The Nscale IPO is shaping up as the boldest test yet of Wall Street's appetite for the artificial intelligence build-out, and the numbers behind it reward a hard look.

A $35 billion bet on contracts that don't exist yet

The pitch is audacious. A company incorporated in May 2024, with barely 55 megawatts of live computing capacity, has filed to go public on the New York Stock Exchange under the ticker NSCL and is reportedly seeking a valuation of as much as $35 billion. Eighteen months ago it did not exist as an operating business. Today it wants Wall Street to price it like an established hyperscaler.

What justifies the number is not revenue. It is a backlog. Nscale's registration statement, filed on September 18, points to roughly $103.4 billion in total contract value, a figure that would be staggering for any company and is close to fiction for one this young. The catch sits in the fine print. Only about $2.6 billion of that total is tied to capacity that actually exists. The rest is a promise to build.

Two customers carry almost all of it. Microsoft and Anthropic together account for around 85 percent of the contracted value. Microsoft's commitments run to roughly $43.8 billion and stretch out to December 2033. Anthropic signed on August 25 for up to $44.6 billion of capacity at the Monarch campus in West Virginia. Strip those two names away and the Nscale IPO looks far more like a startup and far less like a $35 billion enterprise.

The Norwegian site that caught fire

If you want to understand the gap between Nscale's promise and its delivery, look north. The company's flagship facility in Glomfjord, Norway, was supposed to anchor its reputation for execution. Instead it has become the clearest evidence of how hard the buildout really is. Of 30 planned megawatts, about 6.4 are running. The project is roughly six months behind schedule.

The delays were not the usual permitting friction. According to public records, a transformer caught fire in December. The cooling system sprang coolant leaks. A Norwegian safety agency logged ten separate safety violations before closing its inspection case in September, once Nscale completed its fixes. For a business whose entire pitch is that it can pour concrete, string power, and rack GPUs faster than anyone else, a burning transformer at the showcase site is not a good look heading into the Nscale IPO.

The consequences were financial as well as reputational. ByteDance, the TikTok owner that had contracted capacity at the site, was owed penalties running into the millions as the delays stacked up. Rather than pay them out, Nscale reportedly steered the customer into fresh agreements on more favorable terms. Microsoft was also owed penalties tied to late delivery. These are the hidden costs of growth that a headline backlog number never shows.

Nvidia on every side of the table

The most revealing detail in the Nscale story is the role played by Nvidia. The chipmaker is an investor in Nscale. It is the supplier of the GPUs Nscale buys. It is a customer for some of the capacity. And in parts of the arrangement it acts as a guarantor. One company sits on four sides of the same table.

This is the circular financing that regulators have started to name out loud, and a pattern Blumefield has tracked across the entire AI build-out. When the dominant chip vendor funds the buyers of its chips, backs their contracts, and books the resulting sales as revenue, the line between genuine demand and manufactured demand blurs. The International Monetary Fund warned this week about exactly this pattern, flagging a roughly $1.4 trillion web of vendor financing knitting hyperscalers and AI suppliers together. The Nscale IPO is that warning rendered as a prospectus.

None of this is illegal, and none of it is concealed. Nscale discloses the relationships plainly. The question investors must answer is whether a valuation built on contracts, underwritten in part by the same company selling the hardware, reflects real economic value or a hall of mirrors. The answer will not be obvious until the capacity is built and the cash actually arrives.

Burning $1 billion to chase $103 billion

The financials sharpen the stakes. In the first half of 2026 Nscale reported revenue of $140.6 million, up an eye-watering 1,252 percent year over year. In the same period it lost $1.02 billion. The growth rate is real and the loss is real, and both point to a company spending enormous sums today against income it hopes to collect years from now.

The Anthropic deal illustrates the tightrope. Nscale has acknowledged it does not yet have binding financing commitments for the capital needed to deliver that contract. Anthropic, for its part, can walk away from a tranche without liability if Nscale misses a delivery window. So the single largest piece of the backlog is a contract Nscale cannot yet fund, with a counterparty that can exit if construction slips, at a company whose flagship project just ran six months late. That is the risk the Nscale IPO asks public investors to absorb.

Scaling from 55 megawatts of live capacity to the more than 1.37 gigawatts already under contract is not an incremental step. It is a roughly twenty-five-fold expansion, executed across sites in Norway, the United Kingdom, the United States, Portugal, and Iceland, in an environment where power, transformers, and skilled labor are all scarce. Every one of those locations carries its own version of the Glomfjord risk.

Why the Nscale IPO is bigger than Nscale

It would be a mistake to read this as one company's story. The Nscale IPO is a referendum on the entire AI infrastructure trade. If public markets reward a two-year-old with a $35 billion valuation on the strength of unfunded promises, the signal to every other neocloud, every hyperscaler, and every chipmaker is that backlog is as good as cash. If they balk, the repricing could ripple far beyond one London-based startup.

The underwriters understand the symbolism. Goldman Sachs and J.P. Morgan are steering the listing, and the roster of names already attached to Nscale, from Nvidia and Dell as backers to a board that includes Sheryl Sandberg, Susan Decker, and Nick Clegg, is designed to project permanence onto a company that is still mostly a plan. The talent is real. The ambition is real. The open question is whether the market still believes that contracted demand will convert into delivered capacity and collected revenue before the capital runs dry.

For now, Nscale is asking investors to take the backlog on faith and treat the cracks as noise. Whether they do will tell us how much conviction is left in the AI buildout, and how much of it has quietly curdled into hope. As this publication argued in its coverage of the IMF's stability warning, the most dangerous number in finance is the one everyone has agreed not to question. The Nscale IPO puts that number on the ticker tape.