Nscale, the London-based AI cloud company racing to turn contracted GPU demand into live megawatts, said it raised $3.36 billion in pre-IPO convertible loan notes — a financing sized for data-center steel and chips more than for a quiet private mark-up. The company announced the package on September 25, 2026, one week after filing to list on the New York Stock Exchange under the ticker NSCL, according to its press release and contemporaneous coverage of the S-1.
Third Point led the convertible round. Nscale said new and existing backers also included Nvidia, funds managed by Apollo, Citadel, Hudson Bay Capital, the Abu Dhabi Investment Council, and 8090 Industries, with further participation from Davidson Kempner, Qube Research & Technologies, Wellington Management, and others named in the release. Goldman Sachs acted as placement agent.
The headline figure is split. Nscale said $2.36 billion funded at closing, with an additional $1 billion Nvidia commitment expected to fund in mid-November 2026 — making the chipmaker the largest single commitment in the package. The notes convert automatically into ordinary shares when the IPO completes; Nvidia’s notes convert into non-voting shares, the company said. Nscale did not disclose a coupon, maturity, conversion discount, or valuation cap in the announcement. Bloomberg, citing people familiar with the matter, later reported the notes were expected to convert at a double-digit percentage discount to the IPO price, with that discount adjusting up to a $30 billion valuation — terms that remain unverified by the company.
What the capital is meant to buy is capacity, not optionality. In the same release, Nscale cited more than $103 billion in total contracted value and said proceeds would accelerate a vertically integrated stack — behind-the-meter power, liquid-cooled AI data centers, and large-scale GPU clusters — for hyperscalers, frontier labs, and enterprises. Founder and CEO Josh Payne called the raise a milestone for scaling “full-stack AI infrastructure” and accelerating global data-center buildouts.
The S-1 numbers underneath that TCV claim show how unfinished the factory still is. CNBC, summarizing the prospectus, reported about 25,000 active GPUs as of August 31, 2026, against roughly 461,000 active and contracted GPUs; five data-center sites active and 12 contracted; and line of sight to about 10 gigawatts of potential computing power. Cap Table and other S-1 reads put active-plus-contracted power near 1.37 GW and say only about $2.6 billion of the $103.4 billion TCV was “active,” with the rest dependent on sites being financed, built, powered, and accepted. First-half 2026 revenue was $140.6 million — up more than twelvefold from a year earlier — against a net loss of about $1.02 billion, CNBC reported. Microsoft and Anthropic anchor much of the contracted book in S-1 analyses; Nscale’s earlier public deals include multi-country Microsoft GB300 deployments and a West Virginia campus letter of intent for up to 1.35 GW of Nvidia Vera Rubin systems.
That gap between signed demand and live clusters is why a convertible this size matters in the neocloud cohort. Peers such as CoreWeave, Nebius, Crusoe, and Lambda also rent Nvidia GPUs by the hour against long customer contracts; CoreWeave, the closest public comp, priced its March 2025 IPO around a $23 billion valuation and later reported multi-billion-dollar annual revenue, while running a debt-heavy build model. Nscale is still pre-listing, with a March 2026 Series C that valued it at about $14.6 billion, and with media reports — not a priced range — putting IPO valuation talk near $25 billion to $35 billion and a possible offering of about $3 billion. Against that backdrop, a $3.36 billion convertible is less a vanity round than bridge financing for a balance-sheet business: customer prepayments and credit get you partway; convertibles and an IPO fill the rest before contracted megawatts can become revenue.
Convertible notes ahead of a listing also send a market signal. Investors get equity exposure tied to the public offering — often with economics that improve if the IPO clears a hurdle — while the issuer avoids locking a fresh private preferred price weeks before roadshow discovery. For the AI infrastructure boom, the structure says demand for financed AI factories is still large enough that hedge funds, credit shops, a Gulf investor, and Nvidia will underwrite billions against a book that is mostly still under construction. Nvidia’s dual role — supplier of the GPUs, repeat equity backer, and now a $1 billion note commitment — underscores how capital and silicon move together in this market.
The risk sits in the same place as the upside. Concentration, delivery, and financing risk travel with take-or-pay AI contracts; losses are large while depreciation and build costs run ahead of recognized revenue; and conversion math will decide how much of the post-IPO cap table belongs to noteholders. Nscale’s bet is that the convertible buys enough time and steel to turn a nine-figure revenue run-rate and a ten-figure backlog into a public AI cloud company. The IPO will test whether public markets price that backlog as destiny — or as work still left to do.