Elon Musk spent Thursday publicly reversing a year of dismissiveness toward Anthropic, calling the company the current leader in AI and promising he would never use SpaceX’s servers to cripple a rival. The statement matters less as a peace offering and more as an admission of dependency: Anthropic is now one of SpaceX’s largest customers, and Musk’s business has strong reason to keep it that way.
The reversal is notable mostly because of how far it travels from where Musk started. In September 2025 he wrote that “winning was never in the set of possible outcomes for Anthropic,” a claim that was already shaky given reporting that Anthropic held the largest enterprise AI market share at the time. On Thursday he wrote, “I was clearly wrong about Anthropic,” crediting the lab’s Mythos/Fable model line as unmatched and predicting a swift follow-up release.
The compute deal explains the compliment. Anthropic agreed in May to buy the full output of xAI’s Colossus 1 data center near Memphis, Tennessee, a facility that came under SpaceX’s control after the February merger of xAI and SpaceX. The contract runs through May 2029 at $1.25 billion a month, worth roughly $40 billion in revenue to Musk’s xAI unit. Google signed a comparable arrangement for SpaceX infrastructure through June 2029 at $920 million a month. Praising a customer paying that much is a lower-risk move than picking a fight with one.
Musk backed his claim with a list of past restraint: Tesla’s 2014 pledge not to sue good-faith users of its patents, the opened Supercharger network, and SpaceX competing against rival satellite systems without raising prices on shared infrastructure. Those examples describe a pattern of unilateral, revocable choices, not contractual guarantees. Musk has also sued a rival before, taking OpenAI and Sam Altman to court over the terms of its founding mission. A promise not to retaliate carries more weight from someone who has never retaliated against a competitor; Musk does not fit that description.
Anthropic’s real protection is the contract itself, not Musk’s stated character. Cutting off a paying customer mid-term would trigger financial consequences for SpaceX and forfeit a deal that also gives its engineers hands-on exposure to how a frontier lab runs its infrastructure, the same dynamic Amazon’s chip engineers get from supporting Anthropic’s Trainium workloads.
That proximity carries its own tension. During his OpenAI trial, Musk acknowledged that AI “distilling,” the practice of querying a competitor’s model to learn how it works, is common industry behavior. Anthropic itself accused three Chinese model makers of running distillation attacks against Claude in February. Hosting Anthropic’s compute puts SpaceX in a position to see more of the company’s operations than almost any other rival could, regardless of what safeguards are contractually in place.
Read plainly, Musk’s comments are a signal about incentives, not intentions: a rival who profits from a company’s growth has every reason to say flattering things about it in public. Anthropic’s leadership should keep treating the contract’s terms, not Musk’s tweets, as the operative risk document, and any operator relying on SpaceX-hosted compute from a competitor should confirm what data-visibility protections actually exist in writing before the next contract renewal comes due.
Based on reporting by Julie Bort for TechCrunch, published July 9, 2026.
