Is xAI Becoming a Neocloud? What the Anthropic Deal Really Reveals

xAI's Surprising Shift: From AI Developer to Compute Provider

A blockbuster announcement from xAI and Anthropic earlier this week has sparked a fundamental question about where Elon Musk's artificial intelligence company is truly headed. Under the terms of the deal, Anthropic — the company behind the Claude family of AI models — has acquired the entire compute capacity of xAI's Colossus 1 data center, an infrastructure behemoth representing roughly 300 megawatts of processing power. The agreement allowed Anthropic to immediately lift its usage caps and is expected to be worth billions of dollars to xAI.

More than the money, however, the deal signals something far more consequential: xAI appears to be transforming from a consumer of computing resources into a seller of them.

Why xAI Didn't Need Colossus 1 Anymore

Musk offered a straightforward explanation on his own platform, X, noting that xAI had already transitioned its model training operations to a newer facility called Colossus 2, leaving Colossus 1 sitting largely idle. Rather than letting that infrastructure go to waste, xAI opted to monetize it by leasing it out to a competitor — even as that competitor remains locked in legal battles with another Musk-affiliated enterprise.

Some observers have interpreted the Anthropic arrangement as a calculated jab at OpenAI, given the ongoing litigation between the two camps. But the operational rationale is difficult to dismiss. xAI's flagship consumer product, Grok, has experienced a sharp decline in user engagement following a series of image generation controversies earlier this year. If Grok's computational demands fall well short of what Colossus 1 can deliver, parking that capacity with a paying customer makes clear financial sense.

A Neocloud in the Making?

The implications of this strategy become even more striking when viewed against the backdrop of how other major technology players handle compute scarcity.

Google CEO Sundar Pichai recently acknowledged in an earnings call that Google Cloud revenues were being held back by internal capacity constraints. When faced with the choice of renting out GPU power or reserving it for in-house AI development, Google consistently chooses the latter. Meta has taken an even more aggressive stance, constructing an entirely new cloud infrastructure — branded Meta Compute — to ensure Mark Zuckerberg's AI ambitions are never bottlenecked by hardware shortages. Zuckerberg himself framed this investment as a long-term "strategic advantage," recognizing that compute access will determine who wins the AI race of tomorrow.

xAI appears to be moving in the opposite direction. By offloading large volumes of compute to external customers, the company is beginning to look less like a frontier AI lab and more like a neocloud provider — a business that procures GPU hardware from suppliers like Nvidia and leases it out to model developers and other AI companies.

The Economics Are Tougher Than They Look

The neocloud model carries its own set of challenges. These businesses operate in a compressed margin environment, squeezed from above by chip suppliers and from below by the unpredictable ebb and flow of demand cycles. A glance at market valuations tells the story: xAI was valued at approximately $230 billion during its January fundraising round. CoreWeave, a dedicated neocloud operator managing a comparable volume of computing power, is valued at less than one-third of that figure.

Musk's vision for xAI's infrastructure play is, unsurprisingly, more ambitious than a standard neocloud operation. The company has floated plans for orbital data centers — potentially operational by 2035 — and is developing its own semiconductors through a facility called Terafab, a move designed to reduce dependence on Nvidia's pricing power. These are bold initiatives, but they do not fundamentally alter the underlying economics of a business built on renting out compute.

What Happens to xAI's Software Ambitions?

Perhaps the most pressing concern raised by this strategic pivot involves xAI's own long-term product roadmap. As recently as a company-wide all-hands meeting in February, xAI was showcasing serious ambitions beyond hardware. The presentation revealed meaningful investments in AI-assisted coding — further reinforced by a subsequent partnership with Cursor — and ambitious concepts such as leveraging computer use capabilities into full-scale digital twin environments, an initiative internally dubbed the "Macrohard" project.

These are precisely the kinds of forward-looking, compute-intensive projects that demand sustained and dedicated hardware resources. If xAI continues committing large portions of its data center capacity to external clients like Anthropic, it raises legitimate questions about whether the company retains sufficient runway to develop these next-generation products.

The Bigger Picture

The xAI-Anthropic partnership may be strategically logical in the short term — particularly as xAI, now merged with SpaceX, accelerates toward a public offering. Having Anthropic as an anchor customer also lends credibility to SpaceX's orbital data center ambitions. But the long-term trajectory deserves scrutiny.

Building the infrastructure of AI is a worthy enterprise. But there is a meaningful difference between owning the infrastructure to power your own innovations and building infrastructure primarily to service your competitors. As xAI leans further into the neocloud model, the question of where Musk's AI ambitions truly lie becomes harder — not easier — to answer.