Analysis
Nscale Limited, a London-based builder of GPU cloud infrastructure for AI workloads, filed a Form S-1 registration statement with the SEC on September 18, seeking a New York Stock Exchange listing under the ticker "NSCL," according to the SEC filing and Dealroom's coverage of the run-up to the listing. Goldman Sachs and J.P. Morgan lead a 23-bank underwriting syndicate that also includes Morgan Stanley.
A Backlog That Doubled On One Deal
The headline number in the filing is total contract value: $103.4 billion as of August 31, up from roughly $51 billion before Nscale signed a $45 billion, five-year computing-capacity agreement with Anthropic around August 26, covering 460 megawatts at Nscale's upcoming West Virginia campus. That single deal now makes up 44% of the entire disclosed backlog. TCV is Nscale's own reported figure for contracted future revenue, not audited or recognized revenue, and its realization depends on Anthropic and Nscale's other counterparties actually drawing down capacity over the life of the contracts -- a distinction Pulse previously flagged when the $103 billion figure first leaked ahead of this filing.
“That single deal now makes up 44% of the entire disclosed backlog.”
From $33 Million To A $103 Billion Pitch
Nscale's actual revenue tells a much smaller story than the backlog headline: roughly $33 million for all of 2025, about $37 million in the first quarter of 2026, and just over $100 million in the second quarter. That's real, fast-growing revenue for an infrastructure company at this stage, but it's a rounding error against the $103.4 billion figure investors will see on the cover of the roadshow deck -- the gap between contracted future value and cash actually collected today is the central question every analyst covering this listing will have to model.
What Nscale Actually Builds
Nscale provides GPU compute, data-center capacity, networking, storage and orchestration for AI workloads, plus managed services including inference endpoints and fine-tuning. The company is scaling toward 10 gigawatts of data-center capacity and reports roughly 289,000 active and contracted GPUs, including about 194,000 Nvidia Vera Rubin chips. Nvidia itself has invested approximately $2 billion directly into Nscale, and Daniel Loeb's Third Point led up to $1.5 billion in convertible notes; Nscale separately closed a $3.5 billion pre-IPO round and, in September, issued another $3.1 billion in unsecured convertible loan notes -- a stack of debt and equity financing layered on top of each other in the run-up to a public listing.
The Neocloud Comparables
Nscale enters public markets in the same week Crusoe closed a $3.9 billion Series F at a $30.9 billion private valuation, and it will be benchmarked directly against CoreWeave, the neocloud category's first major public listing, and against Nebius. All three sell GPU access rather than owning the AI labs consuming it, and all three carry the same structural question: Nvidia investing directly in companies that resell access to its own chips is now a recurring pattern across the category, raising the recurring question of whether that's Nvidia underwriting demand for its own hardware or simply backing the fastest distributor.
The Bear Case
The gap between TCV and recognized revenue is the risk sitting under this entire filing. A $103.4 billion backlog built substantially on a single $45 billion Anthropic contract is a customer-concentration bet as much as an infrastructure bet -- if Anthropic's own compute demand slows, or if buildout delays push capacity online later than contracted, the backlog figure investors are being pitched today doesn't automatically convert to the cash flow a public market values. Nscale has not disclosed what happens to the TCV figure under a delay or renegotiation scenario, leaving that as the open diligence question once the roadshow starts.
What To Watch
The next filing to watch is the S-1/A that sets an actual price range -- until then, Nscale's targeted valuation remains investor speculation built on the same $103 billion figure the company itself is pitching. Whether public-market investors price this on the backlog or on the $100 million-a-quarter run rate will say a lot about how far AI-infrastructure valuations can still stretch ahead of matching revenue.
