Musk Reverses on Anthropic as SpaceX Inks $15B/Year Compute Deal

Summary

Anthropic has secured a landmark computing partnership with SpaceX, gaining access to the full 220,000-GPU capacity of SpaceX's Colossus 1 data center in Memphis for $1.25 billion per month—a total of $15 billion annually. This deal comes just three months after Elon Musk publicly disparaged the company as "evil" and predicted it would lose the AI race. In a dramatic reversal on July 9th, Musk admitted he was "clearly wrong" and declared Anthropic the current leader in AI, praising their Mythos/Fable models.

Musk's sudden shift in tone tracks closely with SpaceX's dire financial situation. The company posted a $4.9 billion net loss in 2025, with Starlink's $4.4 billion profit offset by the rocket division's $657 million loss and the xAI/X division's $6.4 billion burn. SpaceX's recent $75 billion IPO and $25 billion bond sale are fueling an estimated $40 billion capital spend in 2026, making the recurring $1.25 billion monthly revenue from Anthropic critical to the company's cash flow.

The partnership reflects broader dynamics in AI competition, particularly Musk's ongoing rivalry with Sam Altman and OpenAI. While tensions between Anthropic and OpenAI have dominated AI market discourse, Musk's financial dependence on Anthropic—combined with his documented animosity toward Altman—appears to have overcome his previous public skepticism. Industry observers note that while Musk's praise may be genuine regarding Anthropic's technical progress, the scale and timing of his reversal closely correlate with the signing and commencement of the lucrative compute deal.

• Musk's reversal appears financially motivated, coinciding with SpaceX's public listing and urgent need for revenue to fund $40 billion in 2026 capital expenditures
• The partnership underscores Musk's rivalry with OpenAI's Sam Altman and emerging structural relationships between AI compute providers and frontier model companies