The numbers that will greet investors in Nscale’s offering documents are hard to hold together at once: roughly $103 billion in signed customer contracts, and $33 million in revenue last year. The London-based AI cloud company is asking the public market to value it near $35 billion on the strength of the first figure, while the second shows how little of that promise has arrived.
Nscale, which counts Nvidia among its backers, has filed with the U.S. Securities and Exchange Commission to list on the New York Stock Exchange under the ticker NSCL and is seeking about $10 billion in pre-listing financing, according to people familiar with the matter. The filing caps a two-year climb from an obscure corner of the cryptocurrency industry to the center of the AI buildout.
The company was spun out of Arkon Energy, an Australian crypto miner, roughly two years ago and turned its data-center capacity toward the more lucrative business of renting out the chips that train AI models. In March it raised $2 billion in a Series C round that valued it at $14.6 billion, the largest equity raise in European history. A listing at $35 billion would more than double that figure in six months.
The backlog is the engine of the re-rating. Nscale’s signed contracts total about $103 billion, up from $38 billion at the end of last year, a figure that has tripled in a matter of months. Most of it, roughly 85 percent, comes from a single multi-year agreement to supply Microsoft with $43.8 billion of compute through 2033, with a second large supply agreement making up much of the remainder, according to the filing.
That concentration is the tension at the center of the offering. The Financial Times reported that the prospectus is sparing in what it discloses about how the orders are spread, noting that hundred-billion-dollar commitments rest with a small number of buyers. When one customer holds the key to most of the revenue, the company’s fate is, in effect, that customer’s decision.
The customers have reasons to sign. The largest technology companies are in a race to lock up computing power before the chips arrive, and they are willing to commit years ahead of delivery. For a supplier, a contract like the Microsoft deal acts as a credit line with a very high limit and a very concentrated balance.
The risk is not that the demand is invented. It is that the revenue is delayed, and that a single renegotiation can rewrite the entire story. A company that books $33 million in a year while promising $103 billion is asking investors to trust that the backlog converts on schedule and that no one customer reconsiders.
The model behind that gap is the neocloud. Nscale is part of a crop of companies that borrow heavily to buy GPU clusters, then rent the capacity out to the technology giants that need compute and do not want to wait for it. The economics work only if the customers keep renewing and the chips keep appreciating; both assumptions are being tested as the AI buildout matures.
Rivals have already tried the market’s patience with this model. CoreWeave and other GPU-rental companies have listed or filed, and the reception has been mixed, with investors questioning whether a business built on leased chips and borrowed money deserves software-style multiples. Nscale’s filing, with its thin trailing revenue, is the sharpest version of that test yet.
The pre-listing raise matters because it sets the number the public is later asked to beat. If early investors put capital in at a valuation well below the $35 billion target, the offering becomes a transfer of upside from the insiders who funded the March round to whoever buys at the top. That is a normal feature of a hot market, and a caution sign in one.
Nvidia’s stake gives the offering a tailwind, tying Nscale to the same supply chain that has carried the chipmaker’s value past $5 trillion. But the debate over AI capital spending now runs in both directions. Howard Marks of Oaktree warned this week that the roughly $5 trillion AI infrastructure buildout could keep interest rates higher for longer, and some investors have begun to ask whether the spending can ever pay for itself.
Nscale is betting it will. The offering arrives as the most valuable names in AI delay their own listings, leaving Nscale to test whether public investors still have appetite for a bet on concentrated compute. The question the listing poses is whether they will pay $35 billion for a backlog whose headline figure depends on a handful of relationships and whose revenue has yet to catch up. For now, the answer rests on how public investors weigh a promise against a projection.
