Elon Musk just went from writing Anthropic’s obituary to calling it the leader in AI. According to TechCrunch AI, Musk took to X on Thursday to walk back his September 2025 claim that “winning was never in the set of possible outcomes for Anthropic,” writing instead: “I was clearly wrong about Anthropic.” The reversal matters because Anthropic now runs a huge chunk of its compute on servers Musk controls.
What set this off: users on X floated a nightmare scenario. Since Anthropic rents infrastructure from SpaceX, couldn’t Musk just wake up one morning and boot a rival off his servers? Musk’s answer was blunt. “I would never cut them off in a way that hurt them badly, even as a competitor. That’s not my style,” he wrote, per TechCrunch AI.
The deal behind the drama
This isn’t a spat between strangers. It’s a spat between a landlord and one of his biggest tenants.
Here’s the setup TechCrunch AI lays out:
- Anthropic signed a deal in May to buy 300 megawatts of compute, the full output of xAI’s Colossus 1 data center near Memphis.
- The price: about $1.25 billion per month through May 2029, roughly $40 billion in revenue for SpaceX’s xAI unit.
- xAI merged with SpaceX in February, folding Musk’s AI ambitions into his rocket company.
- Google signed a similar deal, renting SpaceX infrastructure through June 2029 at $920 million per month.
So when Musk praises Anthropic, there’s $40 billion riding on the relationship. “They are obviously currently the leader in AI,” he wrote. “No company has released a model as good as Mythos/Fable and they will undoubtedly have Mythos 2 ready soon.”
Why this matters
The interesting part isn’t the flattery. It’s the dependency. Anthropic, an AI lab that competes directly with xAI, is now paying its rival to keep the lights on. That’s an unusual amount of trust to hand a competitor, and it’s fair to ask whether “style” is enough of a guarantee.
Musk pointed to his track record as proof he plays fair. He cited Tesla’s 2014 pledge not to sue anyone using its patents in good faith, Tesla opening its Supercharger network, and SpaceX launching competing satellite systems without jacking up prices. “Even my worst enemies can attack me on this platform,” he wrote.
Still, Musk isn’t exactly known for turning the other cheek. He sued OpenAI, the company he co-founded. So the reassurance carries an asterisk.
The distillation question
Here’s what stands out to me. Anthropic doesn’t have to rely on Musk being nice. There would be contractual consequences if SpaceX pulled the plug, and SpaceX has every financial reason to keep the deal alive. Anthropic pays handsomely, and its engineers give SpaceX’s team a front-row seat to building and supporting fast-growing AI workloads, the same way Amazon’s engineers learn from hosting Anthropic.
But proximity cuts both ways. During his OpenAI trial, Musk admitted AI “distilling” is real. That’s when one model maker spins up fake accounts to prompt a competitor’s model and reverse-engineer how it works. Asked if xAI had ever distilled from OpenAI, Musk said, “Generally AI companies distill other AI companies,” TechCrunch AI reports, citing the New York Times. Anthropic itself accused three Chinese model makers of distilling Claude back in February.
Hosting a rival’s compute gives SpaceX far more visibility into how that rival operates than any outside competitor would ever get. Anthropic and Google presumably have safeguards. But the risk surface is real.
What to watch
For now, TechCrunch AI notes there’s “nothing but upside” for Musk in this arrangement. He collects the checks, his engineers level up, and he gets to look magnanimous while doing it.
The question is what happens as a three-year contract ages and the competitive stakes climb. Compute is the scarcest resource in AI right now, and control of it is leverage. Watch whether Anthropic starts diversifying its infrastructure to avoid single-vendor dependence, and whether the Mythos 2 launch changes the temperature between these two.
Today it’s praise. Contracts, though, outlast moods. You can read the full report at the original source.