xAI Compute Economics Come Into Focus as Anthropic Reportedly Buys Colossus Capacity

Abstract illustration of AI data centers, compute racks and orbital infrastructure used for frontier model training. Abstract illustration of AI data centers, compute racks and orbital infrastructure used for frontier model training.
Abstract illustration of AI data centers, compute racks and orbital infrastructure used for frontier model training.
Original AIFeed illustration: frontier AI economics are increasingly defined by compute ownership and utilization.

Opening summary

xAI’s infrastructure strategy is becoming clearer after TechCrunch reported details from SpaceX IPO filings. One report says xAI lost $6.4 billion from operations on $3.2 billion of revenue in 2025. Another says Anthropic will pay xAI $1.25 billion per month for compute capacity through May 2029, potentially generating more than $40 billion if the contract continues. The numbers are reported from filing details and should be read as a snapshot of a fast-moving business, not a final judgment on xAI’s long-term prospects.

Key Takeaways

  • TechCrunch reports xAI’s operating losses widened sharply in 2025 while revenue grew more slowly.
  • The company is investing heavily in AI capital expenditure, including Colossus and Colossus II data-center capacity.
  • Anthropic’s reported compute contract turns xAI into both a frontier AI competitor and a supplier of infrastructure to another model company.
  • The story shows why AI economics are no longer only about model quality; utilization, power, financing and hardware control matter.

What Happened

According to TechCrunch, SpaceX filings gave one of the first public looks at xAI’s finances after Musk combined xAI, X and SpaceX into a broader public-market story. The reporting says xAI generated $3.2 billion of revenue in 2025 but lost $6.4 billion from operations. It also describes rapid capital expenditure and ambitions to scale Grok to multiple-trillion-parameter systems. In a related TechCrunch article, Anthropic is said to have secured 300 megawatts of output from xAI’s Colossus 1 data center near Memphis, with a reported monthly payment of $1.25 billion.

Why It Matters

The AI industry is entering a phase where compute strategy can be as important as model strategy. If a company overbuilds capacity, it needs customers or its capital costs can become a drag. If it underbuilds, it may be constrained by GPU availability, inference demand and power access. xAI’s reported Anthropic arrangement highlights a hybrid model: a frontier model lab can also act like a neocloud when capacity is available. That can create revenue, but it also raises strategic questions about selling compute to a rival.

Market Impact

For AI infrastructure investors, the reports validate the scale of demand for large contiguous compute clusters. For enterprise buyers, they reinforce that model prices and availability depend on physical infrastructure, not only software efficiency. For startups, the lesson is uncomfortable: competing directly in frontier models requires balance-sheet strength, power access and exceptional utilization. More narrow AI applications may be better positioned if they can avoid owning the most expensive parts of the stack.

What to Watch Next

Watch whether xAI signs more services contracts, whether Grok usage grows enough to absorb capacity internally, and whether orbital compute plans become more than a long-range narrative. Also watch Anthropic’s own infrastructure strategy: buying large external capacity may help near-term growth, but model labs will keep weighing cost, reliability, independence and speed.

FAQ

Is Anthropic becoming dependent on xAI?

The reported contract is significant, but the article says it includes termination terms. Dependence would depend on how much of Anthropic’s total compute portfolio this represents.

Why would xAI sell capacity to a competitor?

If capacity is unused or temporarily oversupplied, selling it can help monetize infrastructure while xAI continues scaling its own products.

What is the broader signal?

The broader signal is that AI companies are becoming infrastructure companies, with data centers, power, utilization and financing at the center of competitive strategy.

Sources