xAI Emerges as a New Cloud Computing Contender

On Wednesday, xAI and Anthropic announced an unexpected partnership, with the creator of Claude acquiring "all of the compute capacity at [xAI's] Colossus 1 data center." This transfer of roughly 300MW of power allowed Anthropic to immediately raise its usage limits. The deal is a major win for xAI, likely valued in the billions. More significantly, it instantly monetized one of the company's most significant achievements, transforming xAI from a consumer of compute into a provider.

It's easy to view this arrangement as a strategic move against OpenAI amid their ongoing legal dispute. However, Elon Musk explained on X that xAI had already shifted its training operations to a newer facility, Colossus 2, and simply no longer required both data centers.

In the short term, the logic is clear. xAI's current products are largely centered on Grok, whose usage has declined sharply since the image generation controversies earlier this year. If xAI's data center capacity far exceeds what Grok needs, partnering with Anthropic provides a significant revenue stream. This is particularly valuable as the newly combined xAI-SpaceX entity accelerates toward an IPO. More broadly, securing Anthropic as a customer lends credibility to SpaceX's ambitious orbital data center initiative.

Beyond the immediate financial benefit, this partnership with Anthropic reveals an unusual priority within Elon Musk's strategy. It hints that the company's core business may be shifting toward constructing data center infrastructure rather than solely developing AI models.

This approach is uncommon among major tech firms. Companies like Google and Meta, which are also heavily invested in model training, are aggressively building more data centers for their own use. The distinction is subtle, as many of these companies operate simultaneously as enterprise AI vendors, online service providers, and cloud platforms. Yet, when forced to choose between selling excess compute capacity or reserving it for internal tool development, they consistently choose the latter.

Just last month, Sundar Pichai acknowledged on an earnings call that Google Cloud revenue was lower than potential because the company was "capacity constrained." Faced with the choice of leasing out their GPUs or using them to advance their own AI products, Google chose its products.

Facebook has encountered an even more pronounced version of this constraint, establishing an entirely new cloud infrastructure to ensure sufficient GPU power for Mark Zuckerberg's AI ambitions. As he stated when announcing Meta Compute in January, "How we engineer, invest, and partner to build this infrastructure will become a strategic advantage."

The crucial term is "strategic." Both Zuckerberg and Pichai are planning for a future where AI drives the world's most popular and profitable systems. Computing power isn't just about meeting today's inference demands; it's the foundation for building tomorrow's products. A shortage of compute means missing that opportunity.

By concentrating on data centers—both terrestrial and orbital—xAI is positioning itself more as a neocloud business: purchasing GPUs from Nvidia and leasing them to model developers like Anthropic. This is a far more challenging sector, pressured by chip suppliers and fluctuating demand cycles. The valuations of active neocloud companies reflect this reality: while xAI was valued at $230 billion in its January funding round, CoreWeave, which manages a comparable scale of computing power, is worth less than a third of that amount.

As one might expect, Musk's vision for a neocloud is more ambitious. Some data centers might eventually be in space—by 2035, if plans proceed. xAI will also manufacture its own chips at the Terafab, which could mitigate, though not eliminate, Nvidia's pricing influence. However, these innovations don't alter the fundamental economics of the neocloud business.

As recently as the February all-hands meeting, xAI demonstrated serious software ambitions. That presentation unveiled the orbital data center project but also hinted at major initiatives in coding (later reinforced by the Cursor partnership) and innovative concepts like creating full-scale digital twins from computer use (the unfortunately named Macrohard project). These are long-term ventures that require dedicated computing resources to succeed. As long as xAI is selling large volumes of compute to its competitors, it's difficult to see how such ambitious new projects can have a viable future.