The Reversal
Elon Musk didn’t just change his tune about Anthropic—he rewrote the script entirely. In a July 2026 series of posts, he called Mythos/Fable “the best model released so far” and swore under oath (of sorts) he wouldn’t cut Anthropic off, even as a competitor. “That’s not my style,” he said flatly. What made this more than just spin was the timing: less than a year after he publicly declared Anthropic’s success “impossible,” he was now backing them with $40 billion of SpaceX compute. There’s something deeply odd—and commercially shrewd—about that pivot.
From “Impossible” to Enthusiast
Fast-forward to September 2025: Musk posted on X that Anthropic’s winning was “never in the set of possible outcomes.” At the time, he wasn’t just wrong—he wasn’t even close. Anthropic had already secured the biggest enterprise AI market share, according to contemporaneous reports. That gap between Musk’s public skepticism and Anthropic’s actual momentum would later look comical, except for the stakes involved.
The shift didn’t stop at words. By July 2026, Anthropic was one of SpaceX’s largest infrastructure customers—mainly because it signed a deal in May 2026 to rent the entire output of xAI’s Colossus 1 data center near Memphis, Tennessee: 300 megawatts at $1.25 billion per month through May 2029. That totals roughly $40 billion in revenue for SpaceX’s xAI unit, especially notable after Musk merged xAI with SpaceX in February.
For context, Google is also leasing infrastructure from SpaceX at $920 million per month through June 2029. So Anthropic’s contract isn’t just big—it’s the largest single compute agreement SpaceX has publicly disclosed.
Why Trust a Rival? The Musk Playbook
Musk wasn’t relying on vague promises. He pointed to Tesla’s 2014 patent pledge—open sourcing the electric-vehicle maker’s technologies—and the decision to open its Supercharger network to competitors. His argument boiled down to this: if he’s willing to help rivals, then Anthropic should feel safe hosting its models with him.
He added that SpaceX routinely launches competing satellite systems without raising prices or imposing unfair terms. “Even my worst enemies can attack me on this platform,” he said, referring to X. It’s a theatrical flourish—part branding, part legal reassurance.
But here’s the catch: Musk isn’t known for avoiding tough tactics against rivals. He sued OpenAI just months before these comments. And he once called an early investor “a petty man” in a public court filing. So why would Anthropic trust him? Not because he promised not to—because the contract likely includes robust remedies, and because SpaceX gains more by keeping Anthropic happy than it would by alienating it.
Those benefits go beyond the monthly checks. Engineers at SpaceX will learn how to support Anthropic’s rapidly growing AI, much as Amazon’s teams have learned from serving OpenAI and Microsoft. That knowledge transfer—intentional or not—is valuable infrastructure intel.
The Distillation Question
Musk acknowledged something unsettling during his OpenAI trial testimony: AI companies commonly “distill” rival models. He defined it as one team setting up many fake accounts to send prompts to a competitor’s API in order to understand how it works. “Generally AI companies distill other AI companies,” he said, when asked whether xAI had done this to OpenAI.
Anthropic itself accused three Chinese model developers of distillation against Claude in February 2026. That history makes the hosting arrangement more delicate: SpaceX now has direct access to Anthropic’s compute traffic, which gives it a clearer window than most competitors could ever get.
The question isn’t whether SpaceX could distill Claude—it’s whether it would. There are obvious safeguards: contractual non-compete clauses, third-party audits, maybe even air-gapped engineering teams. But if SpaceX’s engineers start tweaking GPU allocation for Anthropic, they’ll naturally learn what works and what doesn’t. That’s just how systems work.
Musk’s “not my style” line is a promise he can break without legal trouble. The real guardrail is self-interest: Anthropic’s $1.25 billion per month is too rich a prize to risk for speculative distillation.
The Upside for Everyone—Especially SpaceX
SpaceX gains three major things from hosting Anthropic:
- High-margin, long-term revenue—$45 billion over three years (before adjustments) is almost guaranteed.
- Engineering insights: Operating Anthropic’s models at scale teaches SpaceX how to support AI workloads, helping it price and package future offers.
- Strategic positioning: As a trusted cloud provider for high-profile AI labs, SpaceX becomes the fallback choice when Azure or GCP face political headwinds.
That last point matters. Governments worldwide are nervous about AI control. If Anthropic ever needs an offshore compute provider, SpaceX’s Texas-based data centers may look safer than a Big Cloud vendor’s facilities in Northern Virginia.
The contract runs until May 2029, but all three-year AI deals are renegotiated—and often rewritten—at year two. Anthropic may walk away with better terms, SpaceX may demand equity, or Musk could pivot again and cut the deal short. For now, though, everyone wins.