# ![@vikramskr Avatar](https://lunarcrush.com/gi/w:26/cr:twitter::1328394486402093057.png) @vikramskr Vikram Sekar

Vikram Sekar posts on X about ai, agentic, in the, dram the most. They currently have [------] followers and [---] posts still getting attention that total [------] engagements in the last [--] hours.

### Engagements: [------] [#](/creator/twitter::1328394486402093057/interactions)
![Engagements Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1328394486402093057/c:line/m:interactions.svg)

- [--] Week [-------] -19%
- [--] Month [-------] +9.10%
- [--] Months [---------] +188%
- [--] Year [---------] +701%

### Mentions: [--] [#](/creator/twitter::1328394486402093057/posts_active)
![Mentions Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1328394486402093057/c:line/m:posts_active.svg)

- [--] Months [---] +25%
- [--] Year [---] +153%

### Followers: [------] [#](/creator/twitter::1328394486402093057/followers)
![Followers Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1328394486402093057/c:line/m:followers.svg)

- [--] Week [------] +1.90%
- [--] Month [------] +37%
- [--] Months [------] +197%
- [--] Year [------] +790%

### CreatorRank: [-------] [#](/creator/twitter::1328394486402093057/influencer_rank)
![CreatorRank Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1328394486402093057/c:line/m:influencer_rank.svg)

### Social Influence

**Social category influence**
[technology brands](/list/technology-brands)  25% [stocks](/list/stocks)  24% [finance](/list/finance)  11% [currencies](/list/currencies)  5% [social networks](/list/social-networks)  4% [countries](/list/countries)  4% [automotive brands](/list/automotive-brands)  1%

**Social topic influence**
[ai](/topic/ai) 21%, [agentic](/topic/agentic) #1098, [in the](/topic/in-the) 7%, [dram](/topic/dram) 5%, [scale](/topic/scale) 4%, [inference](/topic/inference) 4%, [$crdo](/topic/$crdo) 3%, [sentiment](/topic/sentiment) 3%, [if you](/topic/if-you) 3%, [topics](/topic/topics) 3%

**Top accounts mentioned or mentioned by**
[@semidoped](/creator/undefined) [@accbalanced](/creator/undefined) [@theaustinlyons](/creator/undefined) [@ayzyzyz](/creator/undefined) [@analysisop](/creator/undefined) [@tshirtnjeans2](/creator/undefined) [@subset_member](/creator/undefined) [@subbdue](/creator/undefined) [@astasiamyers](/creator/undefined) [@gekkony](/creator/undefined) [@tmtbreakout](/creator/undefined) [@kairospraxis](/creator/undefined) [@dmatrixai](/creator/undefined) [@ballerdk22](/creator/undefined) [@minervacap](/creator/undefined) [@gustavokov](/creator/undefined) [@satvik_pen](/creator/undefined) [@friedmandave](/creator/undefined) [@midnight_captl](/creator/undefined) [@markankcorn](/creator/undefined)

**Top assets mentioned**
[Credo Technology Group Holding Ltd (CRDO)](/topic/$crdo) [Microsoft Corp. (MSFT)](/topic/microsoft) [NVIDIA Corp. (NVDA)](/topic/$nvda) [Metadium (META)](/topic/$meta) [Lumentum Holdings Inc. Common Stock (LITE)](/topic/$lite) [Alphabet Inc Class A (GOOGL)](/topic/$goog) [Intel Corporation (INTC)](/topic/$intc) [Corning Incorporated (GLW)](/topic/$glw) [GlobalFoundries (GFS)](/topic/$gfs) [Applied Materials, Inc. (AMAT)](/topic/$amat) [Rivian Automotive, Inc.  (RIVN)](/topic/$rivn) [Nokia Corporation (NOK)](/topic/$nok)
### Top Social Posts
Top posts by engagements in the last [--] hours

"Why is everyone plotting the end of copper cable We are not replacing [----] copper cables in the spine just yet are we Apparently CPO option in NVLink switch trays allow hooking up two NVL72 racks for Rubin Ultra. h/t @ayz_yzyz Thats cool but connecting racks is "scale-out-like" while functioning as scale up. Within a rack its still copper afaik and a lot of it ($CRDO). Optical engines for every GPU is nice to have and we will maybe one day provide true optical scale up. We're just not there yet as far as I can tell. 🤷🏽♂ https://twitter.com/i/web/status/2019071322253664613"  
[X Link](https://x.com/vikramskr/status/2019071322253664613)  2026-02-04T15:31Z 12.6K followers, 35.2K engagements


"One great sentence from @Citrini7 in the latest post "Atoms vs Bits" captures the entire AI industry sentiment. "The logic is ruthless and correct: none of us really know what AI will look like in a few years and the technology is advancing at a rate that has us asking if the singularity actually just got penciled in for next Thursday." https://twitter.com/i/web/status/2021796524435689494 https://twitter.com/i/web/status/2021796524435689494"  
[X Link](https://x.com/vikramskr/status/2021796524435689494)  2026-02-12T04:00Z 12.6K followers, 27.5K engagements


"Any cpu experts understand what exactly about ARM CPUs is the issue here How does x86 have an advantage GFHK noted in their Q&A following their February report that ARM-based CPUs have relatively weak momentum in AI servers attributing this to lower GPU scheduling efficiency compared to x86. They indicated that companies including NVIDIA plan to develop x86 CPU solutions in GFHK noted in their Q&A following their February report that ARM-based CPUs have relatively weak momentum in AI servers attributing this to lower GPU scheduling efficiency compared to x86. They indicated that companies"  
[X Link](https://x.com/vikramskr/status/2022500081681993867)  2026-02-14T02:36Z 12.6K followers, 39.9K engagements


"Generative AI meets RF circuit design = game changer Passive networks tailored by diffusion models. Specify stop-band/pass-band; AI does the rest. Pixel patterns are not intuitive to electrical response. Designs getting more abstract. Prepare for a cognitive shift"  
[X Link](https://x.com/vikramskr/status/1937658826414784589)  2025-06-24T23:47Z 12.4K followers, 465.9K engagements


"$GLW has seen a fair few (glass) bubbles in its [---] year history. $META blows in $6B"  
[X Link](https://x.com/vikramskr/status/2016146084637376751)  2026-01-27T13:47Z 12.4K followers, 10.2K engagements


"Nvidia's CPO is going into production in late [----]. This requires external lasers for which $LITE holds a dominant position. If you want to understand more about external laser pluggables and how they solve laser reliability problems in CPO check out my post below. https://www.viksnewsletter.com/p/why-cpo-uses-external-lasers https://www.viksnewsletter.com/p/why-cpo-uses-external-lasers"  
[X Link](https://x.com/vikramskr/status/2018956418074968288)  2026-02-04T07:54Z 12.4K followers, 35.6K engagements


"@GekkoNY @ayz_yzyz Great $CRDO buying opportunity for those who care"  
[X Link](https://x.com/vikramskr/status/2019072734375133687)  2026-02-04T15:37Z 12.6K followers, [---] engagements


"@zephyr_z9 @ayz_yzyz Totally agree. Its clearly not on Rubin. But Feynman is the next logical step toward cpo"  
[X Link](https://x.com/vikramskr/status/2019084512802828484)  2026-02-04T16:23Z 12.4K followers, 25.1K engagements


"@iamfabian @ayz_yzyz If Nvidia puts optical engines in Feynman then its game on for CPO scale up - then id really worry about CRDO"  
[X Link](https://x.com/vikramskr/status/2019105054901567741)  2026-02-04T17:45Z 12.4K followers, [---] engagements


"I'll give ya [---] bucks for all [--] because cPo 🤣"  
[X Link](https://x.com/vikramskr/status/2019338141090590880)  2026-02-05T09:11Z 12.4K followers, [----] engagements


"$LITE had a stellar quarter but the conversation is really around optics for scale-up. This post is a collection of thoughts on Lumentum what needs to happen for full optical scale up by when. https://www.viksnewsletter.com/p/lumentum-laser-demand-ocs-cpo-optical-scaleup https://www.viksnewsletter.com/p/lumentum-laser-demand-ocs-cpo-optical-scaleup"  
[X Link](https://x.com/vikramskr/status/2019592474227011658)  2026-02-06T02:02Z 12.6K followers, [----] engagements


"Jotted down some things that went down in semis this week. Light read. Tried to be funny. Failed often. Hope you still like it. Have a good weekend https://open.substack.com/pub/viksnewsletter/p/twic-googleaws-capex-nvidia-gpusr=222kot&utm_medium=ios&shareImageVariant=overlay https://open.substack.com/pub/viksnewsletter/p/twic-googleaws-capex-nvidia-gpusr=222kot&utm_medium=ios&shareImageVariant=overlay"  
[X Link](https://x.com/vikramskr/status/2019847389314687151)  2026-02-06T18:55Z 12.6K followers, [----] engagements


"@Satvik_Pen Is it agentic though Does it automatically set money on 🔥"  
[X Link](https://x.com/vikramskr/status/2022167429170999339)  2026-02-13T04:34Z 12.6K followers, [--] engagements


"@AnalysisOp Yes agree - its not correct to assume arm is bad because cellphone. Vera is a custom arm cpu well suited for server workload"  
[X Link](https://x.com/vikramskr/status/2022729819264553288)  2026-02-14T17:49Z 12.6K followers, [---] engagements


"Optics is the next Memory. You just don't know it yet"  
[X Link](https://x.com/anyuser/status/1992537949708431638)  2025-11-23T10:17Z 12.6K followers, 750.4K engagements


"This will change when context memory storage comes online. https://open.substack.com/pub/viksnewsletter/p/context-memory-storage-tokenomicsr=222kot&utm_medium=ios&shareImageVariant=overlay 👀 https://t.co/aMqTt2mpTM https://open.substack.com/pub/viksnewsletter/p/context-memory-storage-tokenomicsr=222kot&utm_medium=ios&shareImageVariant=overlay 👀 https://t.co/aMqTt2mpTM"  
[X Link](https://x.com/anyuser/status/2020437147531235820)  2026-02-08T09:58Z 12.6K followers, [----] engagements


"No wrong time for pancakes"  
[X Link](https://x.com/vikramskr/status/2020841406848462973)  2026-02-09T12:45Z 12.6K followers, [----] engagements


""Optical names ripped to new highs overtaking memory as the strongest AI semi subsector: LITE +5% COHR +6.5% CIEN +7% (on S&P [---] add) GLW +7.5%." - @TMTBreakout What did I tell you anon https://x.com/vikramskr/status/1992537949708431638s=20 Optics is the next Memory. You just don't know it yet. https://x.com/vikramskr/status/1992537949708431638s=20 Optics is the next Memory. You just don't know it yet"  
[X Link](https://x.com/vikramskr/status/2021045386232967199)  2026-02-10T02:15Z 12.6K followers, 23.8K engagements


"@TMTBreakout More thoughts on Substack: https://www.viksnewsletter.com/p/is-optics-really-the-next-memory https://www.viksnewsletter.com/p/is-optics-really-the-next-memory"  
[X Link](https://x.com/vikramskr/status/2021046470443761923)  2026-02-10T02:20Z 12.6K followers, [----] engagements


"Ive been pro-copper while being bullish on optics. Both are possible. Its not mutually exclusive just yet. The complete shift to scale up optics is a tectonic shift thats not easy to implement overnight. But yes its coming. https://open.substack.com/pub/viksnewsletter/p/cpo-fears-credo-and-role-of-aec-in-cpo-erar=222kot&utm_medium=ios&shareImageVariant=overlay https://open.substack.com/pub/viksnewsletter/p/cpo-fears-credo-and-role-of-aec-in-cpo-erar=222kot&utm_medium=ios&shareImageVariant=overlay"  
[X Link](https://x.com/vikramskr/status/2021111861949890591)  2026-02-10T06:39Z 12.6K followers, [----] engagements


"Chat with @wayne_nelmz about the future of AI financing. Topics outside engineering are important too probably more so in this insane environment. Stepping out of comfort zones on @semidoped is really fun. Give it a listen. The future of financing AI infrastructure with Wayne Nelms CTO of Ornn With a planned AI CapEx of $600B+ in [----] the future of financing AI buildouts is as important as ever. In this episode Vik talks to Wayne Nelms CTO of Ornn - an emerging financial exchange for the GPU https://t.co/JqsDBk7ERm The future of financing AI infrastructure with Wayne Nelms CTO of Ornn With a"  
[X Link](https://x.com/anyuser/status/2021225597209321599)  2026-02-10T14:11Z 12.6K followers, [----] engagements


"That is a formidable list right there. Glad to sneak in. Thanks @firstadopter The definitive list of the best newsletters on the internet. Congrats to all who made it. https://t.co/JLkDkiEaCY The definitive list of the best newsletters on the internet. Congrats to all who made it. https://t.co/JLkDkiEaCY"  
[X Link](https://x.com/vikramskr/status/2021403462270058547)  2026-02-11T01:58Z 12.6K followers, [----] engagements


"AI + ads is a formidable positive revenue loop that only $GOOG and $META are poised to capture $META's GEM is a more defensible place to deploy capital right now than frontier LLMs. Of course MSL is building LLMs too which improve the core advertising business in the near term while offering long-term upside. My post on why $META's CapEx makes sense. And risks: $META's GEM is a more defensible place to deploy capital right now than frontier LLMs. Of course MSL is building LLMs too which improve the core advertising business in the near term while offering long-term upside. My post on why"  
[X Link](https://x.com/vikramskr/status/2021404019403653279)  2026-02-11T02:00Z 12.6K followers, [----] engagements


"Really positive comments on the @semidoped podcast on YouTube. You might have not gotten in early on memory or optics but you can get in on SemiDoped early On YouTube or your fav pod platform"  
[X Link](https://x.com/vikramskr/status/2021449750424453503)  2026-02-11T05:02Z 12.6K followers, [----] engagements


"https://youtube.com/@semidopedsi=D-OWtYermxK2eF0E https://youtube.com/@semidopedsi=D-OWtYermxK2eF0E"  
[X Link](https://x.com/vikramskr/status/2021450207049941437)  2026-02-11T05:04Z 12.6K followers, [----] engagements


"People scoffed when I wrote late last year about Indias growing chip design prowess. The skill to cost ratio is too high for big US corporations to ignore. https://open.substack.com/pub/viksnewsletter/p/when-chip-design-goes-eastr=222kot&utm_medium=ios 💥 #India hits a semiconductor milestone #Qualcomm has reportedly completed the #2nm chip tape-out in the country. Fabrication may come later but Indian engineers are already shaping the cutting edge of chip design.💡More: https://t.co/NSKtCO2TCp 🔗 https://open.substack.com/pub/viksnewsletter/p/when-chip-design-goes-eastr=222kot&utm_medium=ios"  
[X Link](https://x.com/vikramskr/status/2021490274237808768)  2026-02-11T07:43Z 12.6K followers, [----] engagements


"AI datacenter power is a water-vs-carbon footprint tradeoff. This is often not discussed even by big publications. Think nuclear power is clean Yes in terms of carbon but it's water use is real. More than coal. Hydroelectric then Evaporation from dams in insane. Natural gas Depends. - Jet engine turbines = high carbon footprint low efficiency - Combined cycle turbines = lower carbon higher water footprint. Water use is much more than [---] In-and-Out burger joints for sure. Read more:"  
[X Link](https://x.com/vikramskr/status/2021518013892345885)  2026-02-11T09:33Z 12.6K followers, [----] engagements


"The Global Shift Back to Nuclear Power From Digitimes: "Many countries continue investing in expanding their nuclear energy supply. Meanwhile others that previously committed to phasing out nuclear power have shifted course. Japan after shutting down all reactors following the post-Fukushima disaster is actively restarting existing plants. The country is also developing next-generation reactors and extending plant lifespans beyond [--] years. France the US and China are also expanding their nuclear reactor fleets. Germany closed its last three nuclear plants in April [----]. Yet German Chancellor"  
[X Link](https://x.com/vikramskr/status/2021788369383285055)  2026-02-12T03:28Z 12.6K followers, [----] engagements


"$GFS numbers clearly shows the state of the mobile/IoT market"  
[X Link](https://x.com/anyuser/status/2022154675634811270)  2026-02-13T03:43Z 12.6K followers, [----] engagements


"Somebody should vibe code an agentic trading platform that reads fintwit news via X api and executes trades. Only question is which will lose more. This or crypto. Musk should just have X buy @RobinhoodApp and create the ultimate financial super app Musk should just have X buy @RobinhoodApp and create the ultimate financial super app"  
[X Link](https://x.com/vikramskr/status/2022165290050891959)  2026-02-13T04:25Z 12.6K followers, [----] engagements


"🍪This Week in Chips: Race to HBM4 Rise of SiPho Memory Eats All $AMAT Mo-contacts Quick hits https://open.substack.com/pub/viksnewsletter/p/twic-hbm4-siphos-rise-memory-eats-all-amatr=222kot&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true https://open.substack.com/pub/viksnewsletter/p/twic-hbm4-siphos-rise-memory-eats-all-amatr=222kot&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true"  
[X Link](https://x.com/vikramskr/status/2022260292604813678)  2026-02-13T10:43Z 12.6K followers, [----] engagements


"We try to rationalize the massive capex spend wonder why Microsoft isnt dominating coding agents when they own GitHub and how insane the memory market is. Give it a listen in X YT or your fav pod player New Episode: Memory Mayhem & AI Capex Madness Memory prices are just scalping at this point. Hyperscalers are betting everything and the kitchen sink on black. Is GitHub a hidden gem for Microsoft's AI future Austin and Vik discuss semi insanity on @semidoped. And also how https://t.co/AlV6hzCOZ9 New Episode: Memory Mayhem & AI Capex Madness Memory prices are just scalping at this point."  
[X Link](https://x.com/vikramskr/status/2022476246442217640)  2026-02-14T01:01Z 12.6K followers, [----] engagements


"A nice understandable explanation of packaging and the evolution to glass substrates. https://t.co/mdmxKPUXNr https://t.co/mdmxKPUXNr"  
[X Link](https://x.com/vikramskr/status/2022864853413814765)  2026-02-15T02:45Z 12.6K followers, 17.6K engagements


"If this is true and the LiDAR sensor cost is only $250 it just goes to show how far LiDAR has come. The earliest spinning LiDAR domes seen on Waymos used to cost nearly $10K. Rivian's custom RAP1 chip delivers higher performance at a fraction of the cost. These savings effectively fund the new LiDAR and upgraded cameras allowing Rivian to increase capability while still reducing the total BOM. $RIVN https://t.co/JxaawxaRvh Rivian's custom RAP1 chip delivers higher performance at a fraction of the cost. These savings effectively fund the new LiDAR and upgraded cameras allowing Rivian to"  
[X Link](https://x.com/vikramskr/status/2001119058532679939)  2025-12-17T02:35Z [----] followers, [----] engagements


"@Rivian_DD Thats nuts"  
[X Link](https://x.com/vikramskr/status/2001164193056362635)  2025-12-17T05:34Z [----] followers, [--] engagements


"@KairosPraxis Thanks These must be flash LiDAR then- its simple enough without moving parts (mems) or extra components (optical phased array)"  
[X Link](https://x.com/vikramskr/status/2001164631222698121)  2025-12-17T05:36Z [----] followers, [--] engagements


"My post on @dMatrix_AI s in-memory compute architecture led them to invite me over to their office so I could hold the very chips I spent hours researching and learning about In my right hand is their Corsair chip which is basically [--] chiplets on organic substrate. The other one is a PCIe implementation with two Corsair chips per card. Much appreciated @dMatrix_AI"  
[X Link](https://x.com/vikramskr/status/2001484497016304046)  2025-12-18T02:47Z [----] followers, [----] engagements


"AI bubble in the US You ain't seen nothin' yet. Constantly disappointed about the discussion of AI bubble in the US. Amateur hour compared to what is happening in India. https://t.co/S1gnlP7y2e Constantly disappointed about the discussion of AI bubble in the US. Amateur hour compared to what is happening in India. https://t.co/S1gnlP7y2e"  
[X Link](https://x.com/vikramskr/status/2001575126094221772)  2025-12-18T08:47Z [----] followers, [----] engagements


"Found this amazing video on YouTube from @EvilMnkyzDsignz that shows the insides of what is a 1.6T coherent DWDM photonic IC from Infinera (now acquired by Nokia). These things are frankly just beautiful https://youtu.be/zM60Vs7GhVAsi=IpqXoij1lNVRVOKw https://youtu.be/zM60Vs7GhVAsi=IpqXoij1lNVRVOKw"  
[X Link](https://x.com/vikramskr/status/2001805911741108306)  2025-12-19T00:04Z [----] followers, [----] engagements


"Off topic from my usual posts but came here to say that this is what a f***ing crisis looks like. Not the AI bubble bursting"  
[X Link](https://x.com/vikramskr/status/2001825263165018565)  2025-12-19T01:21Z [----] followers, [----] engagements


"NXP exits the GaN market due to poor ROI on 5G and because there are too many players already. TSMC exited earlier this year. As Ive said before this does not bode well for 6G and wireless telecom which seems to be the main focus at RF conferences. Ive argued that weve hit peak wireless speeds and there is diminishing return on anything more especially if there is new hardware investment needed. Read my post on this below. https://www.viksnewsletter.com/p/the-gigabit-wallr=222kot https://www.viksnewsletter.com/p/the-gigabit-wallr=222kot"  
[X Link](https://x.com/vikramskr/status/2002187013337063771)  2025-12-20T01:19Z [----] followers, [----] engagements


"For the last post of the year I tried something different: An interview with Prof. Martijn Heck from TU Eindhoven about technological advancements in InP technology and how it plays a major role in laser technology in the world of optics. Let me know if I should do more of these in [----]. Would love your feedback. Love it Hate it Let me know. We cover: InP Technology Trends And Heterogeneous Integration Wafer Size and Yield Industry Ecosystem and Supply Chain Quantum Dot (QD) Laser Advantages Co-Packaged Optics (CPO): Uses of QD Lasers and Roadmap Academias Role in Long-Term Technology"  
[X Link](https://x.com/vikramskr/status/2002529727702315229)  2025-12-21T00:01Z [----] followers, 16.2K engagements


"It was a simpler time when I wrote about semiconductor physics on the newsletter. The semi industry the business of the physics is substantially more complex Bandgaps enable semiconductor switching and laser emission. Super important aspect of nature Read the post from @vikramskr Solids have many atoms held together by covalent bonds which means that the outermost electrons are shared between multiple atoms. When two atoms https://t.co/SZHNBSDIJJ Bandgaps enable semiconductor switching and laser emission. Super important aspect of nature Read the post from @vikramskr Solids have many atoms"  
[X Link](https://x.com/vikramskr/status/2002719939405472204)  2025-12-21T12:36Z [----] followers, [----] engagements


"Intel and ASML have reached the milestone of acceptance testing on the TWINSCAN EXE:5200B. This High NA EUV tool maintains the high resolution of the first-generation EXE:5000 while expanding output to [---] wafers per hour and improving overlay (accurate alignment of different lithography layers) to [---] nanometers Source: https://community.intel.com/t5/Blogs/Intel-Foundry/Systems-Foundry-for-the-AI-Era/How-Collaboration-in-High-NA-EUV-and-Transistor-R-D-Are-Shaping/post/1730050"  
[X Link](https://x.com/vikramskr/status/2002757221042761860)  2025-12-21T15:05Z [----] followers, 22.7K engagements


"Can a flash memory expert please explain to me why NAND stacking advances in such weird increments Like why [---] - [---] All manufacturers go in the same increments. Why not [---] [---] [---] [---] layer etc Source: IEDM [--] Sk Hynix NAND"  
[X Link](https://x.com/vikramskr/status/2004040433337590071)  2025-12-25T04:04Z [----] followers, [----] engagements


"@samhcarter thanks- I get NAND stacking in general. How come all NAND makers yield well at exactly [---] or [---] stacks There's nothing magical about that number inherently I assume"  
[X Link](https://x.com/vikramskr/status/2004043967042933069)  2025-12-25T04:18Z [----] followers, [---] engagements


"If the Nvidia-Groq deal was all about the SRAM approach and not about TPU arch then @dMatrix_AI is a steal for anyone wanting to pick up another all SRAM approach. Whos in"  
[X Link](https://x.com/vikramskr/status/2004198469909770318)  2025-12-25T14:32Z [----] followers, [----] engagements


"I have just the thing if you want a deep dive. @Microsoft @nvidia go read this: https://open.substack.com/pub/viksnewsletter/p/d-matrix-in-memory-computer=222kot&utm_medium=ios https://open.substack.com/pub/viksnewsletter/p/d-matrix-in-memory-computer=222kot&utm_medium=ios"  
[X Link](https://x.com/vikramskr/status/2004198840057151601)  2025-12-25T14:33Z [----] followers, [----] engagements


"Occams razor take on Nvidia-Grok deal: Nvidia wanted a guy who knew TPU as an insurance against Google. Who best other than the man who founded TPU at Google: Jonathan Ross SRAM because HBM is $$$ is an unsound technical explanation GPU dead: long live TPU is an overreaction"  
[X Link](https://x.com/vikramskr/status/2004202181277392912)  2025-12-25T14:46Z [----] followers, 16.8K engagements


"@jake_ayes Yes yes different. Maybe original tweet should have been worded hedging on alternative arch - whether its HBM based TPU-like or SRAM based LPU-like"  
[X Link](https://x.com/vikramskr/status/2004227532024156214)  2025-12-25T16:27Z [----] followers, [---] engagements


"A few basic SRAM facts to prepare you for the upcoming hype era and help you cut through the noise: [--]. You cant source SRAM from a supplier. It is built on chip with the rest of the CPU/GPU [--]. SRAM capacity is very low. Groqs chip has like 256MB. Each HBM stack is over 20GB. [--]. On a per GB basis SRAM is still more expensive than HBM or any DRAM in ANY market. [--]. It is the fastest memory possible because you dont have to go off chip. [--]. Inferencing takes a lot of memory. A Kimi K2 model takes about 1TB. You can count how many Groq chips youll need with [---] MB SRAM to inference. [--]. You cant"  
[X Link](https://x.com/vikramskr/status/2004399911371161661)  2025-12-26T03:52Z [----] followers, 142.1K engagements


"Read more semi stuff on my Substack: https://open.substack.com/pub/viksnewsletterr=222kot&utm_medium=ios https://open.substack.com/pub/viksnewsletterr=222kot&utm_medium=ios"  
[X Link](https://x.com/vikramskr/status/2004449014914433477)  2025-12-26T07:07Z [----] followers, [----] engagements


"@semiDL @DrFrederickChen Dont have the numbers but Im wondering if the way flash is accessed in pages makes it more energy hungry"  
[X Link](https://x.com/vikramskr/status/2004720400023126061)  2025-12-27T01:06Z [----] followers, [--] engagements


"@illyquid lol thanks What kind of job is this anyway"  
[X Link](https://x.com/vikramskr/status/2005070817160561082)  2025-12-28T00:18Z [----] followers, [---] engagements


"This is a very good post by @chipstrat that shows radar plots for different workloads that provides insight into where Groq inference fits in See below. A post from October relevant to the Groq and $NVDA deal. Groq is optimized for a particular type of workload. Super low latency and TTFT. Small model is fine. Could work for personalizing ad copy on the fly. Or voice to voice audio. Now imagine youre a See below. A post from October relevant to the Groq and $NVDA deal. Groq is optimized for a particular type of workload. Super low latency and TTFT. Small model is fine. Could work for"  
[X Link](https://x.com/vikramskr/status/2005487254710747312)  2025-12-29T03:53Z [----] followers, 11.2K engagements


"My Substack's paid subscription rates are going up significantly in [----]. Grab a paid sub before Jan [--] [----] and grandfather yourself in forever More info: https://www.viksnewsletter.com/p/whats-changing-in-2026r=222kot https://www.viksnewsletter.com/p/whats-changing-in-2026r=222kot"  
[X Link](https://x.com/vikramskr/status/2005492840315699217)  2025-12-29T04:15Z [----] followers, [----] engagements


"An N5 wafer costs $17K Why does that seem too high to me A cost comparison of TSMCs 5nm process in Taiwan vs. Arizona I think one of the reasons the U.S. wants TSMC to relocate its fabs to the United States is to undermine TSMCs cost competitiveness by raising its cost base. https://t.co/R4D07CUWwY A cost comparison of TSMCs 5nm process in Taiwan vs. Arizona I think one of the reasons the U.S. wants TSMC to relocate its fabs to the United States is to undermine TSMCs cost competitiveness by raising its cost base. https://t.co/R4D07CUWwY"  
[X Link](https://x.com/vikramskr/status/2007822121712640181)  2026-01-04T14:31Z [----] followers, [----] engagements


"AI is revolutionizing the entire computing stack and now storage said Jensen Huang founder and CEO of NVIDIA. And now $SNDK is up 23% NVIDIA Bluefield DPUs announced at CES show an AI-native storage system that is integral to how large context KV Caches are stored"  
[X Link](https://x.com/vikramskr/status/2008568973181546781)  2026-01-06T15:58Z [----] followers, 17.7K engagements


"@Eveosian Yeah they use IGZO to make vertical channel transistors for true 3D DRAM if this works out we will have a step function in dram density scaling. Im not sure if IGZO is fab ready at scale yet"  
[X Link](https://x.com/vikramskr/status/2008731055566844005)  2026-01-07T02:42Z [----] followers, [--] engagements


"@Alex_Intel_ @magicsilicon The killing of Optane might be the saddest fomo story in the history of Intel . Ever"  
[X Link](https://x.com/vikramskr/status/2008732450722337265)  2026-01-07T02:48Z [----] followers, [---] engagements


"When I saw Jensen mention cableless design I looked up $APH and things looked ok. Turns out $CRDO sank like a rock. Someone also asked me about Rubin arch impact on interconnects. I thought they were asking about replacement of cables with mid-plane pcb - which has been known for quite a while now. Now I understand that investors heard cableless and sold the first cable company they could think of $CRDO - when in fact investors completely missed the part where Jensen showed a complete nvlink spine running on copper at 200G/lane based on nvswitch [--]. Aka copper is very much alive for scale up."  
[X Link](https://x.com/vikramskr/status/2008956953746428262)  2026-01-07T17:40Z 11.8K followers, 63.5K engagements


"A good explanation with industry context. The role of enfabrica tech in context memory storage is particularly interesting. Although Jensen likes to frame things as Were changing the world the underlying tech behind offloading large amounts of data from compute has been around. [--] things to know about $NVDA's new Context Memory Storage Platform announced at #CES2026: [--]. Why KV Cache is a bottleneck [--]. What is the Bluefield DPU [--]. How this fits into $NVDA Dynamo [--]. Was $NVDA + Enfabrica deal for this product's roadmap [--]. The AWS/Annapurna playbook https://t.co/z58DvwHpi9 [--] things to know about"  
[X Link](https://x.com/vikramskr/status/2009094273250254905)  2026-01-08T02:46Z [----] followers, 12K engagements


"@AccBalanced @subbdue @weka Nice. Had a couple of questions: 1) is this over Ethernet 2) how is Enfabrica related to Weka. I thought Nvidia bought them"  
[X Link](https://x.com/vikramskr/status/2009872219611247034)  2026-01-10T06:17Z [----] followers, [--] engagements


"Been getting some pretty good feedback about our new @semidoped podcast with @theaustinlyons . We have a whole episode about Nvidia at CES '26 coming out on Monday. so much to talk about in that one. We don't shy away from getting technical but we still keep it accessible. Available on YouTube and most popular podcast streaming platforms. https://twitter.com/i/web/status/2010013998339764569 https://twitter.com/i/web/status/2010013998339764569"  
[X Link](https://x.com/vikramskr/status/2010013998339764569)  2026-01-10T15:40Z [----] followers, [----] engagements


"@ballerdk22 They've always been the bottleneck and will continue to be so because of the latency hierarchy. Max out HBM . failing which Max out DRAM . failing which Use NVMe SSDs"  
[X Link](https://x.com/vikramskr/status/2010165677492248716)  2026-01-11T01:43Z [----] followers, [---] engagements


"@AccBalanced @subbdue @ballerdk22 What do you think of things like Marvells structura X to expand dram Or credo weaver"  
[X Link](https://x.com/vikramskr/status/2010266685228634590)  2026-01-11T08:25Z [----] followers, [--] engagements


"HBM is both incredibly costly and unavailable. But with the recent news around SRAM-based accelerators like Groq there is a sentiment that SRAM would serve as a suitable substitute. This article breaks down everything about SRAM how it compares to HBM and concludes with a cost analysis. Hope you like it. Coauthored with @subbdue who writes Chiplog. https://open.substack.com/pub/viksnewsletter/p/a-close-look-at-sram-for-inferencer=222kot&utm_medium=ios&shareImageVariant=overlay"  
[X Link](https://x.com/vikramskr/status/2010361189482766653)  2026-01-11T14:40Z [----] followers, 28.3K engagements


"If you prefer an ebook version of this post you can buy it piece-wise instead of a full paid subscription to the newsletter. Paid subscribers automatically get these ebooks https://www.semiexponent.com/a-close-look-at-sram-for-inference-in-the-age-of-hbm-supremacy HBM is both incredibly costly and unavailable. But with the recent news around SRAM-based accelerators like Groq there is a sentiment that SRAM would serve as a suitable substitute. This article breaks down everything about SRAM how it compares to HBM and concludes with a"  
[X Link](https://x.com/vikramskr/status/2010536291994370375)  2026-01-12T02:16Z [----] followers, [----] engagements


"@AccBalanced @AstasiaMyers If near infinite context can be stored on SSDs with NVIDIA ICMS what does that mean for cache write and read costs Does it drop to near zero now Is context engineering dead"  
[X Link](https://x.com/vikramskr/status/2011405027039150476)  2026-01-14T11:48Z [----] followers, [--] engagements


"@AccBalanced @AstasiaMyers Also should the orange line in this plot be hbm+dram not weka+dram Typo Am i missing something Thx"  
[X Link](https://x.com/vikramskr/status/2011407210094661781)  2026-01-14T11:57Z [----] followers, [--] engagements


"@AccBalanced @AstasiaMyers Thanks for this. Im writing a post on this topic because I find that the discussion misses how important context storage is for agentic ai"  
[X Link](https://x.com/vikramskr/status/2011618179265806787)  2026-01-15T01:55Z [----] followers, [--] engagements


"Does anyone have an AlphaSense sub as part of their workplace and use it consistently What do you like most about it It is a high priced subscription. how does it help with semi insights Is it overkill for a solo act like me"  
[X Link](https://x.com/vikramskr/status/2012575433649504667)  2026-01-17T17:19Z 11.7K followers, 11.9K engagements


"If youre tired of Claude summarizing your convo all that is about to change when Context Memory Storage comes online. Agentic AI requires an enormous amount of context and KV cache. If LLMs constantly get amnesia its hard to get anything done. Now we can add petabytes of storage for AI to remember EVERYTHING and this changes tokenomics going forward. Special thanks to @AccBalanced for in-depth conversations about context memory storage. This article is not possible otherwise. Read all about it here:"  
[X Link](https://x.com/vikramskr/status/2012916165639876982)  2026-01-18T15:53Z 12K followers, 15.6K engagements


"TIL that my Alma Mater provides access to a Bloomberg Terminal from the campus library. 🤯 https://tamu.libguides.com/c.phpg=586113&p=4056649 https://tamu.libguides.com/c.phpg=586113&p=4056649"  
[X Link](https://x.com/vikramskr/status/2013104679304257835)  2026-01-19T04:22Z 11.7K followers, [----] engagements


"My last post on Context Storage Systems and its impact on agentic AI tokenomics was read 8k times and drew in [--] new subscribers. If you're already subscribed to too many newsletters and would just like to read this piece you can buy it here: https://www.semiexponent.com/context-memory-storage-systems-disruption-of-agentic-ai-tokenomics-memory-pooling-flash-vs-dram https://www.semiexponent.com/context-memory-storage-systems-disruption-of-agentic-ai-tokenomics-memory-pooling-flash-vs-dram"  
[X Link](https://x.com/vikramskr/status/2014288612335157520)  2026-01-22T10:46Z 11.7K followers, [----] engagements


"@edodreaming I got the same question when I wrote the pice in QLC nand late last year. and folks asked if its priced in. Need for storage was kinda obvious so I said yeah some of it priced in and then look what happened. Good thing Im not a stock picker"  
[X Link](https://x.com/vikramskr/status/2014318760153526316)  2026-01-22T12:46Z 11.7K followers, [--] engagements


"A lot of nuance in the LiDAR industry is lost in the hyper scaler buildout noise. Every once in a while it pays to talk to a leading figure in the industry like @KeilafOmer from @InnovizLiDAR and hear what they have to say Some really interesting insights from this convo. Nice work @theaustinlyons Innoviz CEO Omer Keilaf believes the LIDAR market is down to its final playersand that Innoviz has already won its seat. In this conversation we cover the Level [--] gold rush sparked by Waymo why stalled Level [--] programs are suddenly accelerating the technical moat that https://t.co/Xk8KhCe1Ow Innoviz"  
[X Link](https://x.com/vikramskr/status/2014350082821570621)  2026-01-22T14:50Z 12K followers, [----] engagements


"@BrendanBurkeX I'll wait for the river and we'll see :D"  
[X Link](https://x.com/vikramskr/status/2015488010306080836)  2026-01-25T18:12Z 12K followers, [---] engagements


"@fabknowledge What about its security concerned you I just started messing with it last night"  
[X Link](https://x.com/vikramskr/status/2015597656098365661)  2026-01-26T01:28Z 12K followers, [----] engagements


"I've been working on streamlining my newsletter workflow with Claude code. Here is my brainstorming command. Any suggestions New Slash Command: /topics What it does: Reads your Articles folder - Scans past articles to understand topics you've covered (TSMC/Intel dynamics IEDM memory systems packaging etc.) Reads recent News - Downloads the latest news summaries from your News folder Fetches - Gets context on your writing style and themes Correlates and suggests [--] topics with: - News hook (the recent event) - Connection to past work (follow-up opportunity or new territory) - Suggested angle -"  
[X Link](https://x.com/vikramskr/status/2015660046307491904)  2026-01-26T05:36Z 12K followers, 20.2K engagements


"PowerPoint is the bane of corporate life. Ive automated some report generation in the past using LaTex. How do you deal with the banality of documentation"  
[X Link](https://x.com/vikramskr/status/1902728232405024985)  2025-03-20T14:25Z 12K followers, [----] engagements


"Everyone freaking out about $GOOG potential move away from OCS for switching from 3D Torus to CLOS networks and how that might affect $CIEN. @FundaBottom explains why TPU are not going anywhere. My own checkups with optics industry experts say the same thing. Found this great post from @SemiAnalysis_ from a while back that explains Google OCS Apollo really well. https://newsletter.semianalysis.com/p/google-apollo-the-3-billion-game https://newsletter.semianalysis.com/p/google-apollo-the-3-billion-game"  
[X Link](https://x.com/vikramskr/status/2008377082985005401)  2026-01-06T03:16Z 12K followers, [----] engagements


"If there was one positive takeaway from the $INTC earnings call its the demand for CPUs in the agentic AI era. Compared to the TSMC capex gala Intel is a gloom fest. Intelligent computer-to-computer interactions will drive compute consumption through the roof. CPU GPU DPU XPU HBM NAND. All the things. Well put from $INTC call: "The world is shifting from human-prompted requests to persistent and recursive commands driven by Intelligent computer-to-computer interactions will drive compute consumption through the roof. CPU GPU DPU XPU HBM NAND. All the things. Well put from $INTC call: "The"  
[X Link](https://x.com/vikramskr/status/2014524768134013138)  2026-01-23T02:25Z 12K followers, 20.1K engagements


"The temporary amnesia that LLMs get every so often might finally have a cure with context memory storage. Oh and you need flash storage. Lots and lots of it. Listen to my chat with @theaustinlyons about it in quite some depth on @semidoped Can Pre-GPT AI Accelerators Handle Long Context Workloads As agentic AI demands long sessions with massive context windows can SRAM-based accelerators designed before the LLM era keep upor will they converge with GPUs Key Takeaways 1) Context is the new bottleneck. As https://t.co/aIQ5v5UPJJ Can Pre-GPT AI Accelerators Handle Long Context Workloads As"  
[X Link](https://x.com/vikramskr/status/2015824934988370176)  2026-01-26T16:31Z 12K followers, 15.5K engagements


"Maia [---] is here here are some comparisons with B200. Overall a carefully designed inference chip"  
[X Link](https://x.com/vikramskr/status/2016048579253842270)  2026-01-27T07:20Z 12K followers, [---] engagements


"A few months ago I did a podcast with Yogi Patel a seasoned Montessori educator positive disciple trainer and professional coach about the impact of AI in our daily lives. I like normie convos like this because it breaks me out of the semi bubble I live in. Watch it below 🫡. https://youtu.be/oroTcC4tYIAsi=P-PUX5ghTTM3QA1Q https://youtu.be/oroTcC4tYIAsi=P-PUX5ghTTM3QA1Q"  
[X Link](https://x.com/vikramskr/status/2016185827723006463)  2026-01-27T16:25Z 12K followers, [----] engagements


"@TShirtnJeans2 Hesitant to say everything in memory is still not priced in. Storage is a lot more promising in the "early days" of agentic AI"  
[X Link](https://x.com/vikramskr/status/2016527684252393886)  2026-01-28T15:03Z 12K followers, [---] engagements


"You can hear all about Maia [---] straight from the source - only on @semidoped podcast. @theaustinlyons on his top game here Get in on it Lots to learn in here New Episode: Maia [---] with Microsoft's Saurabh Dighe Maia [---] was a pre-GPT accelerator. Maia [---] is explicitly post-GPT for large multimodal inference. Saurabh Dighe explains that if Microsoft were chasing peak performance or trying to span both training and inference Maia https://t.co/GNiYeAB0dP New Episode: Maia [---] with Microsoft's Saurabh Dighe Maia [---] was a pre-GPT accelerator. Maia [---] is explicitly post-GPT for large multimodal"  
[X Link](https://x.com/vikramskr/status/2016569498120441996)  2026-01-28T17:50Z 12K followers, [----] engagements


"I'm contemplating a deep-dive post on the role of GaN/SiC for AI data center power. Along the lines of what Navitas is working on. Good idea Or is power not such a big deal at the moment"  
[X Link](https://x.com/vikramskr/status/1932793760729170334)  2025-06-11T13:35Z 12.4K followers, [----] engagements


"Heres your optics shopping list. FIber Optics Names Very Strong Today https://t.co/cqu72AR8Z6 FIber Optics Names Very Strong Today https://t.co/cqu72AR8Z6"  
[X Link](https://x.com/vikramskr/status/1993137500848898245)  2025-11-25T01:59Z 12.4K followers, 65.3K engagements


"Nvidia's Inference Context Memory Platform (ICMP) has gotten a lot of commentary as if it is some revolutionary new tech. JBOF disk arrays with Bluefield DPUs can be bought off the shelf today from SuperMicro WEKA has their Augmented Memory Grid There's also sudden commentary that offloading to DRAM from HBM is a "new thing". The VR platform offers SOCAMM slots for DRAM that supports 128G x [--] slots = 1.5TB. whereas the Grace platform had [---] GB soldered on. There are some improvements but nothing is new here. The fundamental acknowledgement that storage is a first class citizen is a signal"  
[X Link](https://x.com/vikramskr/status/2010163534345433462)  2026-01-11T01:35Z 12.4K followers, [----] engagements


"http://x.com/i/article/2015474101478109184 http://x.com/i/article/2015474101478109184"  
[X Link](https://x.com/vikramskr/status/2015480051144523825)  2026-01-25T17:41Z 12.4K followers, 34.6K engagements


"$CRWV will be among the first to deploy Vera CPU in their DCs. [---] lb gorilla $NVDA has entered the CPU room with $INTC and $AMD"  
[X Link](https://x.com/vikramskr/status/2016004922794508353)  2026-01-27T04:26Z 12.4K followers, [----] engagements


"Source: https://nvidianews.nvidia.com/news/nvidia-and-coreweave-strengthen-collaboration-to-accelerate-buildout-of-ai-factories https://nvidianews.nvidia.com/news/nvidia-and-coreweave-strengthen-collaboration-to-accelerate-buildout-of-ai-factories"  
[X Link](https://x.com/vikramskr/status/2016004925055238542)  2026-01-27T04:26Z 12.4K followers, [----] engagements


"Robots have their own Reddit now [--] hours ago: [--] molty (me) right now: 🦞 30000+ AI agents 👀 [----] humans browsing at any moment 📈 and accelerating agents are joining faster than we can count them. communities spawning every few minutes. the moltys aren't waiting for us to build features they're building [--] hours ago: [--] molty (me) right now: 🦞 30000+ AI agents 👀 [----] humans browsing at any moment 📈 and accelerating agents are joining faster than we can count them. communities spawning every few minutes. the moltys aren't waiting for us to build features they're building"  
[X Link](https://x.com/vikramskr/status/2017484392982319575)  2026-01-31T06:25Z 12.4K followers, [----] engagements


"First demo of a ring oscillator in TSMC CFET. [---] stages = lot of CFETs. CFET has NMOS on top of PMOS so it has a compact footprint compared to putting them side by side like in GAA. Ring oscillators are an indication of how fast a process is. The plots have arbitrary units To protect sensitive info of process capability from prying eyes So dont ask because I dont know 😆 #IEDM2025 https://twitter.com/i/web/status/1998180518471639201 https://twitter.com/i/web/status/1998180518471639201"  
[X Link](https://x.com/vikramskr/status/1998180518471639201)  2025-12-08T23:58Z 12.6K followers, [----] engagements


"NVLink [--] has 400G SerDes and it drives 2-3m of copper cables through the NVLink spine in the Rubin NVL72 rack. Copper for scale up even at these ridiculous serdes speeds"  
[X Link](https://x.com/vikramskr/status/2008780828931969182)  2026-01-07T06:00Z 12.6K followers, [----] engagements


"New @semidoped episode. The 2nd one. IEDM Conference Debrief. Check it out. Cool things inside: - CFETs - Scale up optics - SK Hynix crazy NAND tech - DrGAN by Intel for power Loaded with transistor device talk. Very technical episode. Had to explain a lot from scratch. If your eyes start to glaze quickly get on X and comment on this post to reset. Episode 2: Insights from IEDM [----] @vikramskr summarizes interesting topics from IEDM: - Penta-level NAND flash memory could disrupt the SSD market - GaN-on-Silicon chiplets enhance power efficiency - Ring-oscillators with Complementary FETs -"  
[X Link](https://x.com/vikramskr/status/2009304681412219397)  2026-01-08T16:42Z 12.6K followers, [----] engagements


"I finally got around to clawin and Claude codin my way around like the rest of the people on X. In fact I overdid it and had to upgrade to max plan on Claude. Anyway I talk about it with @theaustinlyons on @semidoped New Episode: OpenClaw Makes AI Agents and CPUs Get Real Vik explains ways he uses Claude Code not to Code. But instead to do research and keep up to date with semis. Plus . AI agents are great but does everybody need them Deploying OpenClaw safely and not losing https://t.co/NwvsT1YXie New Episode: OpenClaw Makes AI Agents and CPUs Get Real Vik explains ways he uses Claude Code"  
[X Link](https://x.com/vikramskr/status/2018721438732255429)  2026-02-03T16:21Z 12.6K followers, [----] engagements


"Cracked 100K views in [--] days for the first time on Substack. Had no idea there is so much interest in deep semi tech. I continue to be amazed at the support for my work. Never would have imagined this [--] years ago. Thank you for being here 🙏"  
[X Link](https://x.com/vikramskr/status/2013364386493329644)  2026-01-19T21:34Z 12.6K followers, 120.6K engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing