Elon Musk praises Mythos/Fable, promises not to ‘cut off’ Anthropic

Elon Musk admitted he was wrong about Anthropic on Thursday. The concession came wrapped in praise for Mythos and Mythos/Fable, the lab's flagship models, and a promise that he would never "cut them off" from SpaceX's compute. The timing is not mysterious. Anthropic now pays SpaceX $1.25 billion per month for the entire 300-megawatt output of xAI's Colossus 1 data center near Memphis. The contract runs through May 2029. That is roughly $40 billion in committed revenue for a unit that did not exist as a combined entity before February, when xAI merged into SpaceX. Google pays $920 million a month for its own slice through June 2029. Anthropic is not a supplicant. It is the anchor tenant.

Musk's September 2025 post declared that "winning was never in the set of possible outcomes for Anthropic." At the time, Anthropic already held the largest enterprise AI market share. The error was not analytical. It was strategic. Musk underestimated a rival that chose to build rather than posture. Now that rival funds a meaningful fraction of SpaceX's capital-intensive compute build-out. The economics are blunt: SpaceX needs the cash flow to amortize Colossus 1 and whatever comes next. Anthropic needs guaranteed inference capacity at a scale no other provider can currently match. Each side holds a gun to the other's head. Musk calls this trust. The market calls it a contract.

His "not my style" argument leans on three exhibits. Tesla's 2014 patent pledge, since deleted from the company blog but preserved in legal archives, promised not to sue good-faith users. Tesla opened its Supercharger network and charge-port design to competitors. SpaceX launches rival satellite constellations without price discrimination. These are real actions. They also belong to a different Musk — one building market share in cars and launch, not defending an AI moat. The same man sued OpenAI, alleging betrayal of a nonprofit mission while pursuing a for-profit competitor in xAI. He has used X to throttle traffic to links he dislikes. He has threatened to ban Apple devices from his companies over an integration he mischaracterized. Style is what a man does when the incentives align. When they diverge, style bends.

Anthropic's lawyers did not rely on style. The contract almost certainly includes specific performance obligations, change-of-control provisions, and remedies for constructive termination. Shutting down 300 megawatts of dedicated capacity would trigger litigation that SpaceX cannot afford to lose. But contracts are paper. Infrastructure is physics. If Musk wanted to degrade Anthropic's service — delayed maintenance, "unplanned" outages, firmware pushes that favor xAI workloads — the contractual path to relief is slow and uncertain. The leverage runs on mutual dependence. SpaceX engineers sit beside Anthropic's workloads daily. That proximity teaches SpaceX how to build for the frontier of model inference, just as Amazon's engineers learned by hosting the early cloud winners. The knowledge transfer may be worth more than the rent.

Musk acknowledged during the OpenAI trial that AI distilling is real — competitors spinning up fake accounts to probe a rival model's behavior. He knows the value of close observation. Hosting Anthropic gives SpaceX a front-row seat to the lab that currently sets the pace. Mythos 2 is coming. SpaceX will see its compute profile before almost anyone else. That intelligence is an asset. It also means Anthropic's most sensitive workloads run on a competitor's metal, monitored by engineers who report to the competitor's CEO.

The praise for Mythos/Fable is not empty. Anthropic's models have set the enterprise standard for reasoning and code generation. Musk's admission that "no company has released a model as good" is a rare public concession of technical inferiority. xAI's Grok series has not closed the gap. The money flowing from Anthropic to SpaceX subsidizes the effort to try. That is a rational trade. It is not a friendship.

Anthropic should not trust Musk's style. It should trust the $40 billion invoice schedule, the penalty clauses, and the fact that SpaceX's balance sheet cannot absorb a default. It should also build its own capacity as fast as capital allows. Compute independence is the only guarantee that survives a change of heart. Musk's heart changes when his interests do. The current alignment is sturdy. It is not permanent. The wise operator treats the Memphis lease as a bridge, not a home.