Two Billion From Nvidia, $103B in Contracts: Inside Nscale's $3.5B Pre-IPO Sprint

Four weeks ago, Nscale’s contracted revenue book stood at $51 billion — already an audacious number for a two-year-old company. This week, as the London-based AI cloud firm races toward a New York listing, it is telling prospective investors that the figure has roughly doubled to $103 billion, and that it wants $3.5 billion more in fresh capital before the IPO even prices, according to Reuters and Bloomberg reports published September 4.

The pre-IPO package, if completed as described, would be one of the largest private financing rounds ever assembled on the runway to a public debut — and its structure says a great deal about where AI infrastructure financing is heading in late 2026.

The Structure: Converts, a Discount, and a $30 Billion Ceiling

The raise splits into two halves. The first is up to $1.5 billion in convertible notes, with Daniel Loeb’s Third Point set to lead the group of note buyers. The second is roughly $2 billion from Nvidia, the chipmaker whose accelerators sit at the heart of everything Nscale builds. Goldman Sachs is working on the financing.

The pricing of the converts is the detail worth pausing on. They are being offered at a double-digit percentage discount to the eventual IPO price — an unusually aggressive sweetener that signals how much bridge capital Nscale needs, and how much return institutional lenders are demanding to provide it ahead of a listing that could still move. The discount isn’t fixed: it adjusts as the company’s valuation rises, and stops adjusting entirely once Nscale reaches a $30 billion valuation. In other words, the earlier and riskier the money, the bigger the payoff if the IPO lands well; beyond $30 billion, note holders simply ride the listing price.

Deliberations are continuing, and the final size and investor lineup could still change, sources cautioned.

How the Book Doubled: One $45 Billion Signature

A month is a short time for a contract book to double. It took exactly one deal. On August 26, Anthropic agreed to pay Nscale $45 billion over six years for AI computing power — roughly 460 megawatts of capacity at Nscale’s flagship Monarch campus in Mason County, West Virginia, running on Nvidia’s next-generation Vera Rubin chips, which start coming online in late 2027.

The backstory makes it more interesting. According to Semafor, that capacity was shopped elsewhere first: Microsoft walked away during a summer review of its data center portfolio, and Google passed after examining its own capital spending. Anthropic, locked in a pre-IPO compute scramble of its own, took the capacity — its fourth major compute commitment in recent months, following $50 billion with Fluidstack, $10 billion with Volta Infra Holdings, and $45 billion with SpaceX. That single signature accounts for most of the difference between the $51 billion book reported in early August and the $103 billion book being briefed to IPO investors now.

On the strength of that contracted revenue, Nscale is telling investors it could produce roughly $18.1 billion in annual revenue and $13.6 billion in adjusted earnings — figures the company explicitly labels as illustrative rather than formal guidance. For perspective on the trajectory being underwritten: Nscale booked about $33 million of revenue in all of 2025, and its most recent quarter passed $100 million for the first time.

Nvidia Financing Its Own Customer

The $2 billion Nvidia leg deserves its own scrutiny. Nscale has contracted for roughly 194,000 Vera Rubin GPUs. Nvidia’s capital would sit alongside that order book — meaning the chipmaker would be helping finance one of its largest buyers, effectively subsidizing demand for its own silicon while taking equity exposure to the buyer’s success.

This is now a familiar pattern across the AI buildout: Nvidia has deployed tens of billions into neoclouds, hyperscalers, and AI labs through equity stakes, vendor financing, and backstop commitments, recycling capital into its own revenue line. For Nscale, Nvidia’s participation is both capital and endorsement — a signal to public-market investors that the company’s chip supply is strategically secure. For Nvidia’s critics, it is another data point in the argument that the AI infrastructure boom is partly self-igniting.

A European Company Listing in New York

Nscale occupies a particular position in the European AI story: it is the continent’s largest AI infrastructure company, part-funded by a Nordic state lender and written into UK sovereign AI plans — and it intends to sell its shares in America.

The European half of the business runs on European money. The company’s Narvik project in northern Norway — which trades on cold air and Nordic hydropower, with Microsoft as its customer — drew $790 million in committed debt from ABN AMRO, DNB, Nordea, SEB, and Eksfin, Norway’s state export finance agency, which exists to support projects using Norwegian power and labor. That financing foundation stands in deliberate contrast to the New York listing, and the tension is real: Europe funds the assets, America captures the equity upside. It is a pattern European policymakers have begun to complain about aloud.

The company’s governance adds establishment credibility: its board includes Sheryl Sandberg and Nick Clegg, the former UK deputy prime minister.

The IPO That Everything Is Building Toward

Nscale has been working with Goldman Sachs and JPMorgan on a US IPO seeking up to $3 billion, with Bloomberg reporting the listing could come as soon as this month. The pre-IPO raise is best read as the last private-market brick in that wall: lock in $3.5 billion of cheap, discounted capital now, enter the IPO with a $103 billion contracted book, and let the public markets price the spread between contract and cash flow.

The bet investors are being asked to make is straightforward to state and hard to verify: that $103 billion of contracted revenue from counterparties including Anthropic and Microsoft will convert into real revenue on the schedule the contracts assume. Anthropic’s own forward compute obligations now exceed $150 billion across four counterparties, none of it contingent on revenue materializing at the assumed pace. If the AI agents economy grows into its compute budget, Nscale’s contracts are a money-printing machine and the converts are a bargain. If demand discipline arrives — as Microsoft’s summer walk-away and Google’s pass hinted it might — the backlog becomes a schedule of obligations against capacity nobody fully needs.

Either way, the next data point arrives soon: the S-1 filing, where the contracted-obligations table meets trailing revenue in public view for the first time.