Is xAI a neocloud now?

xAI and Anthropic announced on Wednesday a surprising partnership where the Claude maker “buys up all the computing capacity”. [xAI’s] Colossus 1 data center”, about 300 MW, which allowed Anthropic to immediately increase its usage limits. It’s a huge deal for xAI, likely worth billions of dollars. More importantly, it delivered immediate revenue one of the company’s most impressive achievementstransforming xAI from a consumer to a computing vendor.

It’s tempting to see the arrangement as an opportunity for OpenAI amid the ongoing litigation. But Musk’s statement about X was that xAI had already done that moved training to a newer data center, Colossus 2and xAI simply didn’t need both.

In the short term, there is a clear logic at work. xAI’s existing products are mainly aimed at Grok, who has seen that declining use since the image generation debacles earlier this year. If the build-out of xAI’s data center is so much more than what Grok needs to operate, the partnership with Anthropic adds a lot of green to the balance. This is especially useful as the company, now combined with SpaceX, is heading towards an IPO. More broadly, having Anthropic as a customer makes it easier to believe that SpaceX’s orbital data center play could actually work.

But beyond the short-term benefit, the Anthropic partnership sends an unusual message about where Elon Musk’s priorities really lie. It suggests that the company’s real business may be more about building data centers than training AI models.

It’s rare for a major tech company to handle computing resources in this way, while companies like Google and Meta, which also train models, are building more data centers. It’s an easy point to miss because so many of these companies work simultaneously as enterprise AI vendors, online services, and cloud providers. But when they’re forced to make a choice between selling more available compute to customers and keeping some of it to build their own tools, they reliably choose door #2.

Just last month, Sundar Pichai admitted on a call that Google Cloud revenue was lower than it could have been because the company had “capacity constraints.” When given the choice of renting out their GPUs or using them to develop AI products, Google chose the AI ​​products.

Facebook has faced a more extreme version of the same limitation, setting up an entirely new cloud appliance to ensure they would have enough GPU power to pursue Zuckerberg’s AI ambition. As he put it when he announced Meta Compute in January: “The way we develop, invest and collaborate on this infrastructure will become a strategic advantage.”

The key word there is ‘strategic’. Both Zuckerberg and Pichai look to a future where AI powers the world’s most popular and lucrative systems. Computing power is not only a way to meet today’s inference demands, but also to build tomorrow’s products. And if you don’t have enough computing power, you miss that opportunity.

By focusing on data centers (earthbound and otherwise), xAI is positioning itself more as a neocloud company: buying GPUs from Nvidia and leasing them to model developers like Anthropic. It’s a much tougher business, under pressure from both chip suppliers and changing demand cycles. The valuations for most active neoclouds reflect that reality: xAI was valued at $230 billion in the January funding round; Coreweave, which has a similar amount of computing power, is worth a look less than a third of that.

Musk’s version of a neocloud is more ambitious, as you might expect. Some of the data centers may be in space – at least by 2035, if all goes according to plan. xAI will make its own chips at the Terafabwhich will take away some, but not all, of Nvidia’s pricing power. But nothing changes the fundamental economics of the neocloud business.

Even during the February all-hands, xAI had real ambitions in the field of software. That was the presentation that unveiled the orbital data center project, but it also teased significant ambitions in coding (since supported by the Cursor partnership) and interesting ideas such as harnessing computing in full digital twins (in the unfortunately named Macrohard project). These are the kinds of long-horizon projects that require dedicated computing resources to succeed. As long as xAI sells large amounts of computing power to its competitors, it’s hard to think such new ambitions have much of a future.