[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]  Dulce [@litigious_dulce](/creator/twitter/litigious_dulce) on x 2734 followers Created: 2025-07-13 16:01:35 UTC I deliberated for several hours yesterday and today regarding the following: 1) Although $IREN ’s data centers in Canada can operate air-cooled B200s, IREN can only commission about 100K GPUs there. $CRWV will buy about 500K Blackwells this year alone. 2) In all likelihood, IREN can air-cool Hoppers but not B200s in Texas due to the high ambient temperature (rack density is irrelevant). 3) IREN probably has disclosed all sites that will be energized in 2026, meaning there isn’t some other site in a colder climate suitable for CSP expansion. Therefore, if IREN plans to scale CSP, it must do so in Texas with liquid-cooling. 4) I then wondered if IREN could build a Horizon-lite (e.g., XXX KW hybrid-cooled data center) for $3m/MW? The answer is possibly, but 5) Rack-scale systems like the GB200 are configured to be primarily liquid-cooled. For example, CRWV’s GB200 racks will be deployed with an XX% to XX% ratio of liquid-to-air cooling capabilities. So I concluded that for rack-scale systems (e.g., including the next-gen Rubin), robust liquid cooling is mandatory; anything less is insufficient. This means that Horizon is IREN’s next-gen data center, the one with the best technical specifications and capacity for upgrades. However, after talking with @TheKamaHsutra, we asked (and not for the first time): 6) Can IREN add rear door heat exchangers (RDHx) to its air-cooled data center to operate the B200s in Texas? Adding this feature is estimated to increase capex by $0.5-1m/MW, which is extremely competitive. Further, IREN can update its design so that newly built air-cooled data centers can be easily retrofitted with RDHx. This begged the question: 7) If IREN is building new air-cooled data centers in SW for bitcoin mining, will IREN expand its CSP (in response to eventual demand) by retrofitting those data centers with RDHx, facilitating the deployment of B200s? If so, then IREN’s BATNA for SW is not bitcoin mining, but CSP! This matters because 8) Although SW is one of the most valuable sites in the entire world, it is so large that it only appeals to hyperscalers that are notorious for being aggressive negotiators, They expect best-in-class performance at minimal cost and are unwilling to pay a premium unless absolutely necessary. Consequently, 9) SW is paradoxically difficult to monetize. The asset’s value is real, but its liquidity is constrained by the outsized influence of a small group of hyperscalers. Bitcoin mining is not an effective BATNA for SW because 10) Bitcoin mining does not generate enough revenue per MW. This use case is simply not accretive enough. However, CSP is the most accretive use case for IREN, even more accretive than colocation (the downside is that it is also more risky). 11) IREN exploring a CSP buildout at the SW site functions as a strategic BATNA. It signals to hyperscalers that if acceptable terms aren’t offered, IREN is willing to internalize the asset and monetize it themselves through high-margin AI compute services. Similar to bidding on your own product at an auction to set a floor price, IREN can use CSP to deter lowball offers and regain negotiating leverage. This doesn’t guarantee a better deal, but it ensures the site doesn’t clear below perceived value simply because the buyer pool is narrow and powerful. That said, I bet that 12) $ORCL, $GOOG, and $AMZN are not very happy because they all offer CSP and do not want another serious competitor. This conflict of interest is almost certainly a source of tension in ongoing negotiations. However, 13) $META doesn’t offer CSP. Further, META is willing to pay top dollar for AI/HPC assets. For the above reasons, 14) I believe META is the optimal candidate for SW. IREN can build turnkey data centers similar to Horizon for META and at the same time gradually build low-cost, scalable data centers that are perfectly optimized for current-gen GPUs. If I’m right about this, then I’m a seer, IDK what else to say. XXXXXX engagements  **Related Topics** [500k](/topic/500k) [100k](/topic/100k) [iren](/topic/iren) [canada](/topic/canada) [$iren](/topic/$iren) [$crwv](/topic/$crwv) [stocks technology](/topic/stocks-technology) [bitcoin](/topic/bitcoin) [Post Link](https://x.com/litigious_dulce/status/1944426992776155412)
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
Dulce @litigious_dulce on x 2734 followers
Created: 2025-07-13 16:01:35 UTC
I deliberated for several hours yesterday and today regarding the following:
Although $IREN ’s data centers in Canada can operate air-cooled B200s, IREN can only commission about 100K GPUs there. $CRWV will buy about 500K Blackwells this year alone.
In all likelihood, IREN can air-cool Hoppers but not B200s in Texas due to the high ambient temperature (rack density is irrelevant).
IREN probably has disclosed all sites that will be energized in 2026, meaning there isn’t some other site in a colder climate suitable for CSP expansion. Therefore, if IREN plans to scale CSP, it must do so in Texas with liquid-cooling.
I then wondered if IREN could build a Horizon-lite (e.g., XXX KW hybrid-cooled data center) for $3m/MW? The answer is possibly, but
Rack-scale systems like the GB200 are configured to be primarily liquid-cooled. For example, CRWV’s GB200 racks will be deployed with an XX% to XX% ratio of liquid-to-air cooling capabilities.
So I concluded that for rack-scale systems (e.g., including the next-gen Rubin), robust liquid cooling is mandatory; anything less is insufficient. This means that Horizon is IREN’s next-gen data center, the one with the best technical specifications and capacity for upgrades.
However, after talking with @TheKamaHsutra, we asked (and not for the first time):
Can IREN add rear door heat exchangers (RDHx) to its air-cooled data center to operate the B200s in Texas? Adding this feature is estimated to increase capex by $0.5-1m/MW, which is extremely competitive. Further, IREN can update its design so that newly built air-cooled data centers can be easily retrofitted with RDHx. This begged the question:
If IREN is building new air-cooled data centers in SW for bitcoin mining, will IREN expand its CSP (in response to eventual demand) by retrofitting those data centers with RDHx, facilitating the deployment of B200s? If so, then IREN’s BATNA for SW is not bitcoin mining, but CSP! This matters because
Although SW is one of the most valuable sites in the entire world, it is so large that it only appeals to hyperscalers that are notorious for being aggressive negotiators, They expect best-in-class performance at minimal cost and are unwilling to pay a premium unless absolutely necessary. Consequently,
SW is paradoxically difficult to monetize. The asset’s value is real, but its liquidity is constrained by the outsized influence of a small group of hyperscalers. Bitcoin mining is not an effective BATNA for SW because
Bitcoin mining does not generate enough revenue per MW. This use case is simply not accretive enough. However, CSP is the most accretive use case for IREN, even more accretive than colocation (the downside is that it is also more risky).
IREN exploring a CSP buildout at the SW site functions as a strategic BATNA. It signals to hyperscalers that if acceptable terms aren’t offered, IREN is willing to internalize the asset and monetize it themselves through high-margin AI compute services. Similar to bidding on your own product at an auction to set a floor price, IREN can use CSP to deter lowball offers and regain negotiating leverage. This doesn’t guarantee a better deal, but it ensures the site doesn’t clear below perceived value simply because the buyer pool is narrow and powerful. That said, I bet that
$ORCL, $GOOG, and $AMZN are not very happy because they all offer CSP and do not want another serious competitor. This conflict of interest is almost certainly a source of tension in ongoing negotiations. However,
$META doesn’t offer CSP. Further, META is willing to pay top dollar for AI/HPC assets. For the above reasons,
I believe META is the optimal candidate for SW. IREN can build turnkey data centers similar to Horizon for META and at the same time gradually build low-cost, scalable data centers that are perfectly optimized for current-gen GPUs.
If I’m right about this, then I’m a seer, IDK what else to say.
XXXXXX engagements
Related Topics 500k 100k iren canada $iren $crwv stocks technology bitcoin