If Neocloud Demand Is 'Incredible', Why Did Microsoft Decline to Double Its Nscale Contract?

Microsoft will not exercise an option to more than double its existing agreement with Nscale to rent computing power at a Texas data centre. The Financial Times, which first reported the decision, said a potential extension would have been worth tens of billions of dollars for Nscale. The existing Texas contract is unaffected, and the choice was made several months ago, before Nscale filed for its IPO in September. This was not a sudden change of heart. It was a quiet, deliberate decision by one of the largest buyers of AI compute to stop short of expanding one of its biggest neocloud commitments.

Set that against what Jensen Huang has been saying. Speaking to press at GTC Berlin, Nvidia's chief executive described demand for neocloud capacity as 'incredible', with operators 'completely sold out'. He also said Nvidia had 'invested enough' in neocloud providers. Those are strong words. They are also the kind of words that shape how investors read a sector heading toward public markets. Nscale is one such company, UK-based and moving toward a US listing, with Josh Payne as founder and chief executive. When the largest customer declines to expand, the claim of a market that cannot keep up with demand becomes harder to take at face value.

The pattern is not confined to Texas. Microsoft withdrew from a deal to anchor Nscale's flagship Loughton, Essex data centre project after power connection delays pushed expected supply into the early to mid-2030s. That is a power problem, not a demand problem, and it is worth being precise about the difference. But it is also the second time Microsoft has stepped back from a significant Nscale commitment in a matter of weeks. Two pullbacks, one of them explicitly tied to grid delays, tell a more complicated story than 'sold out'.

Nvidia's own behaviour adds to the unease. The company has begun guaranteeing utilisation floors for neocloud partners such as Sharon AI, agreeing to rent idle capacity if customer demand falls short. That is not the action of a supplier watching capacity disappear before it can be built. It is the action of a supplier underwriting the risk that some of that capacity might sit empty. Huang's public confidence and Nvidia's private backstops can coexist, but they point in different directions.

The strongest case against this reading is straightforward, and it deserves to be stated plainly. Microsoft's decision was made months before it became public, and it reflects Microsoft's own capital discipline and power constraints rather than any market-wide shortfall in demand. A company can decline to expand one contract while demand across the sector remains genuinely strong. Power availability, not customer appetite, is the binding constraint in many markets, and Microsoft may simply have concluded that the Texas expansion did not clear its internal hurdles at the price and timeline on offer. On this view, Huang's claim and Microsoft's pullback are both true, and there is no contradiction to explain.

That argument has force, but it does not dispose of the question. If demand were as overwhelming as 'completely sold out' implies, the option to more than double capacity at an existing site, with an existing partner, would be an odd thing to let lapse. Expansion at a site already under contract is usually the cheapest and fastest way to add capacity. Declining it suggests either that the economics did not work, that power could not be secured, or that Microsoft's own forecasts did not justify the commitment. Any of those explanations is compatible with strong sector demand in the aggregate. None of them is compatible with the idea that buyers are scrambling for every available chip and every available megawatt.

The more useful question is what happens next. Nscale is heading toward a US listing with a marquee expansion lost and its flagship UK anchor tenant gone. Fidji Simo, formerly OpenAI's CEO of AGI Deployment, has joined its board as an independent director, which lends credibility but does not answer the demand question. Whether Nscale can find alternative tenants for the Texas capacity it planned to expand, and what valuation investors are willing to put on the business, will be the real test. Huang's words will not settle it. Microsoft's decision, made quietly months ago and only now surfacing, is the more informative signal.