + ecosystem friction theyll remain the dominant choice. But lets see come 1H26 Place your bets"
X Link 2025-11-25T15:17Z [----] followers, [---] engagements
"I take back my TPU comments Market Gods please 🙏 keep $googl green 😉"
X Link 2025-11-25T16:02Z [----] followers, [----] engagements
"I should have specific that I meant price / margin declines would only happen in an oversupplied world in which case the marginal cost (power) would set the price. But in a world of LT supply constraint (my view) multiple suppliers can and will still earn high margins. And if the supply imbalance is large incremental supply (TPUs) reduces the line out the door but doesnt bring down the price. That would change if total capacity moves close to demand"
X Link 2025-11-25T23:00Z [----] followers, [---] engagements
"@dcyydn NVDA contracts HBM annually and are an insane % of the market so have fun negotiating them down vs stiffing $googl"
X Link 2025-11-25T23:07Z [----] followers, [----] engagements
"@Newell_Street @joeflaum Why do you need decades to justify the stock We dont underwrite decades in any tech stock. No one (especially in the age of AI) has a 20-year guarantee on their current unit economics / mkt position. And yes ASML"
X Link 2025-11-25T23:17Z [----] followers, [---] engagements
"@dcyydn If you are the dominant player in a market you have leverage over your suppliers who will cater to you more than GOOGL on account of your volume. And they contract annually"
X Link 2025-11-25T23:30Z [----] followers, [---] engagements
"@BlestToBePerson @Newell_Street @joeflaum And you are not overpaying today for that risk"
X Link 2025-11-25T23:39Z [----] followers, [--] engagements
"@buccocapital @thexcapitalist Oh no 200bps over risk free what will ORCL do More likely: yield hogs into the cuts will gobble gobble"
X Link 2025-11-28T20:57Z [----] followers, [---] engagements
"@RealJimChanos James River had execution and governance errors and knowingly under reserved for years. Insurance gets built into the fair price and with operating costs so low theres plenty of room for it. Odds are v low youve even ridden in a $Tesla with FSD [--] or later"
X Link 2025-11-30T04:21Z [----] followers, [---] engagements
"$GOOGL this $NVDA that $TPUs are (nano) bananas $OAI cant pay Sam is bad. Meanwhile $TSLA just drives itself. If youre interested in AI rent a $Tesla with the latest FSD for a week. I just did it had zero interventions (highway burbs rain) insane parking-lot nav and of course mad max mode (wow). Its THAT good. And is improving rapidly through recursive learnings as the dataset and compute capabilities (Blackwell clusters) grow. Regardless of what you think about the stock cars drive themselves now and most people dont even really know. Open your minds imagine and get ready. (And yes Waymo"
X Link 2025-12-01T00:36Z [----] followers, [--] engagements
"$GOOGL this $NVDA that $TPUs are (nano) bananas $OAI cant pay Sam is bad. Meanwhile $TSLA just drives itself. If youre interested in AI rent a $TSLA with the latest FSD for a week. I just did it had zero interventions (highway burbs rain) insane parking-lot nav and of course mad max mode (wow). Its THAT good and is improving rapidly through recursive learning as the dataset and compute capabilities (Blackwell clusters) grow. Regardless of what you think about the stock cars drive themselves now and most people dont even really know. Open your minds imagine and get ready. @elonmusk (And"
X Link 2025-12-01T00:37Z [----] followers, [----] engagements
"Everyone debates which AI is smarter but nobodys talking about memory. $GOOGL Gemini is stateless by design (no persistent context). Thats why its fast (and cheaper to serve) but also why its like having an advisor with amnesia. Built to retrieve and orchestrate vs remember. @OpenAI / @sama is going in a MUCH different direction: compounding context into a recursive relationship. The model actually knows you better over time. As an investor I find this more helpful to my process. These are VERY different theories of what AI should be"
X Link 2025-12-04T14:12Z [----] followers, 14.6K engagements
"The edge AI hype ignores memory economics. Valuable tokens (multimodal tool use long context) need a lot of DRAM. In the cloud that capacity is shared and highly utilized = low cost per token. Put it on billions of $$ phones where it sits idle 95% of the time = much higher effective cost per token. But everyone needs a hot take. Btw wheres the $ARM pitch from the edge crew They license the CPU ISA in basically every phone on earth for local Ai and crickets 😂"
X Link 2025-12-10T06:02Z [----] followers, [---] engagements
"@buccocapital I like how you truncated the otherwise very substantive response to create maximum FUD. Good job value add stuff"
X Link 2025-12-11T03:27Z [----] followers, [---] engagements
"Annddd $NVDA is swamped by Chinese H200 orders Huaweii is not the way 😝 China wants $NVDA chips; they just want the real thing not the hand-me-downs. Theyre saying they dont to save face in the event the US doesnt cave. In modern Seinfeld they would likely call this stealing the breakup (I dumped you first). 4/4 China wants $NVDA chips; they just want the real thing not the hand-me-downs. Theyre saying they dont to save face in the event the US doesnt cave. In modern Seinfeld they would likely call this stealing the breakup (I dumped you first). 4/4"
X Link 2025-12-12T13:58Z [----] followers, [----] engagements
"Theres a reflexive loop in $ORCL CDS. Its thin and technical with forced buyers (bank hedging new issue pipeline hedges duration/insurer buyers managing concentration limits supply chain) and far less natural liquidity than people think. Marginal buying pushes it wider then the CDS widening story snowballs. But the reality is cash bonds (6-7%ish) arent signaling anything close to a funding problem. If new paper has to clear closer to 8% thats maybe the best rel-val ever in IG given todays brutal opportunity set. Run the math on the success case for 10y paper with 8y duration.just 150bp"
X Link 2025-12-14T15:56Z [----] followers, 29K engagements
"@Richard89022768 The mkt is short compute in aggregate. Hyperscalers and enterprises are the real demand and they are balance sheet buyers who are capacity-constrained right now"
X Link 2025-12-14T17:51Z [----] followers, [---] engagements
"$AVGO has now erased all TPU hype gains. Easy come easy go"
X Link 2025-12-15T16:05Z [----] followers, [----] engagements
"To be fair $AVGO is the biggest sell in the rebalance this week (127% ADV) followed by $aapl (25%) and $MSFT (20%). $META / $TSLA are large adds. Go figure. Good day for rebal books $AVGO has now erased all TPU hype gains. Easy come easy go. $AVGO has now erased all TPU hype gains. Easy come easy go"
X Link 2025-12-15T19:15Z [----] followers, [----] engagements
"I kind of always suspected the A in AI wasnt for Artificial. $LMT"
X Link 2025-12-16T20:16Z [----] followers, [----] engagements
"So youre saying theres a chance I can use my $AMZN prime in / with GPT Oh my 🫢"
X Link 2025-12-17T02:33Z [----] followers, [----] engagements
"This year Santa probably says MoMo not HoHo 😝 (gulp) *thats the L/S momentum index for the unfamiliar $nvda"
X Link 2025-12-17T18:22Z [----] followers, [----] engagements
"BlueOwl is facing a lot of pressure on other fronts (says more about them than $orcl). Consider in the same article regarding the Abilene DC in TX Blue Owls own targeted returns on that project are as high as 25%. If Michigan looks anything like that (or $ORCL flexes deal terms to match) there will be a v long line to fill Blue Owls shoes at that kind of return (levered equity facing IG credit should earn 15% so the extra [--] gives you a lot of room as an investor to hedge / wedge your own protection and still materially outperform). Said differently PNLs reset to ZERO soon and lot of capital"
X Link 2025-12-17T18:43Z [----] followers, [---] engagements
"Useful signal = helping other investors predict the future vs state the present. Regulatory designation doesnt tell us about real-time capabilities OR trajectory. Compounding AI gains on expanding compute changes the probability distribution fast. Have you used recent versions How did they compare to earlier ones Do you have an opinion on how more coherent clusters may improve the model You may disagree on the answers but skipping those questions leads to incomplete analysis. Also dunking on retail investors is a bit dclass 😉 $tsla"
X Link 2025-12-18T06:46Z [----] followers, [--] engagements
"@tszzl Markets love irony. The AI itself can dissolve your confusion about it if you ask the right questions. And yet all the Bubble takes read like AI generated slop. $nvda"
X Link 2025-12-21T17:04Z [----] followers, [---] engagements
"@edzitron Wrong. $ORCL had a better bid and Blue Owls house is on fire (anyone paying attention can see that)"
X Link 2025-12-23T17:37Z [----] followers, [---] engagements
"@Midnight_Captl #3 yea it turns out demand IS huge BUT you cant meet it - ex Bubble Guy. The deep irony of course makes it the most likely 🤦🏻♂ (FML) $nvda"
X Link 2025-12-24T15:56Z [----] followers, [---] engagements
"@mcuban Tax the abundance so you get less of it Also higher taxes into deflation that will save everyone especially the poor"
X Link 2025-12-25T14:40Z [----] followers, [----] engagements
"Thats the first paragraph of the piece. If you anchor on todays chatbot usage the capex looks high. But no one is building for 3-4 messages a day and I walk through the assumptions of the multi modal ramp. BTW as AI improves = use increases. Historical growth data is from when the products were worse (less coherent compute context windows memory). You dont extrapolate linearly from that"
X Link 2025-12-25T15:54Z [----] followers, [---] engagements
"More like appeal to common sense. You cited Tesla AI5. I showed the full picture which is the same guy building massive GPU clusters saying we need more compute. You also dont seem to know much about what $MSFT is up to (I notice no response there) The existential race framing also implicitly means you think the smartest guys in the space dont know about edge potential and are just racing to the bottom in a pointless charge. While some guy on X (who confuses attention compute with memory bandwidth) knows better"
X Link 2025-12-25T18:40Z [----] followers, [---] engagements
"@shakeorgetbaked @grok evaluate this debate and also the points on efficiency gains from subquadratic"
X Link 2025-12-25T18:44Z [----] followers, [---] engagements
"If distillation is about to solve this why is the guy you cited (putting edge inference in robots) building Colossus AND talking about data centers in space instead of letting competitors waste $600B on stranded capex He could just wait. This is a fair question and not an appeal to authority. Either he knows something you dontor youve cracked what he (and everyone else) hasnt. 🤔🤔🤔"
X Link 2025-12-25T20:50Z [----] followers, [--] engagements
"@mcuban If AI actually causes mass displacement + deflation the last thing you want is contractionary fiscal policy. 🤔 And if it causes a productivity boom the tax base grows anyway. Robot taxes solve a problem that either fixes itself or gets worse with the fix"
X Link 2025-12-25T21:09Z [----] followers, [---] engagements
"Good framing but youre conflating memory bandwidth with memory capacity. Groqs SRAM is fast but tiny (230MB/chip). A 100K context window needs 6-8GB of KV cache alone without it you cant do long context multimodal agentic workloads. You cant really interconnect your way out of a 100x memory capacity gap either. $nvda https://twitter.com/i/web/status/2004343600394744163 https://twitter.com/i/web/status/2004343600394744163"
X Link 2025-12-26T00:08Z [----] followers, [----] engagements
"@Midnight_Captl @benitoz Cool point"
X Link 2025-12-26T02:29Z [----] followers, [---] engagements
"Famously difficult to raise capital when there are [--] buyers for every unit you can produce 😂😝 $ORCL $crwv $NBIS From an interview with an exec right in the middle of the data center buildout selling power contracts to the hypersalers neoclouds and shell builders. The demand ratio for GPUs is [--] to [--] projected to go to [--] to [--]. I thought there was no demand 🤣 https://t.co/HsP1cn171v From an interview with an exec right in the middle of the data center buildout selling power contracts to the hypersalers neoclouds and shell builders. The demand ratio for GPUs is [--] to [--] projected to go to [--] to"
X Link 2025-12-26T16:58Z [----] followers, [----] engagements
"@ShanuMathew93 I suspect it refers to nameplate capacity of equipment sold (often backup gensets / power systems) as opposed to 39GW of new always-on generation"
X Link 2025-12-27T21:50Z [----] followers, [---] engagements
"Nice engagement bait But SoftBank has $33Bn cash post $NVDA sale to fund OAI. They have other assets worth 10s of billions too (TMUS PayPay). The $ARM loan is 8.5% LTV today 😂😂. Even if they maxed out the facility AND $Arm dropped 40% it would be 33% LTV (extremely well collateralized). For 33% to trigger a call youd need loan terms so tight the facility would be unusable from day one (terms contemplate equity vol duh). Also the Archegos comp is nonsense. That was hidden leverage via swaps where counterparties didnt know each others exposure. This is [--] banks knowingly syndicating a"
X Link 2025-12-28T15:24Z [----] followers, [----] engagements
"At the risk of narrative violation looks like theres plenty of room for AI to grab more IG share $orcl $meta $googl $nvda"
X Link 2025-12-28T19:35Z [----] followers, [----] engagements
"@Midnight_Captl @OmerCheeema This is only bc the public treats inference as a monolithic task rather than a spectrum of complexity scale and value. If you understand the gradients and that $nvda is a full system (vs a GPU) built to handle it all efficiently it should strengthen your view of their moat"
X Link 2026-01-01T18:57Z [----] followers, [---] engagements
"@packyM coined a frame thats been stuck in my head since 2021: compounding crazy. Innovation compounds and the time between thats impossible and thats normal keeps shrinking. In his words discoveries become inventions become building blocks ad infinitum. AI is the cleanest example. Chatbots were always just the first visual primitive. And as with past tech diffusion innovation was always going to drive costs down and use up. Adoption was also always going to lag the frontier and move slower than the new tech capability. That gap is where disillusionment periods come from. The news of the last"
X Link 2026-01-06T15:00Z [----] followers, [---] engagements
"@DarioCpx Nothing to see here except a basic timing amendment on a project finance loan for an already disclosed slight delay (heard weather delays are common in construction). I get it though easy FUD since credit stuff seems scary to outsiders. 🤷🏻♂. $CRWV"
X Link 2026-01-06T17:23Z [----] followers, [---] engagements
"This is extremely deceptive and objectively false (if you dont understand finance fine; if you do then I consider it evil). $CRWV does not have 4mo of liquidity. They have numerous delayed draw term loans tied to site specific capex. Taking the cash balance and dividing by neg free cash flow to get 4mo ignores the 3rd section of the cash flow statement (LOL). And the amendment is to align the start of a covenant calc with customer cash flow (1mo delayed data center already known). Silly. https://twitter.com/i/web/status/2008728857717739832 https://twitter.com/i/web/status/2008728857717739832"
X Link 2026-01-07T02:34Z [----] followers, 15.2K engagements
"Behind the Meter is the way. Great sleuthing here Hat tip. $NBIS TL/DR: A fully islanded engine-plus-flywheel power architecture would meaningfully compress time-to-power and allow $NBIS to bypass utility interconnection bottlenecks that are increasingly acting as market gatekeepers for AI infrastructure. In a world where bring-your-own-generation is becoming table stakes this would represent a quiet but structurally important advantage $NBIS - EVIDENCE SUGGESTS VINELAND SITE MAY BE FIRST OF ITS KIND FULLY OFF GRID HYPER SCALE DATA CENTER Evidence points to the systemic innovation teased for"
X Link 2026-01-08T16:56Z [----] followers, [----] engagements
"Ive been thinking a lot about the recent developments in AI post CES and how to contextualize them. Interestingly I think whats happening now mirrors the evolution of human civilization and understanding it in this context helps explain why this moment is potentially so pivotal. Big Bang: Transformers paper etc. The fundamental physics of intelligence changed. Early humans: First LLMs learning basic patterns. Fire language simple tools. Childhood to adolescence: Knowledge compounding. Building structures developing curiosity about the world. GPT-2 to GPT-4 is basically elementary school"
X Link 2026-01-09T17:07Z [----] followers, [----] engagements
"@buccocapital What if you own a stock thats kind of both $INTC 😂"
X Link 2026-01-10T03:30Z [----] followers, [---] engagements
"Excellent points but none of this should be new to anyone doing the work on this theme. The fact that it is shows there is indeed edge hidden in plain sight $crwv CoreWeave $CRWV CEO Michael Intrator on the @BigTechPod providing what I believe is the most compelling rebuttal to the narrative that GPUs only have a [--] to 3-year useful life. CoreWeave's customers are the world's most sophisticated users of GPUs and they are signing https://t.co/SatqbEDVZO CoreWeave $CRWV CEO Michael Intrator on the @BigTechPod providing what I believe is the most compelling rebuttal to the narrative that GPUs"
X Link 2026-01-10T18:31Z [----] followers, [----] engagements
"@Midnight_Captl @jukan05 I think [----] Brain Disease is very sad and we should be kinder to him. $crwv $orcl"
X Link 2026-01-10T22:45Z [----] followers, [---] engagements
"@ContrarianCurse Pre EPS vol is cheap in all mega caps though and liquid call spreads are much better. Same digital call (tight spread) pays 45-50:1 in options and is way more scalable. $googl"
X Link 2026-01-12T00:44Z [----] followers, [----] engagements
"I have a solution for the $GOOGL vs $NVDA debate: Own both (Its not zero sum duh). TPUs are excellent chips especially for $GOOGL. Plus their distribution & multi-modal data flywheel is unmatched. BUT $NVDAs system keeps getting more robustRubin architecture the Groq acqui-hireand is hard to beat. Their build capacity + upgrade cadence also keeps customers loyal (stray too far Jensen pulls allocation). The world is just short compute and both will sell all they can make for a long time. https://twitter.com/i/web/status/2011237127510442090 https://twitter.com/i/web/status/2011237127510442090"
X Link 2026-01-14T00:41Z [----] followers, [----] engagements
"I think hes missing the plot. GPT [--] was a system upgrade (routing + orchestration). It improved reliability and task horizon. A headline IQ jump wasnt the point. The next leg is RL + agents tightening the recursive loop and disaggregated DCs making it scalable (workload routing tiered memory improved networking). None of that suggests slowing. $nvda"
X Link 2026-01-18T02:14Z [----] followers, [---] engagements
"I wonder what AI lab needs some money https://www.reuters.com/business/media-telecom/japan-us-narrow-first-550-bln-investment-picks-including-softbank-linked-plan-2026-01-19/ https://www.reuters.com/business/media-telecom/japan-us-narrow-first-550-bln-investment-picks-including-softbank-linked-plan-2026-01-19/"
X Link 2026-01-20T01:05Z [----] followers, [----] engagements
"The race for power was always destined for keep-away behavior and I believe this is accelerating in earnest After Claude Code (ACC @TMTbreakout). A reinvigorated model race + increased data center coherence (disaggregation tiered memory) makes every incremental unit of compute more valuable. More must-win product moments shorter cycle times new demand unlocks. More unlocks = more tokens = more power demand. Hyperscalers GPU vendors ( $AMD) and others ( $AAPL rumors) are now buying to meet their own growing needs AND make sure the competition cant (zero sum). Bitcoin miners and neos with"
X Link 2026-01-23T17:03Z [----] followers, [----] engagements
"@quantszn @TMTBreakout I like the whole sector Each of them has their own haters / bear pitch I think important to zoom out. All boats rise"
X Link 2026-01-23T19:21Z [----] followers, [--] engagements
"@JaredKubin Also his data is wrong. The gold standard IMF data (linked) sows 57% as of 3Q25. https://data.imf.org/en/news/imf%20data%20brief%20december%2019 https://data.imf.org/en/news/imf%20data%20brief%20december%2019"
X Link 2026-01-24T14:12Z [----] followers, [---] engagements
"$NVDA as central bank of AI. What a novel idea $crwv :) If the Fed were public itd be a multi-trillion-dollar entity controlling liquidity (M2).$NVDA plays the same role in AI central bank of compute.It issues the base currency (GPUs) and does QE (indirectly buys its own stuff) so the ecosystem can expand and multiply. The If the Fed were public itd be a multi-trillion-dollar entity controlling liquidity (M2).$NVDA plays the same role in AI central bank of compute.It issues the base currency (GPUs) and does QE (indirectly buys its own stuff) so the ecosystem can expand and multiply. The"
X Link 2026-01-26T13:27Z [----] followers, [----] engagements
"Interesting that when @sama made equivalently bold predictions people had such a different reaction. Narratives are powerful indeed. Dario predicts trillions in AI revenue likely before [----] and country of geniuses by [----]. I think no one can estimate what this really means. https://t.co/9Phuw78uIP Dario predicts trillions in AI revenue likely before [----] and country of geniuses by [----]. I think no one can estimate what this really means. https://t.co/9Phuw78uIP"
X Link 2026-02-14T16:07Z [----] followers, [----] engagements
"I can teach to the test better than you Bytedance is putting pressure on the American Frontier Labs. China is gaining ground every day. And DeepSeek v4 hasn't even been released yet. Bytedance hasn't just recently revolutionized text-to-video; they're also catching up in regular LLMs. https://t.co/xOmAIOFuGu Bytedance is putting pressure on the American Frontier Labs. China is gaining ground every day. And DeepSeek v4 hasn't even been released yet. Bytedance hasn't just recently revolutionized text-to-video; they're also catching up in regular LLMs. https://t.co/xOmAIOFuGu"
X Link 2026-02-14T18:27Z [----] followers, [---] engagements
"@stevehou Hmm. GOOGL could do $225m EBITDA in [--] leaving it $30-35bn in FCF and they could cut capex in 1/2 and repay all debt in around [--] quarters just for context. Not really scary to have 1/2 turn of leverage during investment cycle. 🤷🏻♂"
X Link 2026-02-15T18:55Z [----] followers, [---] engagements
"We havent seen a blowout capex number like this since insert any quarter $googl $hut"
X Link 2026-02-04T21:16Z [----] followers, [----] engagements
"I think a few things are going on w/ $NVDA (and its mega cap brethren) 1) Lag [--] are all sources of funds 2) burdened by its own success / size 3) AI bulls can get more juice / asymmetry elsewhere given [--] + [--] The other narratives are so conflicting in context of the broader market I find them less compelling. Punish the spenders on ROI yet AI disruption is decimating entire sectors 🤔. I suspect theres a reflexive loop of quant vigilantism (press Ai losers) HF pain and narrative backfill. A little time and maybe some pre NVDA eps FOMO probably helps."
X Link 2026-02-14T01:41Z [----] followers, [---] engagements
"AI regulation is climate [---]. New establishment policy will run entirely through this prism. And China is the new attack on democracy imo. Its a frame that tries to make opposition radioactive for folks who disagree. If preventing China from accessing our chips and winning Ai (whats that mean btw) is so existential shouldnt the govt be doing way more at home to advance the cause Yet one mention of a govt backstop and people had a melt down. Why is Dario Amodei so weirdly hawkish on the Chinese https://t.co/DvXZzNhoxK Why is Dario Amodei so weirdly hawkish on the Chinese https://t.co/DvXZzNhoxK"
X Link 2026-02-14T02:05Z [----] followers, [----] engagements
"AIs Shadow Output Gap While Washington obsesses over debt and inflation AI is already ushering in an age of abundance (Part 1) The political and economic establishment cant stop talking about deficits debt and the CPI. Capitol Hill hearings FOMC minutes and financial news all pulse to the same beat. Yet this fixation ironically coincides with the arrival of the most powerful productivity engine in human history: generative AI. Its impact is creating a shadow output gap an invisible but rapidly widening expansion of supply-side capacity. Policymakers especially at the Federal Reserve act as if"
X Link 2025-07-07T23:57Z [----] followers, 74.9K engagements
"The New Circularity: Ai is compressing software terminal values. But to displace software you need enormous compute memory and power. The lost value should accrue to the physical stack. And the market isnt pricing in nearly enough of that migration. The problem is that we remain in a shoot first regime. Few participants have first principles for how AI works. DeepSeek in 1Q25 was the template: incomplete understanding turns every headline into a FUD crisis. A super BOOM with micro busts if you will. This dynamic may be here to stay since nobodys sure exactly whos swimming naked. As my old"
X Link 2026-02-03T15:59Z [----] followers, 18.8K engagements
"Mr Regulation is also Mr Responsible how convenient. And of course actions words always: TPUs = GPUs. Panic bidding power. Upsizing capital raise. Starting beauty pageant for IPO. 🤔🤔 Dwarkesh asks Dario a fantastic question relating to how he is so bullish on AGI yet so conservative on data center build out - Dario has an amazing take on this: Dario Amodei details the staggering financial risk of the AI race explaining that if growth continues at 10x a https://t.co/Ed99X88zwS Dwarkesh asks Dario a fantastic question relating to how he is so bullish on AGI yet so conservative on data center"
X Link 2026-02-14T00:32Z [----] followers, [----] engagements
"In the spirit of narrative consistency I have to ask is this good or bad for OnlyFans 🤔 Looks great Looks great"
X Link 2026-02-14T18:33Z [----] followers, [----] engagements
"@evrgn11112231 @GavinSBaker Forgot to mention we are still punishing the spenders on ROI questions at the same time 🤯"
X Link 2026-02-16T04:45Z [----] followers, [---] engagements
"@JonahDispatch Oh And HRC was also contesting the results Oh wait she called to concede. Got it"
X Link 2020-11-16T03:23Z [----] followers, [--] engagements
"@carlquintanilla And what did the inflation reduction act do"
X Link 2024-08-24T01:12Z [----] followers, [---] engagements
"@RichardHanania how do you know the context of the eulogy line If you are going to go there you should at least make sure you know the context. For example are you sure the context wasnt about her depression she told me everything was her fault and I assured her it was all mine. Whether thats the context or not I dont know but neither do you. And if you are going to incorporate something very personal like that you should maybe include it or be sure"
X Link 2024-08-26T02:18Z [----] followers, [----] engagements
"@RichardHanania The NYpost article gives a good sense for the context and I think you know you are off or you didnt even read it. Sad"
X Link 2024-08-26T02:21Z [----] followers, [---] engagements
"$CHWY $WOOF largest retailers of pet protection in the USA are surging Protect your dog @elonmusk #Debate2024"
X Link 2024-09-11T16:55Z [----] followers, [---] engagements
"@mcuban @RobertJSalvador Trump manufacturing slowed towards end of [----] because of [--] fed hikes(which the fed admitted later were a mistake). @mcuban thanks for reminding everyone that normal IQ people can become billionaires too :)"
X Link 2024-09-25T03:51Z [----] followers, [---] engagements
"@amitisinvesting I was at dinner with Jensen. It isnt up bc of ROI its up because people are learning about time-inference scaling and the dawn of the AI flywheel. A second and independent scaling law has emerged that will unlock the agentic use case. Many on wall st dont understand the implications of Strawberry Jensen spent time educating. $NVDA $ARM #ai #tech"
X Link 2024-10-14T02:13Z [----] followers, [---] engagements
"Inference Time Scaling will require more General Purpose compute that (dumb) consensus believes this stems from 1) needing much faster reasoning (see o3 time performance) 2) speed of and amount of tokenization and 3) substantial memory needs to increase context windows (KV Cache) as a model that cant remember very far back is materially less helpful (duh). Dylan Patel (@SemiAnalysis_) speaks extensively about this. The increased memory and compute needs will make system level solutions (NVDA) the gold standard (more efficient data transfer power use etc when everything is created together)."
X Link 2025-01-04T21:54Z [----] followers, [---] engagements
"@Taxidriver1031 Its a growth stock thats ebitda positive and trades at 1.5x ARR. People who have zero clue will post anything on something they see on X or Reddit. Get an education clown"
X Link 2025-01-15T03:44Z [----] followers, [--] engagements
"@GavinSBaker @GavinSBaker Japan has the $$ (GPIF). Masa = Japan: Japan is doing FDI into USS hence no more discussion of tariffs on Japanese. So this SoftBank doesnt have the money is missing the v obvious geopolitical context"
X Link 2025-01-23T03:16Z [----] followers, [----] engagements
"Theres a lot of talk about DeepSeek and the impact on NVDA. $5mm training costs (Which is a lie they have 50k+ H100s but cant disclose bc of export control laws on chips to China). But people are also *conflating cost of training with the cost of running improving and scaling AI based services. This is why Google trains the latest Gemini for $150mm but spends $60Bn on capex. Ironically you can explore all of this in ChatGpt 4o which can explain it to any novice the Ai has the answer about. the Ai Heres the GPTs overview on the topic. - Absolutely youre spot on. The discussion around AI costs"
X Link 2025-01-25T04:10Z [----] followers, [----] engagements
"More likely folks are fitting a narrative to todays $NVDA price action the stock was trapped by options gamma in the low $140s + TXN missed huge (backtest all semis against TXN whiffs whether its relevant to AI or not) $nvda $tsla $meta $spy @GavinSBaker"
X Link 2025-01-25T04:10Z [----] followers, [---] engagements
"Because a Chinese HF bro quant firm fake trained a model and said it cost $5mm Googl trained the latest Gemini on 150mm yet is spending 60bn in capex. Because the cost of Ai goes materially beyond training when you want offer a scaled improving inferencing service. How you are using the fake training cost (and therefore negative gross margin) copycat to say multiple large tech CEOs are morons for not knowing they are wasting money is a bit weird. Your tweet will not age well"
X Link 2025-01-25T06:37Z [----] followers, [----] engagements
"The efficiency gain is a lie brother. The training cost is not $5mm. Read their own paper they tell you what they leave out on research and data cost not to mention they have access to thousands of h100s via grey market and neoclouds. You have a serious follower base propogating this stuff is v evil"
X Link 2025-01-25T07:03Z [----] followers, [----] engagements
"Haha. Wrong. Inference time scaling requires integrated system level approach as memory / context windows grow. There is no deepseek is bullish for lower costs their costs are not low and the techniques are not new. They have a distilled model that uses MoE if you cant explain those things right now then you should wonder whether you have a handle on the situation"
X Link 2025-01-25T23:48Z [----] followers, [---] engagements
"DeepSeek isnt bearish for AIits rocket fuel. Cheaper derivative (distilled) models expand the TAM driving more users use cases and exponential inference demand. NEW Scaling Laws (efficiency = more compute NOT less) make this bullish for $NVDA which powers it all neoclouds that scale AI workloads ($NBIS) and the entire ecosystem. The AI revolution just accelerated. Also a counterfactual to ponder: would DeepSeek turn down a [--] million Blackwell GPU cluster or happily use it The answer is important. $spy $qqq $tsla @OpenAI @Banana3Stocks @GavinSBaker"
X Link 2025-01-26T17:47Z [----] followers, [---] engagements
"@Citrini7 Good luck on your other AI tomorrow as if it will outperform. I like how you deleted your prior work experience on LinkedIn by the way ambulance guy in college = Citrini research with a 7yr gap. Whats your AUM again Sick track () scalable ideas runs no $$. Weird"
X Link 2025-01-27T04:31Z [----] followers, [---] engagements
"$AAPL became a consensus short very quickly (valuation bruh). What do shorts miss Phone revenue has gone nowhere Myopic. Thats in the face of a strong dollar which acts as a price cut in EM (I.e. Tim Apple cut price to take some share and once Apple gets you you never leave). Short term strategic pain for long term gain. I see $AAPL as a bond with a coupon that grows (eps + buyback) noting they often finance below UST Also new LLM innovations re: memory improve the AI edge case making the future of Apple intelligence bright. There are of course market cap limits / walls ($200 was hard to"
X Link 2025-02-01T00:13Z [----] followers, [---] engagements
"For example DeepSeek basically jailbroke an $NVDA GPUlike a hacker wiping iOS off an iPhone and running custom software. Cool in theory but not scalable. Almost no one has the technical skill to do it and most people still need the App Store (CUDAs libraries) to actually get things done. Plus it breaks compatibility with everything built for iOS making it impractical for most users. Also note Jensen has been predicting a 1000x drop in the cost of inference. This is also means hes thought it through and planned for it (he told me and a group of other CIOs this directly at a small dinner in"
X Link 2025-02-01T19:40Z [----] followers, [---] engagements
"$NBIS has the most robust catalyst path in AI and is a pure play bet on 1) the rise of AI agents (the customers are new companies building these agents) and 2) lower cost of inference creating more demand (see $AMZN call). Also the stakes in AV Ride and ClickHouse are material ($1.5-2Bn in value) and reduce the core biz ARR multiple (net of cash) to 1-1.5x. Coreweave will come in the low double digit range. Of course the cash will be used to invest in more capacity (expect an update by Arkaday here in 1H25 with a goal to reach [--] GW of capacity in 2026). The difference between $NBIS and the"
X Link 2025-02-08T16:19Z [----] followers, [---] engagements
"$NBIS is the most undercover AI opportunity in the market imo. It also has the most robust catalyst path in AI and is a pure play bet on 1) the rise ofagents (the customers are new companies building these agents) and 2) lower cost of inference creating more demand (see $AMZN call). Note the stakes in AV Ride and ClickHouse are material and reduce the core biz ARR multiple (net of cash) to 2.5-3x. Coreweave will IPO in the double digit range. Of course the cash will be used to invest in more capacity (expect an update by Arkaday here in 1H25 with a likely goal to reach [--] GW of capacity in"
X Link 2025-02-08T18:41Z [----] followers, [---] engagements
"Housing costs down + stocks down = consumer net worth down. Most of main st owns a home theyll be psyched their levered asset is lower in price Asset deflation is good for employment You walked into the casino at Facebook and got lucky. Stick to the podcast leave economics and markets to people who at least have basic understanding"
X Link 2025-02-09T04:40Z [----] followers, [----] engagements
"@RonFilipkowski Like all the dipshits that believed Covid was from a lab More embarrassing than falling for an irrelevant con from Dinesh. #moron"
X Link 2025-02-15T00:23Z [----] followers, [---] engagements
"$NBIS Remember the company is likely to do a large (very bullish) growth capital transaction in [----] (time is of the essence). They won't increase the ARR guide before then -- they'll want to do that when they announce the deal in part for the benefit of the new (or additional) investors (duh). New capital is the path to more compute and an eventual "AI as a Service" cloud platform. This is essentially a private company that happens to be public and so far they've done a Series A (very unique for public investors to have access to something like this). Until the "Series B" capital raise"
X Link 2025-02-20T12:53Z [----] followers, [---] engagements
"People are way too jumpy about AI datapoints theres a lack of first principle analysis and deep skepticism/fear which is actually bullish (Wall of Worry). DeepSeek freakouts showed us who was swimming naked (most people). Its a shoot first and ask s later mentality. This of course is the opportunity for those who take time to understand (which ironically just means asking #Grok3 to help you) For example on the latest datapoint freakout re MSFTs rumored data center pullback: $MSFT $NVDA $NBIS 1.Satyas podcast is in fact massively bullish on AI. Listen to it yourself with an open mind."
X Link 2025-02-22T17:58Z [----] followers, [---] engagements
"Dispersion (stock picking is back baby) is the buzzword Ive heard most at CIO dinners in [----] (kind of like if we say it enough it might come true). But Im starting to think the market will return to narrow leadership large cap tech cash cows + Ai. TPU (trade policy uncertainty NOT a custom ASIC 😉) & DOGE (Deflation of Growth Expectations 😉😉) likely slow the economy (i.e. no trade deals tariffs as revenue lower aggregate demand). In that backdrop the back end likely rallies ($TLT for the win) helping accelerate money flows into large cap tech bond proxies (mag 7) and secular winners"
X Link 2025-02-24T04:00Z [----] followers, [---] engagements
"@StockMarketNerd Dude the cuts are 1% of [----] build plan all tied to not doing training for OpenAi and/or inability liquid cool on site. The Cowen analyst held a call Friday night and Sunday night confirming it. Your analysis is very embarrassing. Sorry to be so blunt"
X Link 2025-02-24T04:28Z [----] followers, [---] engagements
"How embarrassing it is to have so many followers and and not know that when sales occur late in a quarter on a new product launch actual payment can spill over into January while revenue is recognized when the board ships thereby increasing DSO. Blackwell is the largest new product launch ever ($11Bn in 4Q vs the street looking for $4-6Bn) and the entire investment community knows it was December weighted. That you somehow think this is a sign of stretched customers who all just increased capex another 20% above street and claim they are capacity constrained on AI is absolutely disqualifying"
X Link 2025-02-27T02:48Z [----] followers, [----] engagements
"$NVDA EPS: heavy wears the crown. The companys past success has broken investor brains expectations for a mindblowingly explosive quarter every report are distracting many from the extraordinary nominal dollar gains the company is generating. 1) the market spent 6mo hand wringing over Blackwell delays well Blackwell is here and did $11Bn in 4Q with the street expecting $4 to $6. 2) the CIO dinner smart consensus take pitched to me countless times was theyll miss the quarter and the guide for first time but you want to buy the dip bc the back half is bananastan well they beat the quarter and"
X Link 2025-02-27T04:22Z [----] followers, [---] engagements
"Anyone else feel so bearish today that it actually made them bullish Asking for a friend. $NVDA"
X Link 2025-02-28T00:16Z [----] followers, [---] engagements
"Google monetizes low quality queries which are sticky auto insurance options attorney near me which coordinates well with maps and search on $GOOGL returns YouTube videos (lasting / real knowledge repository aka token advantage). Ive noticed a behavioral shift similar to yours and yet $GOOGL grew search 12% two quarters in a row. But the Peter Lynch in me is nervous. Valuation though may protect us 15x [----] is a 7yr fwd multiple trough. PS even if search is a MSD grower $googl is still better than 1/2 the biz in S&P. For example as you said YouTube is undermonetized as an ad biz and AI asset."
X Link 2025-03-01T00:24Z [----] followers, [---] engagements
"Spread trade suddenly YOURE a wall st trader. The irony. By wall st do you mean the 401k and pension class Thinking you can bifurcate between the two and know exactly when wall st weakness impacts main st and that we wont cross that point is hubris. Lot of Americans will suffer if you calculate the inertia of a large economic tanker (USA) wrong"
X Link 2025-03-09T19:42Z [----] followers, [----] engagements
"Lets take stock of what we are trying to fix with the global trade reordering [--]. Unemployment 4% [--]. Wage gains [---] m/m [--]. Market Based PCE (ex portfolio mgmt) 2% y/y with lagged housing coming down [--]. US Led Secular Tech Boom 🤔 Actual problems: [--]. border security [--]. Govt waste / wokeism [--]. Housing availability (caused by #1 + high rates) Why are we trying to fix the latter by undoing the former De regulation is great of course but trade policy uncertainty is roughy equivalent and will negatively impact growth. And tariffs are a tax and deflationary (thats what you call it when the government"
X Link 2025-03-11T02:10Z [----] followers, [---] engagements
"After multiple weeks on the pain train with a ticket to pound town its time to reflect. Rather than guessing on Trade Policy Im focusing on questions I can actually answer: 1) Will demand for all things AI be higher or lower by year-end HIGHER. $NVDA s extended "period of disillusionment" (sideways since last summer) and DeepSeek uncertainty set a classic bear trap just as compute demand is accelerating (contrary to concerns). The AI Flywheel is spinning faster than most realize (from agents to robots) and the negative news drum beat is likely to flip positive on the King (NVDA) as spring"
X Link 2025-03-15T15:13Z [----] followers, [---] engagements
"The reflexivity coefficient between markets and the real economy is approaching 1.0market swings directly impact real-world behavior faster than ever (consequence of household equity exposure / high information flow) Housekeeping data point: A friend rolling up accounting firms shared this weekend Were seeing broad-based caution from SMB clients uniformly across our accounting business. CEOs feeling cautious heading into Q1 earnings have every incentive to temper expectations especially with stocks already down. But AI could be a notable exception. Crucially big AI players report later"
X Link 2025-03-16T14:53Z [----] followers, [----] engagements
"$NVDA. Why the sleepy response post GTC Some thoughts. 1) GTC historically is a boondoggle for nerds who get excited by new acronyms that most people will never (or need to) understand 2) Jensen is burdened by his own meteoric success -- heavy wears the crown of impossible expectations 3) He is asking people to imagine an exponential future while most investors think linearly and are naturally skeptical because of #2 (this must be I think his focus on economics and token math is an attempt to help start exponential thinking. What I hear: - reasoning requires a lot more tokens (more compute);"
X Link 2025-03-19T22:59Z [----] followers, [---] engagements
"@Benioff @UnitreeRobotics That robot costs 50-100k per the website"
X Link 2025-03-22T17:03Z [----] followers, [----] engagements
"@HyperTechInvest LOL"
X Link 2025-03-23T01:35Z [----] followers, [----] engagements
"The current administration needs a broader definition of trade. Its not just goodsits also financial assets. In that sense the U.S. runs a massive trade surplus: We import goods they buy our bonds and stockslowering our cost of capital and reinforcing USD as the global reserve. That system powered the U.S. economic machine: It let companies like Amazon burn cash for a decade reach escape velocity and change the worldwith low rates and deep capital markets. Ironically the admin doesnt acknowledge the trickle-down value of that Even while pushing deregulation capex incentives and corporate tax"
X Link 2025-04-01T04:12Z [----] followers, [---] engagements
"Almost every AI signal lately has been bullish but the markets latched onto one cautious datapoint from MSFT (even though GS raised MSFT [----] capex recently after meeting with them). Flip the scenarioMSFT bullish everyone else soft would the market be excited Of course not. This is a Rorschach test. People want to sell so theyre cherry-picking reasons. Belief will return when the macro vol fades. $NVDA"
X Link 2025-04-02T01:24Z [----] followers, [----] engagements
"@chamath @amirpc Theres multiple teams at Wedbush with multiple analysts and they likely speak to 100s of analysts management teams and expert calls every week given the number of people. That would be a research shop doing its job. Lol Dead"
X Link 2025-04-05T22:09Z [----] followers, [----] engagements
"@AravSrinivas TikTok for finance. 30-60 second clips with research analysts"
X Link 2025-04-12T01:40Z [----] followers, [----] engagements
"@zerohedge *explanation not exclamation"
X Link 2025-04-16T18:03Z [----] followers, [----] engagements
"Random thought: OpenAIs first mover advantage + new unlimited memory = contextual lock-in. Just like Microsoft shipped the PC OS six months earlier and rode that head start into a multi-trillion-dollar monopoly OpenAI is quietly building the most valuable memory graph in history: your data your preferences your interactionsforever. $nvda @OpenAI"
X Link 2025-04-19T14:03Z [----] followers, [---] engagements
"$GOOGL I wonder if we should ditch the ad market valuation (search is screwed) and value it as the only big tech company that owns end-user context the model and real custom silicon (Jensen will tell you he respects TPU but not the others). That combination is likely to increase in value as the AI Flywheel spins faster Also YouTube is an index of cultural memory: [---] hrs/min uploaded the only proprietary video RLHF corpus at global scale. Competitors LLMs may have to license that data eventually for true multi-modal capabilities. Oh it also trades at a notable discount to S&P and its bad biz"
X Link 2025-04-26T00:28Z [----] followers, [---] engagements
"I can see it now: DeepSeek R2 crushes compute needs Huge breakthrough The Great @DanielTNiles linked a presentation by another guy named Niles which claimed to analyze all the latest DeepSeek papers to preview their amazing breakthroughs. $nvda @OpenAI @sama Great headlinesback to reality oh there goes gravity 👇 [--]. Sparse-expert tricks arent so special Only 1-in-20 experts fire per token yet RAM & network pipes stay busy so real-world speed gains are way lower [--]. Training is still pricey Cheaper forward passes dont spare the trillion-token data consumption or months of reward-tuning runs."
X Link 2025-05-09T22:37Z [----] followers, 57.4K engagements
"You are confusing fair value thinking with cost-basis GAAP. You have apparently never heard that you should study accounting before commenting on it :). $CRWV doesnt take a mark to market / impairment because secondary market chips trade cheaper to carrying value. Under GAAP GPUs sit in PP&E at historical cost less accumulated depreciation in this case straight line depreciation. The accounting firm will only require a write-down if an impairment test shows the entire cash generating asset cant earn back its carrying value (revenue has to comfortably cover book value). Also the impairment"
X Link 2025-05-18T13:31Z [----] followers, [---] engagements
"All of the hot takes on the coming US power shortage in response to the Senate bill solar / wind subsidy changes are far too alarmist. An alternative view: - We are short GPUs today NOT power. Power in the US is a merchant market when demand shows up prices spike new capacity comes on. You can see this in PJM mkt dynamics. When the reserve cushion got thin last year (7x prior price) [--] GW of quick start capacity instantly moved to construction. A greenfield CCGT (combined cycle gas) can be built in [--] months and there are DOZENS of 100-MW-plus brownfield sites SITTING EMPTY today waiting for"
X Link 2025-06-29T20:13Z [----] followers, [---] engagements
"I view the housing as the last covid remnant. Existing supply at 40yr lows because of the great refi at zero in 2020-2021. This created fiscal dominance for the first time ever in the USA. Housing prices UP into high rates diluted the impact of the monetary channel on aggregate demand. Interestingly this also means cutting rates may provide less stimulus than in past (housing may hold steady or decline rather than go up)"
X Link 2025-07-26T13:22Z [----] followers, [----] engagements
"@HedgeyeComm @FishtownCap The legacy biz is growing slower than peers bc its 25% larger AI is growing triple digits for Amzn and still below 10%. If you can show their Ai business is growing slower than others great"
X Link 2025-08-01T04:21Z [----] followers, [---] engagements
"Theres a lot of discussion about the Chinese energy advantage (3x the power grid) so I think its time to set the record straight. The TL/DR is that Token per grid MWh is the only measurement that matters and the Chinese system is an ORDER OF MAGNITUDE behind despite their total size. Our advantage is also compounding (faster chip innovation from a higher base) making it difficult to overcome. $NVDA (Note some people will debate about the exactness of numbers like anything its imperfect but the framework and direction are right). A quick thread:"
X Link 2025-08-05T03:24Z [----] followers, [---] engagements
"SemiAnalysis note post mkt highlighting nvda cgx. My summary: $NVDA out with a new leap forward in compute in the last 24hrs called $NVDA CGX. Its a monster leap for inference efficiency and will be available as part of Rubin (after Blackwell). Its 67x improvement and unlocks extremely long context windows by splitting inference into [--] parts pre-fill and decode. In current architectures the model has to re-read the book so to speak at multiple points in an interaction slowing things down including throughput. In this case one accelerator indexes the book (pre-fill) and then all decode can"
X Link 2025-09-10T21:15Z [----] followers, [---] engagements
"1/ The $NVDA / OpenAI deal tells me were closer to the beginning of the AI boom than the end. And todays meager stock reactions across the ecosystem suggests market skepticism or ignorance. Either way its opportunity. (1/5)"
X Link 2025-09-23T03:35Z [----] followers, [---] engagements
"4/To all the Private Credit guys whining about having too much GPU exposure Jensen just debunked your biggest fear. Hes providing credit support by putting in $100Bn equity capital BEHIND YOU. Perhaps actually you have too little"
X Link 2025-09-23T03:36Z [----] followers, [---] engagements
"Everyone is upset about the $NVDA ponzi / circular deals Id love to hear their thoughts on this other big one called fractional banking. Supposedly that one is really something"
X Link 2025-09-23T17:09Z [----] followers, [---] engagements
"China wants $NVDA chips; they just want the real thing not the hand-me-downs. Theyre saying they dont to save face in the event the US doesnt cave. In modern Seinfeld they would likely call this stealing the breakup (I dumped you first). 4/4"
X Link 2025-09-24T03:01Z [----] followers, [----] engagements
"If the Fed were public itd be a multi-trillion-dollar entity controlling liquidity (M2).$NVDA plays the same role in AI central bank of compute.It issues the base currency (GPUs) and does QE (indirectly buys its own stuff) so the ecosystem can expand and multiply. The same core rule also applies: dont fight the Fed"
X Link 2025-09-27T14:14Z [----] followers, 16.5K engagements
"Deepminds new paper on Chain of Frames (reasoning for video models)= dawn of the multi-modal token ramp. Veo3 is basically Gpt3 and as we move to CoF in future models compute demand will explode. This is part Jensens compounding exponential. Every extra second of video = 10s of thousands of tokens adding reasoning (multiple passes) involves another 2-4x. Think about robotics training. A 60s clip at [--] frames per second & 256k tokens / frame = 460k tokens (60x a text prompt). With [--] reasoning passes were at 920k tokens. One synthetic robot training session = hundreds of such clips i.e."
X Link 2025-09-27T20:49Z [----] followers, [---] engagements
"Imagine China buying $NVDA GPUs made by $INTC in the USA. Beijing secures $NVDA supply Trump gets a trade win and $INTC gets the demand boost it needs to scale AI chips and help de-risk US reliance on Taiwan. Everybody wins :) 🤔"
X Link 2025-09-29T18:49Z [----] followers, [---] engagements
"Ironically if the productivity boom from Ai plays out at scale we will all have much more time to create and watch AI slop. A self fulfilling loop #skynet $nvda"
X Link 2025-10-01T22:41Z [----] followers, [---] engagements
"Even if every existing GPU on earth magically became a Blackwell GB300 tomorrow wed still be 10-30x short on compute for consumer inference alone at saturation. The other TAM verticals are large (er) and compound the problem. This is why the big guys are spending. A quick overview and a few things to note: The genesis of this thought experiment is the recent compute demand surge data points / Jensens compounding exponentials from BG2. The goal is to imagine the art of the possible consumer inference TAM + the necessary compute to get a sense for what AI inning were in. 1) this is a SWAG"
X Link 2025-10-06T00:58Z [----] followers, [----] engagements
"For the wheres the ROI Dinosaurs Doomers and Top callers ZOOM OUT: Tokens are the work units the dividend is TIME. AI turns clicks actions. Assistants buy / book / fix / teach so humans dont. That creates a shadow consumer surplus (time saved money not spent) that P&Ls wont show at first (solows paradox is back). Rails get built; yield shows up later as inference-driven revenue. Monetization is things like merchant-funded conversion lift affiliate/booking fees payments/interchange action ads per-action fees.and perhaps most importantly opex deflation when agents (robots) do $50/hr work for"
X Link 2025-10-06T18:13Z [----] followers, [---] engagements
"It is highly likely Jensen blessed AND likes the OpenAI / $AMD deal. It is in $NVDAs interest for the AI flywheel to spin faster rather than have OAI slowdown to wait for Jensens limited cowos supply. Let the LO pre $AMD analyst day chase commence"
X Link 2025-10-06T20:51Z [----] followers, [---] engagements
"@FundaBottom @theinformation They dont get clicks for that. BUT they did us a favor anyone who panicked / believed it showed us they are swimming naked and dont understand a key part of the AI ecosystem at all. Keep the receipts. $nvda $amd $nbis"
X Link 2025-10-08T00:56Z [----] followers, [---] engagements
"Perhaps actually you have too little exposure as h100 pricing goes UP. 4/To all the Private Credit guys whining about having too much GPU exposure Jensen just debunked your biggest fear. Hes providing credit support by putting in $100Bn equity capital BEHIND YOU. Perhaps actually you have too little 4/To all the Private Credit guys whining about having too much GPU exposure Jensen just debunked your biggest fear. Hes providing credit support by putting in $100Bn equity capital BEHIND YOU. Perhaps actually you have too little"
X Link 2025-10-08T23:43Z [----] followers, [---] engagements
"Macro musings #1 Tough to be a dollar bear (even with cuts) when NVDA GPUs (and stock) are priced in USD (note: its a metaphor) #Computedollars $nvda $nbis $crwv $amd"
X Link 2025-10-10T00:35Z [----] followers, [---] engagements
"Musing #4: a disciplined and simple follow Jensen policy since [----] approximated the investment outcome of a multi-year AI deep dive $nvda"
X Link 2025-10-10T02:35Z [----] followers, [---] engagements
"The best hedge is often to make money somewhere else. Perhaps (not so) rare earth stocks are interesting then. If theres no art in the deal the US gov goes full throttle to build domestic capacity (operation warp speed part 2); if theres a deal Ai stocks are back on the train to #ripcity and youre happy to lose on the rocks. You also own the option rare earths work either way and you win on both 🫣 $USAR (lot of daylight between its market cap and peers 🤷🏻♂)"
X Link 2025-10-11T00:56Z [----] followers, [---] engagements
"BTW racks optics power distribution cooling all have their own cycles with multi year lead times. They have to upgrade as well to handle new hardware. So no you cant just compress the hardware cycle to any time frame you want at scale. People are still getting Blackwell and figuring out how to make it work 1yr after initial delays. Also software improvements are not discussed and they can materially change the inference efficiency for old chips over time. Better to do own work than just read seeking alpha pieces. I just brought up multiple things on the fly that are critical to the analysis"
X Link 2025-10-11T02:58Z [----] followers, [----] engagements
"@chamath Behind the meter to a DC campus near you"
X Link 2025-10-11T04:22Z [----] followers, [---] engagements
"@DanielSLoeb1 20x levered 24/7 $NVDA 🤞"
X Link 2025-10-11T15:18Z [----] followers, [---] engagements
"@firstadopter When the bubble stories stop is the tell its a bubble. $nvda $nbis"
X Link 2025-10-12T16:34Z [----] followers, [---] engagements
"@WarrenPies I heard the assets of our time GPUs & QQQ are priced in USD. Debase that $nbis $nvda"
X Link 2025-10-14T23:24Z [----] followers, [---] engagements
"Any specialist in the sector knows you cant compress the hardware cycle fast enough which the most important counter and you did not mention. So while Jensen can put out new stuff every year we cant make the rest of the infrastructure fast enough for 1-2yr full replacement cycles. Jensen will tell you this himself. @PikerCapital @OutsiderInfo76 @RihardJarc BTW racks optics power distribution cooling all have their own cycles with multi year lead times. They have to upgrade as well to handle new hardware. So no you cant just compress the hardware cycle to any time frame you want at scale."
X Link 2025-10-15T16:16Z [----] followers, [----] engagements
Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing