Nscale's IPO Filing Lists $103 Billion in Contract Value. Only 2.5% of It Is Revenue.

Nscale filed to list on the New York Stock Exchange this month, and its S-1 carries the kind of figure that AI-infrastructure headlines are built around: $103.4 billion in contract value. That's roughly the market capitalisation of a mid-sized S&P 500 company, attached to a business that generated $140.6 million in revenue over the first six months of 2026. The gap between those two numbers isn't a rounding error. It's the entire story of how neocloud IPOs get priced right now, and Nscale's filing is specific enough to read closely, line by line, rather than take at the headline.

The company is targeting roughly a $35 billion valuation while raising about $3 billion in the offering, alongside a separate $3 billion arrangement with Nvidia disclosed in the same filing. That last detail fits a pattern that's become familiar this cycle: the chip supplier and the compute buyer increasingly show up on both sides of the same balance sheet, financing the demand that then gets reported back as growth.

The $103 Billion Contract Value Number That's Mostly a Promise

Total contract value is a standard disclosure for capital-intensive infrastructure companies: the sum of everything a business expects to bill across every signed customer agreement, counted over the full life of those contracts and recognised at signing. It's a real number, and regulators expect companies with long-duration agreements to disclose something like it. The trouble starts when a headline treats it as current revenue, because it isn't. It's a forecast that assumes every contract survives, in full, for years.

Nscale's own filing, as reported by TechCrunch and Quartz, breaks the $103.4 billion down further: a supply agreement with Microsoft worth $43.8 billion running through 2033, and an agreement with Anthropic worth $44.6 billion, tied to an eight-gigawatt data centre site in West Virginia. Combined, two customers back roughly 85% of the entire disclosed contract value.

Two Customers, One Unbuilt Site

The Anthropic agreement is the more conditional of the two. Its $44.6 billion figure assumes a specific chain of events: the West Virginia site gets financed, gets built to an eight-gigawatt scale, gets powered, and performs to the specification Anthropic signed up for, all before any of it converts into the kind of revenue that shows up on an income statement. As of the filing, that financing wasn't disclosed. The agreement is also gated on Nscale hitting defined build and performance milestones, which gives Anthropic grounds to walk away if Nscale falls short.

That's a meaningfully different instrument from a signed, funded, already-operating data centre lease. A total contract value figure doesn't distinguish between the two. It counts a contingent, milestone-gated agreement on unbuilt capacity the same way it counts a contract that's already running, which flatters the headline without changing what's actually enforceable today.

What gives these agreements real weight, according to reporting on the filing, is that they're structured as take-or-pay: the customer commits to pay for reserved capacity whether it ends up using it or not. That's a genuinely stronger commitment than an ordinary services contract, and it's the mechanism doing most of the work behind the $103.4 billion figure. It just doesn't override the sequencing problem. A take-or-pay obligation on capacity that hasn't been built yet is only as solid as the milestones standing between here and delivery.

The Two and a Half Percent

Of the $103.4 billion in disclosed contract value, about $2.6 billion was active as of the end of August 2026. That's roughly 2.5% of the headline figure. Over the same six months, Nscale posted a $1.02 billion net loss against $140.6 million in revenue.

None of this makes Nscale an outlier on capital intensity. Building compute infrastructure at this scale requires spending years ahead of the revenue it eventually produces. What it does mean is that the $103 billion headline and the $140.6 million income statement describe two different points on the same company's timeline: one years out and fully contracted on paper, the other today, mid-construction, and burning cash.

A Going-Concern Warning, Resolved by Spending Less

Further into the same filing, Nscale's auditors included going-concern language: the standard wording used when there's substantial doubt about a company's ability to keep operating on its current trajectory. Nscale's own management addressed it in the filing by concluding that delaying planned capital spending would free up enough cash to move past the risk.

That's a striking pairing to find in one document: a headline backlog large enough to be described in the same breath as a mid-sized S&P 500 company's market value, and a risk factor questioning whether the business can keep operating without slowing down its own build-out. Both statements can be true at once, and reading only the first one is how a $103 billion number ends up doing more persuading than it should.

This Isn't Unique to Nscale

CoreWeave is the clearest comparison. It's the highest-profile neocloud IPO of the current cycle, and it told investors at listing that it intended to diversify its customer base over time. Instead, its Microsoft revenue concentration rose, from 62% in 2024 to 67% in 2025, according to its own public filings.

The reason isn't a failure of intent. Landing one hyperscaler-sized contract is usually what gets a neocloud to IPO scale in the first place, and there are only a handful of buyers on earth who can commit tens of billions of dollars to compute at once. Diversifying away from that customer afterward means either winning an equally large deal elsewhere, which is rare precisely because the buyer pool is so small, or growing a long tail of smaller customers fast enough to dilute the ratio, which takes years. The same structural dependency shows up, to varying degrees, across the rest of the neocloud field, wherever a small number of hyperscaler and frontier-lab customers underwrite most of a company's disclosed backlog.

It also explains why Wall Street keeps pricing these listings as a test case rather than treating each one as routine. The growth story is genuinely compelling: demand for GPU capacity is real, and take-or-pay commitments from Microsoft and Anthropic are not nothing. The open question, repeated across coverage of this filing, is whether a valuation built on a decade-long backlog adequately compensates investors for how much of that backlog depends on two counterparties, a site that doesn't exist yet, and a capital-spending plan the company itself has already flagged as tight enough to warrant going-concern language.

The Comparison That Matters More Than the Headline

Set the two documents side by side and the shape becomes clearer. One page of the S-1 makes the case for a $35 billion valuation on the strength of $103.4 billion in future billings. Another page, further back in the risk factors, describes a company whose ability to keep operating depends on spending less than it had planned to. Both are accurate. Neither is the whole picture on its own, and a filing built this way rewards a reader who goes looking for the second page rather than stopping at the first.

Four Numbers Worth Pulling From Any Infrastructure Filing

The backlog headline is the least useful number in a filing like this one. These four say more about what's actually being disclosed:

What This Means Beyond the Stock Ticker

The same four questions apply outside an S-1. Any company signing a multi-year compute, cloud, or infrastructure contract with a vendor whose growth story rests on two or three mega-customers is exposed to a version of the same risk: how much of the vendor's revenue is contracted but not yet active, how concentrated is its customer base, is the capacity it's promising you actually financed and built, and under what conditions can either side walk away. Those questions matter as much to a buyer weighing a long-term vendor commitment as they do to an investor reading a prospectus.

Nscale's roadshow will spend the next few weeks selling the $103 billion number. The four smaller figures underneath it are the ones worth reading twice, by investors and by anyone signing a contract on the strength of a backlog they can't yet see running.