NextFin News - Nscale, the London-based AI infrastructure builder, has raised $3.36 billion in a pre-IPO financing round led by the hedge fund Third Point, converting a nine-figure contract backlog into committed capital days before the company files to list on the New York Stock Exchange. The deal is the final private-money checkpoint before a public-market verdict on whether $103.4 billion of promised compute revenue is worth a $35 billion valuation.
The raise lands at the hinge of Nscale's short history. The company filed its Form S-1 on September 18, seeking an NYSE listing under the ticker "NSCL," with underwriters sizing an offering of roughly $2 billion and a target valuation near $35 billion reported ahead of the filing. The $3.36 billion round is structured as convertible notes — debt that turns into equity at the IPO — offered at a double-digit discount to the eventual offering price, according to people familiar with the terms. That discount adjusts upward as the valuation climbs but stops at a $30 billion ceiling; above that level the conversion price stays fixed. In effect, Third Point and its co-investors are underwriting a roughly two-fold jump from the $14.6 billion valuation Nscale set in its March Series C, with less than six months between the two marks.
The anchor of the financing is Nvidia. On September 15 the company entered a subscription agreement for a minimum of $3.1 billion in convertible financing, comprising $2.1 billion of unsecured convertible loan notes plus a further $1.0 billion portion to be issued to Nvidia as notes or non-voting shares, according to the prospectus. The notes convert automatically into shares on completion of the offering. The Nvidia portion is expected to close on or around November 16. Goldman Sachs is advising on the fundraising.
The timing is not accidental. The capital arrives days after Nscale disclosed that its active and contracted total contract value reached $103.4 billion as of August 31, up from $38 billion at the end of 2025. Nearly all of the increase came from two agreements: a contract with Anthropic signed August 25 for aggregate payments of up to $44.6 billion, and a Microsoft commitment of up to $43.8 billion through 2033. Together those two customers represent roughly 85 percent of the total contract value.
But the backlog and the bank account tell two different stories. Revenue for the six months ended June 30 was $140.6 million — a 1,252 percent year-over-year increase from $10.4 million, but a rounding error against a nine-figure backlog. The net loss for the same period was $1.02 billion, wider than the $368.9 million loss a year earlier. About $457 million of that loss came from fair-value adjustments, with an operating loss of roughly $492 million. Cost of revenue ran $189.6 million, meaning Nscale failed to make a gross profit: it cost more to deliver its services than it collected from customers. The prospectus states that management once had "substantial doubt" about the company's ability to continue as a going concern.
That is the tension this round resolves, at least for now. The $3.36 billion is not ordinary growth capital. It is the bridge between a private valuation built on contracts not yet performed and a public-market verdict that will be cast on revenue not yet earned.
What the $3.36 Billion Actually Buys
The first question is not whether Nscale needed cash. It did, and urgently. A company that lost more than $1 billion in six months, with negative gross margin, cannot fund a multi-gigawatt buildout from operations. The $1.4 billion delayed-draw term loan it signed in February — backed directly by its GPU fleet and led by PIMCO, Blue Owl and LuminArx — was a down payment. The pre-IPO notes are the rest of the bridge.
The structure reveals what the sophisticated lenders were really buying. Convertible notes with a discount and a valuation cap are a hedge against uncertainty: if the IPO prices below expectations, note holders convert at a discount and win; if it prices above the cap, they still participate at the capped price. The $30 billion cap is the point at which the lead investors decided the discount was no longer worth extending. That is a narrower, more cautious signal than the $35 billion valuation the IPO is chasing.
Nvidia's role is the more revealing piece. The chipmaker is simultaneously Nscale's supplier, its customer and its creditor — a triangular relationship that recurs across this cycle's AI infrastructure buildout. Nvidia sells Nscale the Vera Rubin GPUs; Nscale rents the resulting compute to AI labs; and Nvidia's financing effectively helps fund its own customers' purchases. The arrangement de-risks Nscale's balance sheet while securing demand for Nvidia's next-generation silicon. It is vertical integration achieved through financial engineering rather than ownership.
"Management identified that the Company's forecast funding requirements included reliance on uncommitted debt and equity financing, which initially raised substantial doubt about the Company's ability to continue as a going concern," the prospectus said. "Management has evaluated its ability to defer, reduce or cancel capital expenditure if such financing is not obtained as forecast and has concluded that this plan is probable of being effectively implemented and, if necessary, would provide sufficient liquidity to alleviate the substantial doubt."
The proceeds, per the filing, will fund data center projects and deployments, technology development, working capital and possible acquisitions. In plainer terms: they pay for the GPUs, the power plants and the construction required to turn $103.4 billion of contracts into revenue a public market can audit quarter by quarter.
The Backlog Is Real. The Conversion Rate Is the Question.
Nscale's headline number is a total contract value of $103.4 billion. That figure is company-reported and, critically, it is not recognized revenue. Only $2.6 billion of it was active as of August 31. The rest depends on counterparties actually drawing down capacity over multi-year terms — and, in many cases, on data centers, power plants and GPUs that have not yet been built.
The more auditable measure in the prospectus is remaining performance obligations of $56.4 billion as of June 30, spread over the next seven years. Even that number is a promise rather than a deposit, and $7.1 billion of it reflects a financing component on customer prepayments — money clients have advanced ahead of delivery. The filing also discloses that it had no binding commitments for the project financing some of these contracts require when the prospectus was filed.
The concentration risk is stark. One customer accounted for 52 percent of revenue in the first half of 2026, the filing shows. On the backlog side, Anthropic alone represents roughly 43 percent of total contract value, and Microsoft and Anthropic together about 85 percent. The Anthropic agreement gives the AI lab access to 460 megawatts of compute capacity at Nscale's Monarch campus in West Virginia, using Nvidia's Vera Rubin chips, with the facility expected to come online at the end of 2027. If Anthropic's own demand for inference compute slows, or if construction delays push capacity online later than contracted, the backlog investors are being pitched does not automatically convert into the cash flow a public market values.
There is a second, quieter concentration: Nvidia. The chipmaker has invested more than $2 billion cumulatively in Nscale, supplies the GPUs, and now holds a convertible position in the financing. A shock to Nvidia's supply chain — export controls, fabrication bottlenecks, or a generation delay — would hit Nscale on three axes at once.
The hardware gap makes the conversion question concrete. Nscale reported roughly 25,000 active GPUs against 461,000 active or contracted units at August-end — a ratio of about one to eighteen. The distance between those two numbers is the distance between the contract book and the revenue line.
The comparison with CoreWeave, the neocloud category's first major public listing, is instructive and mostly unflattering. CoreWeave delivered $5.13 billion of revenue in 2025 and carries a market capitalization of roughly $48 billion to $50 billion, with a revenue backlog of $66.8 billion reported at its 2025 results. Nscale is asking the public market to value a $35 billion company that has collected $140.6 million in six months — roughly one thirty-sixth of CoreWeave's full-year revenue — and has never posted an annual profit. The gap between the two is not a valuation dispute; it is a business-model question. CoreWeave sold proven, delivered capacity. Nscale is selling a promise of capacity that must still be built.
One independent analyst captured the bear's framing before this financing closed: the single largest contract in the backlog "runs through a campus that isn't built, on power that isn't powering yet, funded by money that hasn't been raised." The money has now been raised. The campus and the power remain the test.
Cyclical Losses on a Structural Shift
The easy read of Nscale's losses is that they are start-up losses — front-loaded capital expenditure that flips to profit once the data centers are live. That is partly right, but it misses the deeper change. The AI infrastructure buildout is a structural shift in how compute is financed and owned, and Nscale is one of its purest expressions.
Three pieces of evidence support the structural call. First, demand is locked in by contracts rather than spot purchases: the Anthropic and Microsoft agreements, a letter of intent with Microsoft for 1.35 gigawatts at the West Virginia campus, and the Stargate Norway project with Aker and OpenAI are multi-year commitments that would survive a cyclical downturn in chip sentiment. Second, the financing side has been rewired: GPU-backed term loans, convertible notes from the chip supplier, and sovereign and hedge-fund capital flowing into physical infrastructure form a capital stack that did not exist five years ago. Third, the asset itself has become the collateral — Nscale's February term loan was backed directly by its GPU fleet, treating chips as bankable infrastructure the way a utility treats turbines.
The cyclical counterweight is real and must be kept separate. The pace of AI capital expenditure is cyclical; hyperscalers can and do slow orders when utilization lags. Nscale's negative gross margin is a start-up condition that should mean-revert as capacity comes online and fixed costs spread across more billable hours. But the regime underneath — AI labs renting dedicated capacity from specialized infrastructure owners rather than building it themselves — is the structural leg. It will not revert on its own.
That distinction matters for valuation. A cyclical read says losses will shrink and the stock will be judged on earnings. A structural read says the market will pay for contracted capacity growth, the way it paid for cloud buildouts in the 2010s, and tolerate losses longer. Nscale is pricing itself on the structural story while reporting cyclical start-up financials.
The company's own founder frames the opportunity in those terms. "This is the fourth industrial revolution; the world is changing at a rapid pace," Josh Payne, Nscale's chief executive, said when the Series C closed in March. "Nscale is leading this buildout. We are building this foundation that the market sits on, the engine of superintelligence."
The Strongest Case Against Nscale
The bear case is not that Nscale will run out of money. After this round, it will not. The bear case is that the company is a highly leveraged bet on a small number of customers' demand curves, financed in part by its own supplier, and that the public market will refuse to pay a premium for backlog it cannot yet audit.
The concentration numbers do the heavy lifting. Roughly 85 percent of the backlog rests with two customers. Anthropic, in turn, has been racing to secure capacity after its own infrastructure strain hit reliability, signing agreements with Advanced Micro Devices, SpaceX, Google and Broadcom in recent months. A customer diversifying its supply chain is a rational customer; it is also a customer with optionality that weakens any single provider's pricing power. If Anthropic renegotiates, delays drawdowns, or shifts workloads to its own builds, a large share of the $44.6 billion moves with it.
The second leg is the gross-margin problem. Cost of revenue of $189.6 million against $140.6 million of revenue is not a temporary accounting quirk; it is the signature of a company buying GPUs at premium prices and renting out capacity before facilities are fully utilized. The path to positive gross margin requires utilization rates the company has not yet demonstrated at scale. CoreWeave proved that model in public with $5.13 billion of revenue before the market gave it a roughly $48 billion valuation. Nscale is asking for a comparable premium with about one thirty-sixth of that revenue.
The third leg is the financing structure itself. When your largest creditor is also your largest supplier and a major customer, the conflicts are inherent. A downturn that hurts Nscale's ability to repay also hurts Nvidia's receivables and its chip demand. The alignment is strong in an up-cycle and fragile in a down-cycle.
This case is serious enough that it cannot be waved away with a growth narrative. It rests on disclosed numbers, not speculation. Nscale's defense is that its contracts are long-term and committed in structure, that its campuses are designed to serve multiple tenants rather than a single anchor, and that the $3.36 billion removes the going-concern overhang that would otherwise force a distressed capital raise. Whether that defense holds depends on one observable outcome: revenue recognition, quarter by quarter.
There is also the question of who gets paid while losses mount. The prospectus shows founder and chief executive Josh Payne receiving roughly $23.2 million in compensation for the period, and president of APAC Jing Yin roughly $24.7 million, while Sheryl Sandberg and Nick Clegg sit on the board with share grants of about 1.9 million and 917,460 shares respectively. The pay is not the story; it is a reminder that in a company burning cash, dilution and governance matter as much as backlog.
What Comes Next
The immediate beneficiaries of this round are clear. Nscale buys time — roughly the next 18 to 24 months — to convert backlog into revenue without returning to the capital markets under duress. Nvidia buys demand security for its Vera Rubin generation and a financial stake in the infrastructure layer that will consume it. Third Point buys a discounted entry into what could be the defining infrastructure listing of the cycle, with the $30 billion cap as its downside marker.
The exposed parties are the public-market investors who will be asked to underwrite the gap between $103.4 billion of contracted value and $140.6 million of collected revenue. The first amendment to the S-1 — the one that sets the price range and share count — will be the first real test. If the offering prices near the reported $35 billion target, the market is endorsing the backlog-as-revenue thesis. If it prices materially below the $30 billion note cap, the private round will have been the smarter money.
The watchlist is specific. First, the S-1/A price range and share count, blank in the initial filing. Second, quarterly revenue recognition against the $56.4 billion of remaining performance obligations; the first two public quarters will show whether backlog converts at the pace the pitch deck implies. Third, gross margin: a move from negative toward positive is the single best indicator that utilization is catching up with capacity. Fourth, any disclosure of how much of the Anthropic and Microsoft contracts is committed versus contingent — that clause, more than the headline dollar figure, determines how much of the $88.4 billion is real.
Split by horizon: in the short term, sentiment will follow the IPO pricing and the initial order book. In the medium term, fundamentals will follow utilization and gross-margin trajectory, not backlog announcements. In the long term, the structural question is whether AI labs continue to outsource capacity to specialized infrastructure landlords like Nscale, or whether hyperscalers and the labs themselves bring more of the buildout in-house.
The falsifying signal is concrete: if Nscale reports two consecutive quarters in which revenue recognized from remaining performance obligations grows by less than 10 percent sequentially while gross margin stays negative, the structural-premium thesis is wrong and the stock should be judged as a capital-intensive utility, not a growth compounder.
Nscale has turned a backlog into a balance sheet. What it must now turn into revenue is the market's patience.
Explore more exclusive insights at nextfin.ai.
