Anthropic’s SpaceX Compute Deal Is the Most Contradictory Bet in AI Right Now
Anthropic built its reputation on being the safety-first alternative to OpenAI. So it raised eyebrows across Silicon Valley when the maker of Claude quietly signed a deal to pay SpaceX the company controlled by its most prominent ideological rival, Elon Musk $1.25 billion per month for compute access through May 2029.
The contradiction is hard to ignore. Every dollar Anthropic pays for Memphis compute goes to Musk’s empire, and the 90-day exit clause cuts both ways. For a company that confidentially filed for its own IPO on June 1, that kind of dependency on a rival’s infrastructure is the sort of concentration risk that tends to earn its own paragraph in a future prospectus.
A Deal Born From Desperation on Both Sides
The partnership exists because of necessity more than strategy. Grok’s monthly app downloads fell from more than 20 million in January to about 8.3 million by April a 60% drop in three months leaving xAI with surplus GPU capacity it needed to monetise. Anthropic, facing a compute shortage of its own, stepped in. The result is a partnership where Musk rents Colossus to the very rival building the model that is beating Grok.ÂÂ
That agreement, revealed in SpaceX’s IPO prospectus, covers capacity at the Colossus GPU cluster in Memphis, with either party able to walk away on 90 days’ notice after an initial period.
Meanwhile, the risks around SpaceX stock itself are also sharpening for investors eyeing the Anthropic SpaceX compute deal from the outside. Historical data covering all U.S. IPOs over the five years ending in 2025 shows only 35% of IPO stocks posted positive returns, with median underperformance versus the S&P 500 of 57.2%. SpaceX’s blockbuster debut has not erased those odds.
For venture investors watching this space, the Anthropic-SpaceX dynamic is one of the more telling case studies in what AI competition actually looks like when the compute runs out.
