Nscale's $103 Billion Book Is Built on a Chokepoint It Doesn't Own

Nscale is heading to a U.S. listing, reportedly as soon as this month, and the news cycle is wrapping it in the flattering frame of Nvidia's blessing. The chip giant is said to be putting roughly $2 billion into Nscale's pre-IPO financing, on top of the $300 million-plus in earlier rounds and the $900 million credit line already tied to the company. An easy read: Nvidia, the arbiter of AI's scarce goods, is betting on this startup. The more useful read starts one layer down, where the bottleneck actually lives.

Because Nscale does not make the thing that is scarce. It rents it.

The company inside the "ai compute provider" label

Nscale is a London-based "neocloud," a two-year-old company that borrows and raises money to buy NVIDIA graphics processors, build data centers around them, and lease the resulting compute to frontier AI labs on long-term contracts. Its customers, such as they are public, include Microsoft and Anthropic, and its sites span Norway, Portugal, Texas, and North Carolina.

The striking thing is the gap between the revenue it has actually earned and the revenue it says it has already sold. For all of 2025, Nscale recorded about $33 million of revenue. Its most recent quarter cleared $100 million. Yet the figure Nscale is waving at prospective IPO investors is a contracted book of roughly $103 billion — up from about $51 billion four weeks earlier, most of the jump coming from a single $45 billion deal with Anthropic signed in late August.

Those are two different things wearing the same-sounding word. The $103 billion is a forward renting commitment, a stack of signed leases for compute that will be delivered and invoiced over years if everything converts. The $33 million is cash that actually changed hands. There is nothing inherently wrong with selling capacity before you build it — CoreWeave built a public company that way and now trades at a $49 billion market cap despite still losing money. But it means the entire Nscale thesis hangs on the gap between paper commitments and realized revenue, and on how much of that spread the company gets to keep.

Follow the money to the node that matters

Now run the chain the way Nscale's own numbers imply. Anthropic wants frontier compute in huge, specific lumps; Microsoft wants the same. Nscale's job is to assemble it: secure land and power, buy GPUs, stand up the facility, and hand the keys to the customer.

At each step, ask who is hard to replace.

Power and land: scarce, but owned by grid operators and utilities, not by Nscale — and there are plenty of developers chasing the same megawatts.

The GPU: NVIDIA's Vera Rubin platform, the next generation after Blackwell. Nscale has contracted for roughly 194,000 Vera Rubin GPUs. That is the irreplaceable node. There is no fast substitute for NVIDIA's top silicon, its allocation queue is the real constraint in this whole industry, and every neocloud on earth is competing for the same allotment.

The middleman: Nscale. It owns the leases and the debt, but it owns neither the chips nor the grid. Its margin comes from the spread between what Anthropic pays for compute and what Nscale pays to assemble it — and that spread is squeezed from both sides by two much larger counterparties.

That is the uncomfortable symmetry at the center of the story. Nscale sits between a concentrated seller (NVIDIA) and a concentrated buyer (about $45 billion of its backlog, roughly 44%, sits with Anthropic alone). A supply-chain node is only as strong as its power to gouge one of those sides, and a two-year-old renter has power over neither. The scarce asset the headline implies Nvidia is blessing — well, Nvidia owns that asset, and Nvidia is its owner of record in every sense.

Why Nvidia would pay for this at all

Which returns to the odd fact worth pausing on. NVIDIA does not need Nscale's money, and it does not need to pay anyone to want its GPUs — demand has outrun supply for years. So why is the company widely reported to be adding another roughly $2 billion to a cloud firm's pre-IPO round?

Because Nvidia's interest here is not that Nscale is irreplaceable. It is that Nscale is a lever for moving Nvidia's own product. Financing a company that has committed to buy 194,000 Vera Rubin units is, in effect, Nvidia underwriting a buyer of its own bottleneck — locking downstream demand and making sure those GPUs ship through this buyer rather than to a competitor's rival chip. The $2 billion is closer to a purchase of the pipeline than a prize for excellence. It is Nvidia spending money to fortify the one chokepoint it already owns, not a signal that Nscale uniquely bosses the supply chain.

Watch what the IPO paperwork reveals about that intent. The pre-IPO raise is said to total about $3.5 billion: roughly $2 billion from Nvidia plus about $1.5 billion in convertible notes led by Third Point, with the converts carrying a double-digit discount to the IPO price that narrows as the valuation rises and caps around a $30 billion valuation. Those terms are the terms of an institution buying certainty of placement, not an arm's-length bet on a bargain. And a cap near $30 billion against the $14.6 billion the company carried in March tells you how much of the rerating the financing itself is pricing in before a single retail share exists.

What the reader actually owns in this trade

For an ordinary investor deciding whether the headline deserves a place on a watch list, the question isn't whether Nscale is important. The question is whether Nscale captures the rent, and whether the price already assumes it does.

The clean exposure to this industry's real scarcity is NVIDIA — a company trading near a $5.5 trillion market cap, near its 52-week high, that owns the chips every neocloud is fighting to buy. Even there, the forward multiple runs around 61 times earnings; nobody is getting Nvidia's structure on the cheap. Nscale is the leveraged, concentrated application of that same scarcity: a thin-middleman business whose economics are dominated by gigantic contracted revenue it has not yet earned, heavy debt it must service, and a single customer accounting for nearly half its book. Its own "illustrative" projections — about $18.1 billion of annual revenue and $13.6 billion of adjusted earnings — are framed as illustrative rather than guidance, and they only matter if the $103 billion converts.

The honest test is conversion. Watch whether Nscale turns contracted leases into invoiced revenue quarter after quarter, how much leverage it carries to do so, and what the IPO's final valuation and ownership look like. The dependency is real — the market genuinely needs these GPUs housed and powered. But the node that actually decides how fast it can ship belongs to Nvidia, and the stock that captures that decision is not the middleman's.