Elon Musk has a new favorite AI lab. It is not his own.

In a surprising reversal on X this week, the SpaceX and xAI owner offered glowing praise for Anthropic, the company he once dismissed as a guaranteed failure. Musk explicitly promised that he would not "cut off" the AI lab from the massive compute resources it currently rents from his infrastructure. For a CEO who has spent years locked in bitter public feuds with rivals like OpenAI, the tone shift is jarring.

It is also entirely logical. Anthropic is currently paying SpaceX $1.25 billion every month for access to the Colossus 1 data center. That is a $40 billion revenue stream. Musk isn't just being magnanimous; he is protecting a massive asset.

The Economics of a Rivalry

Musk’s change of heart follows a period of intense skepticism. In September 2025, he publicly claimed that "winning was never in the set of possible outcomes for Anthropic." Today, he sings a different tune. He now calls Anthropic the "leader in AI," specifically citing the performance of its Mythos and Fable models.

This isn't just talk. In May, Anthropic signed a deal to consume the entire output of the Colossus 1 facility near Memphis. The contract runs through May 2029. Google, another major player, is also renting SpaceX infrastructure through a separate $920 million-per-month agreement.

Musk argues that his decision to host competitors aligns with his past behavior at Tesla. He pointed to the company’s 2014 patent pledge and the opening of the Supercharger network as evidence that he prefers open ecosystems over walled gardens. "SpaceX launches competing satellite systems with no increase in price," he noted. "That’s not my style."

The Risks of Proximity

Despite the public promises, the arrangement creates a unique tension. Hosting a rival’s primary compute workload provides SpaceX with unprecedented visibility into how Anthropic’s models operate.

Musk has previously acknowledged that "AI companies distill other AI companies." Distillation involves using a competitor’s model to train your own, effectively siphoning off their intelligence. While Anthropic and Google likely have rigorous safeguards in place to prevent SpaceX from scraping their proprietary data, the physical proximity of the hardware remains a factor.

SpaceX engineers are now tasked with supporting Anthropic’s scaling needs. This creates a feedback loop. By helping Anthropic build, SpaceX engineers gain deep insights into the architecture of the world’s most advanced models. It is a massive learning opportunity for Musk’s team.

What This Means for Users

For the end user, this partnership is a stabilizing force. It ensures that the compute-heavy training required for the upcoming Mythos 2 model has a home. Without this capacity, the pace of AI development would likely stutter.

Key Takeaways

  • The Financial Reality: Anthropic is paying SpaceX $1.25 billion monthly, making it one of Musk's most important enterprise customers.
  • The Strategic Pivot: Musk has publicly abandoned his previous stance that Anthropic was destined to fail, now calling them the industry leader.
  • The Hidden Benefit: By hosting Anthropic, SpaceX gains deep operational knowledge and engineering experience that could accelerate its own AI efforts.

Musk’s promise to keep the servers running is backed by more than just his word. The contract is ironclad. There would be severe legal and financial consequences if he attempted to throttle Anthropic’s access.

For now, the rivalry is on ice. The money is too good. The real test will come in 2029, when the current contracts expire and the competitive landscape looks entirely different.