Apropos of My Morning Chart of the Day: "Well, I Can Hammer...
This AM I posted: “Well, I Can Hammer the CPUs & the GPUs & the NPUs of an M4Max MacStudio After All!: Chart of the Day” <https://braddelong.substack.com/p/well-i-can-hammer-the-cpus-and-the>. Now come Derek Thompson, Claudeai, Alex Imas, Matthew Zeitlin, and Joe Weisenthal to smartly note that Elon Musk has now pulled the xAI car off to the side, gotten out of the car, is no longer pretending to be a driver, and has given the keys to its NVIDIA engine to Anthropic. I could have this wrong. But I view this as Elon Musk’s trying hard to get some positive cash flow over the period between now and the SpaceX IPO. And then, when the SpaceX IPO happens, he hopes to benefit from the most extraordinary Long Squeeze in all of financial history. Public company SpaceX. Massive market capitalization, at least initially. Small liquid float of available stock. Lots and lots of index funds that need to hold actual or synthetic SpaceX shares. Squeeze. Given mechanical index fund demand and given the lockup of the shares of insiders, is there an equilibrium market price for SpaceX a week after the IPO? Inquiring minds are very curious.
But meanwhile, the Musk empire has to make it to the IPO without running out of cash...
Now, in the center ring of the circus compute‑starved lab rents capacity from a man who hates it, to light up 220,000 idle GPUs and help Musk limp to the SpaceX IPO. The colossal Colossus deal exposes just how desperate frontier labs are for GPUs, how Grok and Optimus are not the AI-frontier, and how dependent Musk’s empire now is on narrative and other people’s belief in magic, as Anthropic buys capacity and Musk sells dark silicon to get some cash flow as he waits for the rescuing cavalry to come over the hill in the form of—dumb-β index funds.
And now into the sphere of public reason come Derek Thompson and Claudeai:
Derek Thompson: <[twitter.com/DKThomp/s...](https://twitter.com/DKThomp/status/2052066312118047201/)>: ‘Wow. The AI supply crunch is real. Frontier labs are desperate for compute. Musk has compute capacity but a meh model, and Anthropic has a fantastic model with weak capacity. Now I wonder if Elon continues to refer to his new business partner as “woke AI,” “Misanthropic,” etc.
Claudeai: ‘We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity. This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API. Effective today, we are: 1) Doubling Claude Code’s 5-hour rate limits for Pro, Max, and Team plans; 2) Removing the peak hours limit reduction on Claude Code for Pro and Max plans; and 3) Substantially raising our API rate limits for Opus models. Our agreement with @SpaceX means we will use all the compute capacity at their Colossus 1 data center. This will give us over 300 megawatts of additional capacity to deploy within the month…
With the funniest and truest comment on this being from:
Alex Imas: <[twitter.com/alexolegi...](https://twitter.com/alexolegimas/status/2052125990407577866/)>: ‘A burn so deep you can see it from space:
Matthew Zeitlin: We always here about frontier labs being compute constrained, [so] why is xai/spacex giving anthropic access to their compute?
Joe Weisenthal: ‘it’s the frontier labs that are constrained…
And the connections with my:
DeLong, J. Bradford. 2026. “Well, I Can Hammer the CPUs & the GPUs & the NPUs of an M4Max MacStudio After All!: Chart of the Day”. DeLong’s Grasping Reality. May 7. <https://braddelong.substack.com/p/well-i-can-hammer-the-cpus-and-the>.
are obvious and straightforward:
For those of you not currently under indenture to Anthropic:
Pro is the $20/month plan that gives you at least 5× the usage of the free tier plus Claude Code and Cowork.
Max5x at $100/month gives you 25× free-tier usage per 5‑hour window.
Max20x at $200/month gives you 100× free-tier usage per 5‑hour window.
Note that Anthropic, at least, claims that “all the compute capacity at their Colossus 1 data center” will now be available to Anthropic. None wil be reserved for xAI. That would appear to mean that not only is Tesla as an auto manufacturer effectively dead as a source of promises of large future cash flow to investors in Elon Musk, but so is the Robotaxi project (it needs AI), the Optimus robot (ditto), and the Grok (ditto ditto) Musk future technology visions as well. That means that all chips are riding on SpaceX—StarLink, and the transfer over of the NASA budget as largely personal profit to Elon.
Still, Musk has looked like he is about to hit the wall before—and survived to place more bets, two of which, Tesla in its glory plague-year expansion days and StarLink, had money.
I would dearly love to see something about the terms of the deal. Is Musk selling computational capacity to Anthropic at below market because he has dark silicon in large amounts because no users for Grok? Is Musk selling above—well, not above, for there is no market: people who more than half-think they are building Digital God are not in the business of selling GPU time at any price—because Anthropic right now has genuine capacity and equally genuine PR problems? It could be going either way.
One very interesting thing here is that Anthropic is compute‑constrained enough to do this. It would be operationally rational—SpaceX has dark silicon; Anthropic could light up a lot more (albeit at prices that do not cover its marginal costs, as best as I can figure). But. It is a fact that it is rarely wise to partner with someone who hates you for ideological reasons. You cannot properly incentivize them to make win-win adjustments to contracts, as they would rather hate than profit. And Musk hates Anthropic.
Anthropic must be feeling both (a) very confident in its lawyers and (b) under truly immense pressure to obtain more capacity RIGHT NOW.
A second very interesting thing is that this underscores just how much x/SpaceX/AI, capital‑rich but utilization‑poor, has already slipped from “frontier lab” into “merchant compute” territory, with the orbital‑data‑center talk best understood as fig leaf. The report <https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/> is that Grok’s TOPS-utilization rate was only ~11%, embarrassingly low. xAI leadership churn meant that without paying inference customers for Colossus I’s 220K NVIDIA GPUs burning 300MW they sat dark and idle because there was nothing for them to do. That is perhaps 2.5% of total hyperscaler computation capability (and 1% of datacenter capability, if only 0.05% of the world’s total ability to shake electrons through NAND gates).
A third interesting thing is that, as Robert Cyran <https://www.reuters.com/commentary/breakingviews/anthropic-deal-clouds-spacexs-ai-value-2026-05-07/> of Reuters notes, this puts the SpaceX IPO story at risk for a small improvement in cash flow. The vision-vibe of xAI as GPU landlord does not compute alongside “WE’RE GOING TO MARS!!!!” as the overarching narrative. The SpaceX IPO will only succeed if the narrative is prime.
Point.
But I really do not see this as having much downside from Musk’s point of view. His story is so much better than Bitcoin’s. In fact, each of Musk’s ten stories is better than Bitcoin’s. I would be surprised if Musk has to seriously worry, as long as there are people who believe in BitCoin.
More important, I believe that Musk only has to make it past the SpaceX IPO in order to become truly cash-rich. For there is in play a very powerful factor, not a fundamental factor, and definitely not a “microstructure” factor—call it a market-macrostructure factor. It is working in favor of the SpaceX IPO. Suppose the SpaceX IPOnget a small initial pop. SpaceX will then have relatively little of its tradable shares outstanding. Instantly, every single index fund not doing smart-β has to buy from that limited liquid amount.
It will be the opposite of a short squeeze: it will be a long squeeze—the biggest long squeeze in history.
It may be truly wild to watch. And, if Musk is smart, it will end with him taking down more than $100 billion in personal cash from it.