Everything About AI
NextFin News - Nscale Ltd. is facing resistance from prospective investors over the roughly $30 billion valuation it is pursuing in its initial public offering, with a cohort of investors arguing that a price tag between $16 billion and $19 billion is far more realistic. The pushback arrives as sentiment toward AI infrastructure companies deteriorates following a string of market setbacks, including the collapse of Australian rival Firmus's IPO plan and Microsoft's withdrawal from an anchor-tenant deal at Nscale's first British data center.
The disconnect is stark: Nscale filed for a U.S. listing on September 18 under the ticker NSCL, and has yet to begin formal marketing even though the required 15-day waiting period has elapsed. A valuation gap of that size - roughly 37 percent to 47 percent below the company's target - rarely closes without one side blinking. For a company that has built its public pitch on a $103.4 billion contract book, the question now is whether investors are pricing the backlog, or the risk that much of it never becomes revenue.
The Numbers Behind the Backlog
Nscale's prospectus, filed with the Securities and Exchange Commission, presents a company that looks enormous on paper and modest in operation. Total contract value stood at $103.4 billion as of the end of August. But only $2.6 billion of that was live, and just 55 megawatts of data-center capacity were switched on - all of it in Europe, while most of the signed capacity and the two largest contracts sit in the United States. That arithmetic works out to roughly $1.9 billion of contracted value for every megawatt actually delivering power today.
The revenue picture is thinner still. In the first six months of 2026, Nscale posted $140.6 million in revenue, a 1,252 percent jump from $10.4 million a year earlier, against a net loss of $1.02 billion. The company also pointed to $56.4 billion in remaining performance obligations. Concentration risk is explicit in the filing: Microsoft and Anthropic together account for 85 percent of the $103.4 billion contract book, or about $87.7 billion. In its own words, "a substantial portion of our revenue is driven by a limited number of our customers."
The two anchor contracts carry their own complications. Microsoft signed agreements worth about $43.8 billion through 2033, but pulled out of the Loughton, Essex project on October 18 after the site ran into power-constraint delays - a blow to the very first British data center Nscale hoped to showcase to public-market investors. The Anthropic agreement, worth $44.6 billion for computing power at a planned eight-gigawatt facility in West Virginia, is not yet financed; Nscale is targeting 2028 as the year it activates the facility's initial two gigawatts. Under the terms of that deal, Nscale must reach defined milestones and sustain reliable levels of computing performance, or Anthropic has grounds to walk away.
The valuation trajectory itself is part of the problem. In March, Nscale raised $2 billion in a Series C round led by Norway's Aker and 8090 Industries, with participation from Nvidia, Citadel, Dell and Jane Street, at a $14.6 billion valuation. The roughly $30 billion IPO target - reports have put the range at $30 billion to $35 billion - asks public investors to pay more than double that private-market mark in a matter of months, while the company's live capacity remains measured in tens of megawatts and its losses run above $1 billion a year.
The Neocloud Debt Reckoning
Nscale is not being judged in isolation. The valuation pushback is the latest symptom of a broader repricing of the "neocloud" model - specialist providers that lease AI computing capacity and borrow heavily to fund construction. On October 14, shares of Oracle, DigitalOcean, TeraWulf and Hut 8 all fell sharply after a chart of CoreWeave's finances circulated widely. DigitalOcean, TeraWulf and Hut 8 each dropped more than 10 percent; Oracle closed 8.8 percent lower; Dell Technologies and Advanced Micro Devices, both suppliers to the neoclouds, fell about 11.3 percent and 5.7 percent respectively.
CoreWeave, the sector's bellwether, saw its shares fall about 25 percent in a single week - their steepest weekly drop since November - after the yield on its bonds touched 15 percent. The debt math is unforgiving: CoreWeave owed $35.6 billion of borrowed principal on June 30, with $10.6 billion scheduled to be repaid by the end of 2027, and its interest expense, net, reached $640 million in the second quarter, or roughly a quarter of the $2.575 billion in revenue it booked. That is the transmission mechanism in plain terms: when borrowing costs rise, the leverage that funded the AI buildout stops being an accelerant and starts consuming the cash it was meant to deploy.
Nscale's own balance sheet reflects the same dependence. The company has raised more than $4.6 billion in debt since the start of the year. On September 15, three days before its IPO filing, it signed a subscription agreement for $3.1 billion in convertible notes - $2.1 billion in unsecured convertible loan notes plus a further $1 billion in notes or non-voting shares that will be issued to Nvidia. Nvidia is simultaneously Nscale's largest shareholder, its primary chip supplier, and the guarantor of approximately $860 million in lease obligations. The filing itself warns that even this close partnership does "not eliminate supply chain risks."
The Power Bottleneck Is the Real Story
Behind the valuation dispute lies a harder constraint: electricity. The Loughton site was cleared and had planning permission, with a provisional 90-megawatt grid connection agreement in the pipeline, yet the connection would not be available in time for the project's 2027 opening. Reports indicate that UK Power Networks has put sufficient grid capacity years beyond the planned start, with a timetable stretching into the early to mid-2030s. The site seeks up to 90 megawatts, and it sits in a UK queue of roughly 315 data centers requesting 73 gigawatts of connections.
This is not a one-company problem. Research from Sightline Climate, now trading as Currence, estimates that around a quarter of global data-center capacity ran into delays during 2025, and that between 30 and 50 percent of the capacity expected online in 2026 could slip because of power constraints, equipment shortages or local opposition. In the UK alone, more than 100 projects are reported to be considering gas generation while they wait in the connection queue. Provisional grid approval and actual power delivery are two separate processes with two separate timelines - a distinction that is foundational to understanding why so many AI infrastructure promises are arriving late.
The implication for Nscale's valuation is direct. A contract tied to a site without a dependable power connection is worth less than a contract tied to a live facility, and investors are beginning to price that difference. When Microsoft - itself an investor in Nscale as well as a client - walks away from the company's flagship British project, it is not just losing revenue; it is removing the anchor tenant that was supposed to validate the entire development model for public-market buyers.
What the Firmus Collapse Changed
Investors now have a direct precedent, and it is uncomfortable. Firmus, the Nvidia-backed Australian AI data-center operator, scrapped its planned $5 billion IPO on the Australian Securities Exchange on October 9, citing "recent market volatility and prevailing market conditions." The intended equity valuation was $30.6 billion - an enterprise value of roughly $60 billion including about $30 billion of debt. Investors labeled the valuation "excessive" and "mind-boggling," and orders pulled back after escrow terms would have allowed existing holders to sell more than half their stock at listing. Firmus said it would raise money privately instead.
The parallel is almost too neat: a Nvidia-backed AI data-center developer, a valuation hovering around $30 billion, a reliance on debt-financed construction, and a public market that refused the price. When a cohort of Nscale investors urges $16 billion to $19 billion, they are not pulling a number from thin air - they are pricing the asset against the most recent observable outcome in the same sector. A discrepancy in expectations between a company and its potential buyers can pressure a firm to slow down the process, and the valuation gap may reflect differing outlooks on where funding costs are likely to go and whether challenges to the broader sector may be quickly resolved.
Cyclical Headwind or Structural Repricing?
The central question for investors is whether this is a cyclical dip in sentiment or a structural reassessment of the neocloud business model. The cyclical case is straightforward: borrowing costs have risen, the IPO window has narrowed, and uncertainty over AI's long-term payoff has grown. If rates fall and AI demand keeps compounding, today's caution could prove to be a temporary discount. CoreWeave's network grew from three facilities to 51 over four years; the demand for AI compute has not disappeared.
But the structural case is stronger, and it rests on the business model itself. Neoclouds borrow at high rates to build data centers, then lease capacity to a handful of hyperscale customers at spreads that must cover debt service, power, and execution risk. The Firmus and CoreWeave episodes show that when financing costs rise, the model's thin margins are the first thing to break - and that is not a cycle that self-corrects on its own. The contracts that anchor these valuations are conditional: they depend on power delivery, on construction milestones, and on financing that has not yet been secured. A contract that can be walked away from is not revenue; it is an option on revenue.
The verdict: the timing of Nscale's valuation gap is cyclical - a function of this quarter's sentiment and rates - but its magnitude is structural. The era of pricing AI infrastructure on the face value of signed contracts, rather than on powered, delivering capacity, appears to be over. Investors are no longer paying for the backlog; they are discounting it for the probability that it never converts.
What Comes Next
In the short term, the pressure is on Nscale to either accept a lower valuation or delay the offering. The company has not set a price range, and with formal marketing not yet begun, it retains the option to wait. A trajectory that asks for more than double the March private valuation, while live capacity stays in the tens of megawatts and losses run above $1 billion a year, is difficult to defend in a market that has just rejected a near-identical $30 billion data-center valuation.
Three scenarios define the path. The base case is a repricing toward the $16 billion to $19 billion range investors are urging, or a postponement until the neocloud credit spread narrows. The upside case requires two things: a fall in borrowing costs and evidence that the Anthropic and Microsoft contracts survive their milestone tests and move from paper to powered capacity. The downside case is a second credit event in the sector - another neocloud default or another scrapped IPO - which would widen the discount further and could force Nscale to lean more heavily on Nvidia's balance sheet.
The beneficiaries of this repricing are the parties with capital and power: hyperscalers like Microsoft and Anthropic, which can dictate terms to leveraged builders, and Nvidia, which sits on both sides of the transaction as shareholder, lender, and chip supplier. The exposed are the highly leveraged neoclouds, their creditors, and equipment suppliers whose order books depend on neocloud capex continuing to expand.
The single signal that would falsify the structural-repricing thesis is observable and specific: if CoreWeave's bond yield falls back below roughly 9 percent and two neocloud IPOs price at or above contracted-value multiples within the next six months, then this is a cyclical dislocation, not a regime change. Until then, the market is telling AI infrastructure companies that a signed contract is not the same as a switched-on data center.
"A substantial portion of our revenue is driven by a limited number of our customers."
Nscale's prospectus may be built on $103 billion of signed contracts, but the market is now pricing the 55 megawatts that are actually switched on - and on that measure, every AI data-center developer is being asked to prove that its backlog is a backlog of revenue, not a backlog of problems.
Explore more exclusive insights at nextfin.ai.
