Nscale's S-1 Shows $103.4 Billion in Contracts – Only $2.6 Billion Are Active

Nscale's S-1 Shows $103.4 Billion in Contracts – Only $2.6 Billion Are Active

AI infrastructure company Nscale has filed its S-1 registration with the SEC ahead of a listing on the New York Stock Exchange under the ticker NSCL — and the document offers the first detailed look at the economics of a so-called "neocloud": a cloud provider building capacity for AI training and inference. The numbers in the filing are striking at both ends: a contracted contract value of $103.4 billion, against first-half 2026 revenue of $140.6 million and a net result of minus $1.02 billion. Roughly 85 percent of the contracts rest on two customers, Microsoft and Anthropic, and the largest agreement is contingent on financing and can be cancelled if the company misses milestones the S-1 itself describes as "stringent." The filing is thus, in practice, a stress test of how public markets value contracted — but not yet earned — GPU capacity.

The News Itself

Nscale, a British company, submitted its S-1 registration to the SEC on September 18, 2026, and is seeking a listing on the NYSE under the symbol NSCL, according to Crypto Briefing and CNBC. No date for the offering, price range, offering amount, or share count has been set yet — this is the window before pricing. The valuation target is inconsistently reported: Crypto Briefing cites "approximately $30 billion," its own ByteDance coverage cites "$30–35 billion," while the Financial Times via TechCrunch has mentioned $35 billion. The figure should therefore be read as a reported range, not a confirmed target. Goldman Sachs, JPMorgan, and Morgan Stanley are leading the offering according to Crypto Briefing.

The announcement lands three days after Nscale, on September 15, 2026, entered a subscription agreement worth a minimum of $3.1 billion — $2.1 billion in unsecured convertible notes plus $1 billion in convertible notes or non-voting shares to Nvidia, depending on the timing of the transaction, according to Pulse 2.0. The company announced a planned acquisition of Anyscale in July 2026.

The Numbers That Define the Company

First half of 2026 (as of June 30), according to CNBC citing the prospectus:

• Revenue: $140.6 million, up 1,252 percent from $10.4 million a year earlier.
• Net result: minus $1.02 billion, versus a loss of $368.9 million the year before.
• GPU capacity (as of August 31): 25,000 active GPUs, and 461,000 that are either active or contracted.
• Data centers: five active and 12 contracted facilities.
• Debt: over $8 billion, excluding a financing arrangement with Dell.
• Employees: over 1,000 full-time staff.

That means the loss is roughly seven times the revenue. The growth rate is real, but it is built on a capital base that today consists mainly of debt and convertible notes — and on contracts that have not yet converted into revenue.

How the $103.4 Billion Backlog Actually Works

The S-1 reports $103.4 billion in "active and contracted total contract value" (TCV) as of August 31, 2026, up from $38 billion at the end of 2025. But as Pulse 2.0 notes about the filing: the figure represents "aggregate contracted revenue across the committed terms of Nscale's signed customer agreements, rather than revenue already earned or the company's valuation." Only about $2.6 billion of the amount is classified as active TCV.

The contracts are take-or-pay agreements with an average term of around 5.7 years. In practice, that means customers commit to paying for capacity regardless of use, but the value is spread over many years ahead. The backlog is therefore a promise of future payments, not a deposit account. (A caveat is warranted here: CNBC reports, in the same coverage, a figure for "remaining performance obligations" of around $56.4 billion, and the tension with the $103.4 billion TCV figure is not resolved in available sources. It may come down to differing definitions, but we cannot verify the filing first-hand.)

The gap between $103.4 billion in contracted value and $2.6 billion in active value, and between 461,000 contracted GPUs and 25,000 active ones, is both the company's entire strategy and its entire risk: Nscale is to build itself up to the contracts, not the other way around.

Concentration and Conditionality

Roughly 85 percent of the contracts come from two names:

• Microsoft: up to $43.8 billion through 2033.
• Anthropic: up to $44.6 billion — but the agreement is contingent on Nscale securing financing, and Anthropic can withdraw or cancel if Nscale fails to meet milestones the S-1 itself categorizes as "stringent," according to TechCrunch.

The company's founder and chief executive Josh Payne (32) writes in a letter to prospective investors, quoted by CNBC from the filing: "From inception, we built Nscale with an infrastructure-first thesis…" The letter thus presents the company as built for contracted demand rather than speculation. But that presentation must be read against the debt figure of over $8 billion, and against the fact that the largest single agreement in the backlog can fall away if the financing does not come through.

The Anonymous Customer

The S-1 does not name all of its customers. One customer accounted for 73 percent of 2025 revenue (around $33 million) and 52 percent of revenue in the first half of 2026, according to Crypto Briefing. It is worth emphasizing that the identification comes from subsequent journalism, not from the S-1 itself: Crypto Briefing reports that ByteDance rents access to 2,304 Nvidia B200 GPUs at Nscale's data center in Glomfjord, Norway, a relationship that began in May 2025 and was partially financed by a $105 million loan from Macquarie. This is a serious vulnerability in the revenue base: it appears to rest heavily on one customer who is neither Microsoft nor Anthropic — and who, if that is the case, carries geopolitical and commercial uncertainty of its own.

The Capital History

Nscale was spun out of the crypto mining company Arkon Energy around 2024, according to Crypto Briefing. In March 2026, the company raised $2 billion in a Series C round that valued the company at $14.6 billion. Nvidia contributed $1 billion in convertible notes as part of the broader $3.1 billion package — the same package formalized in the subscription agreement on September 15, 2026. The company has also announced a planned acquisition of Anyscale.

The architecture is clear: an AI infrastructure provider finances its buildout with debt and convertible notes, partly from the very chipmaker whose GPUs it sells capacity on, while the contracts meant to carry it all remain largely unserved. Nvidia's position as supplier, lender, and (potentially) shareholder is a pattern that recurs across the neocloud sector, and the S-1 makes it visible on a single company's balance sheet.

What the Pricing Will Say

What makes this S-1 worth reading closely is that it forces a question the market has so far been able to bypass in the private markets: How should a share be priced when its value rests on contracted, but not yet earned, revenue, with extreme customer concentration and a major agreement that can be withdrawn?

The questions that will determine the outcome are concrete:

• Pricing and offering size: No date, price range, offering amount, or share count has been set.
• Financing: The Anthropic agreement is explicitly contingent on Nscale securing financing. The IPO is itself part of that equation.
• Delivery: Going from 25,000 to 461,000 GPUs and from five to 12 data centers requires capital, power, and time — the milestones the S-1 calls stringent.
• Customer concentration: Two names and an anonymous third party account for nearly the entire revenue base.

The numbers in the filing show a company that has actually managed to sign contracts worth over $100 billion with the industry's most sought-after customers. They also show a company that today earns $140 million in half a year, loses more than a billion in the same period, and whose largest asset is a promise from customers to pay in the future — if the financing comes through, if the milestones are met, if the capacity gets built. How Wall Street values that combination will say something about how AI infrastructure is priced at all in the years ahead.