A data-center lease, a proposed AI review regime, and a bigger argument about whether the boom is creating a monopoly or a broad economic lift.
A lease of Colossus 1 quickly turned into a much larger argument about who gets to control AI, who gets to profit from it, and whether Washington should step in before the market settles the question. The speakers framed Elon Musk’s deal with Anthropic as both a compute fix and a strategic pivot toward a new kind of hyperscaler. By the end, they were no longer just debating one contract, but the shape of AI markets, regulation, labor, and the economy around them.
Elon Musk’s lease of Colossus 1 was treated as a proof point that the AI bottleneck has shifted. The speakers argued that the fight is no longer about whether demand exists for Anthropic and OpenAI, but about who can secure enough power, land, and chips to serve it. In that reading, the deal was less a one-off transaction than a signal that the scarce input in AI is now compute, not users.
Elon just leased all of Colossus 1, his data center.
4:27
Anthropic and OpenAI's revenue performance has nothing to do with demand. Zero. It is entirely to do with the supply constraints that exist in data centers and specifically in power.
6:11
The conversation also recast the lease as a valuation story for Musk. By landing terrestrial capacity now, the speakers argued, he can weaken one of the biggest objections to the SpaceX and xAI narrative: that future space-based data centers are too far off to justify today’s ambitions. The new capacity, in their telling, gives Musk a business that can generate cash before the more speculative parts of the stack arrive.
By actually landing a bunch of terrestrial capacity, I think you start to blunt that because you can now start to say that even if the orbital data centers get delayed by a few months or a few quarters, he now has this structural core business.
7:57
He's now in the hyperscaler competing against Google Cloud, Amazon Web Services and Azure.
8:55
Elon Musk’s pitch in this stretch is that excess power and data-center capacity are no longer a sunk cost, they are the business. Brad argued that the Anthropic lease is not just a customer win for one model lab but a way to turn electrons into tokens and create a new revenue line that can cushion xAI’s own spend. The deeper claim is that Musk is building a second company inside the first, a quasi-hyperscaler that can help tell a better valuation story while the model business catches up.
There’s nobody better on planet Earth than Elon at converting electrons to tokens. It’s a critically important evolution to the story.
9:10
Brad framed the deal as a practical answer to a balance-sheet problem. xAI had been spending ahead of revenue, he said, while Anthropic got the compute it needed, and Musk got to monetize capacity that would otherwise sit idle or underused. He also cast the move as a broader shift in the company’s identity: not just a frontier-model lab, but an immediate competitor in the hyperscaler trade.
He said he was building AWS all along, or EWS all along.
9:45
The same argument ran through the rest of the discussion: if power is scarce, then whoever controls it can charge for the pipe while waiting for the models to mature. Johnson said the deal solves a major problem because xAI was carrying heavy training costs without yet participating in the revenue that is flowing into enterprise coding products. On that reading, the lease is a bridge, letting Musk lease capacity rather than fund every watt up front.
This deal fixes that problem. Elon’s now able to have a frontier model company, but he’s able to now not have these massive unpaid for capex commitments because he’s able to lease that capacity.
13:08
A harmless-sounding safety debate turned into a fight over concentration. As soon as the speakers started assigning numbers to Anthropic’s growth, the argument jumped from market size to market power, with one voice warning that the company could become a monopoly unlike anything seen before. The pushback was equally blunt: the market is still too early, too competitive, and too fragile to treat that outcome as settled.
I call it the biggest monopoly in human history.
27:17
If they just do that for 18 more months, then I think it will be in this unprecedentedly powerful position.
31:54
Sachs tried to make the claim concrete by comparing Anthropic’s trajectory to Standard Oil. In his telling, regulators would be seduced by the safety story while missing the real issue, which is control of the core technology. He suggested that a safety regime could become the modern version of a moat, protecting the winner while sounding like consumer protection.
Imagine if John D. Rockefeller was way better at public relations, and instead of calling his company Standard Oil, he called it safe oil.
28:13
People might even have called Rockefeller an effective altruist because, of course, he was so concerned about the safety of his product.
28:53
The fight over an “FDA for AI” quickly stopped being about a news report and became a proxy war over what Washington is really trying to build: a narrow cyber review process, or a standing approval regime that could shape the market before it matures. The speakers kept returning to a single fear, that safety language can be used to justify a permanent gatekeeper once the government gets comfortable with the role.
The last thing I want is DC trying to preemptively get in the game of picking winners and losers at the starting line of AI. That would be a disaster.
33:41
The approval regime, this idea that you’re going to have to share every model with an FDA in Washington and they’re going to have to pre-approve the model is a disaster.
37:35
Kevin Hassett’s public language was enough to trigger the panic. He described future AIs that could create vulnerabilities as something that should go through a process before they are “released to the wild,” and Scott Bessent cast the issue as a balance between innovation and safety. The speakers then spent the rest of the segment trying to separate a limited security review from what they said would be a disastrous pre-approval regime.
We’re studying possibly an executive order to give a clear roadmap to everybody about how this is going to go and how future AIs that also potentially create vulnerabilities should go through a process so that they’re released to the wild after they’ve been proven safe just like an FDA drug.
36:31
What we are determined to do is work with our AI companies to allow them to continue innovate. But our charge of the US government is maintaining safety.
37:04
One speaker argued that the White House reaction was less about ideology than cyber fear. In his telling, the catalyst was a model that looked especially dangerous in the wrong hands, and the sensible response was not a bureaucracy in Washington but stronger defenses, more coordination, and tighter controls on access during the preview phase. He also claimed the administration is not actually pushing a full approval system, only a way to harden infrastructure and keep agencies informed.
The argument at the end of the conversation turned on a simple test: whether AI is already showing up in profits, productivity and wages, or whether the boom is still mostly a financing story. One side kept returning to the same point, that investors have spent the money and now need proof. The other insisted the proof is already visible in cloud revenue, cost cuts and a broadening construction surge.
There is literally not a cintilla of evidence that AI has helped lift the operating margins of the S&P 500. There’s all kinds of bluster.
1:08:23
We’ve seen a massive drop. We have one that helped DoorDash with their food pictures. Now it’s all done by AI.
1:13:48
Brad countered with operating-margin data from big cloud providers and the broader market, arguing that efficiency is already showing up in the numbers. He pointed to Azure and Google Cloud growth, then said S&P 500 margins had improved even as payroll growth stayed modest. The deeper claim was that AI is already passing through the stack, from infrastructure to model revenue to application-level savings.
You can’t will profits to go up. Ultimately what happens is they have to sell more beer, they have to sell more shoes.
1:12:34
The people that are paying for all these tokens need to see an actual benefit.
1:09:15
That was the meeting point. Chamath said the market is still in a window where capital can be deployed ahead of measurable returns, but he set a deadline in the hundreds of days, not years. Brad’s answer was that enterprise spending is already monthly and recurring, which is itself a kind of proof, even if the economy at large has not yet caught up.
Why did the Elon-Anthropic deal matter so much?
It mattered because the speakers saw it as a compute solution, not just a lease. In their reading, Anthropic’s growth was being capped by power and data-center supply, and Elon turned his infrastructure into leverage.
Did they think Anthropic was becoming a monopoly?
One side openly said Anthropic could become the biggest monopoly in human history if its trajectory continued. Others pushed back, arguing the market was still early and competition from OpenAI, Google, and others remained real.
What was the dispute over an FDA for AI?
The dispute was over whether models should face a pre-release approval regime in Washington. Critics called that regulatory capture, while supporters said the government needed more cyber coordination and better visibility into model use.
What did they say AI is doing to the economy?
They argued AI is already helping cloud growth, GDP, wage growth, and operating margins. They also said the main test will come later, when companies have to prove that the tokens they buy produce durable profit gains.
Did anyone on the panel want tighter AI rules?
Yes, but mostly in the form of coordination, logging, and KYC for early model access, not a federal approval system. The stronger push was to harden systems against cyber abuse without slowing deployment.
AI-assisted summary of All-In Podcast's podcast, verified against the original transcript.