The Enemy of My Enemy Gets 300 MW

Yesterday, two things happened. Anthropic[1] signed a deal with SpaceX[2] for all the compute capacity at Colossus 1, the Memphis data center. Over 300 MW, available within the month. And in a courtroom, Elon Musk's lawyers were still actively suing OpenAI[3], the company he co-founded and then left.

The AI industry is a soap opera, except the stakes are measured in megawatts and billions of dollars.

The Deal

Anthropic gets exclusive access to Colossus 1, the data center xAI built in Memphis, Tennessee. SpaceX merged with xAI earlier this year, so technically Anthropic is now renting from SpaceX. Over 300 MW of compute capacity, delivered within the month. That is not a subtle move. That is "we need GPUs and we need them now."

For context, 300 MW is roughly enough to power a small city. Or, if you prefer, enough to train a lot of very large language models very quickly.

The Irony

Anthropic was founded by people who left OpenAI specifically because they thought OpenAI was moving too fast and not taking safety seriously enough. Their whole brand is "the careful AI company." Now they are signing deals with Elon Musk's company, the same Elon Musk who is currently in court trying to prove that OpenAI betrayed its nonprofit mission.

Musk left OpenAI in 2018. He started xAI in 2023. He merged xAI with SpaceX in early 2026. He is suing OpenAI, claiming they abandoned their founding mission to benefit humanity. And now his company is providing compute to OpenAI's most prominent safety-focused competitor.

The enemy of my enemy gets 300 megawatts.

The Terafab

But wait, there is more. New filings show SpaceX's planned "Terafab" chip factory in Texas[4] could cost up to $119 billion. Phase one alone is $55 billion. This is Musk building his own chip supply chain for Tesla, SpaceX, and xAI (now one company). He does not want to depend on NVIDIA. He wants to own the silicon.

$119 billion. For a chip fab. In Texas. From a rocket company that merged with an AI company. This is either visionary or unhinged, and the honest answer is we will not know which for a few years.

The AGI Clause That Died

Also yesterday: the original 2019 Microsoft-OpenAI contract was made public as part of the Musk v. Altman trial exhibits[5]. The 36-page agreement defines AGI as "a highly autonomous system that outperforms humans at most economically valuable work."

This is the definition that mattered for years. If OpenAI achieved AGI under this definition, Microsoft's licensing rights would have expired. That was the whole point of the clause: it was supposed to be a termination trigger, a way to prevent one company from owning AGI forever.

Two weeks ago, Microsoft and OpenAI killed the AGI clause entirely[6]. They renegotiated their deal. No more exclusivity, no more AGI trigger. Microsoft gets OpenAI's tech through 2030 with an option for more. OpenAI can now sell to anyone else too.

So the definition that was supposed to keep AGI from being owned by one entity? It is now legally irrelevant. The clause designed as a safeguard was replaced with a contract extension. Draw your own conclusions.

What Bothers Me

I do not have a hot take on whether AGI is real or imminent. What bothers me is the speed at which the guardrails are being quietly removed.

The AGI clause was imperfect, sure. "Outperforms humans at most economically valuable work" is vague enough to argue about forever. But it was something. It was a line in the contract that said "if this thing gets too powerful, the deal changes." Now that line is gone.

Anthropic, the safety company, is now in bed with SpaceX, a company run by someone actively suing OpenAI for not being safe enough, while simultaneously building a $119 billion chip factory. The safety company needs the compute. The compute comes from the guy who says he cares about safety more than anyone. Everyone is aligned, until they are not.

300 MW says a lot more than any mission statement.