Everything About AI
NextFin News - Nscale, the London-founded AI data-center builder backed by Nvidia, has publicly filed for a US initial public offering, seeking a New York Stock Exchange (NYSE) listing under the ticker NSCL as its book of contracted revenue swells past $100 billion. The filing, made public on September 18, 2026, sets up one of the most closely watched tests of the "neocloud" boom since CoreWeave's March 2025 listing: whether public markets will pay a hyperscaler-style premium for a two-year-old company whose revenue is still measured in the low hundreds of millions but whose contract book already rivals the largest infrastructure deals in technology history.
The Filing and the Numbers Behind It
The filing arrives barely 18 months after Nscale was spun out of Melbourne crypto miner Arkon Energy in May 2024 by co-founders Josh Payne and Nathan Townsend. In that short window, the company has assembled what it describes as more than $100 billion in total contracted revenue, up from roughly $51 billion in early August, according to documents shared with prospective investors. The contracts average 5.7 years in duration, implying roughly $18 billion in annualized contracted revenue — a figure that, if recognized, would place Nscale in the same revenue tier as established cloud infrastructure names.
The scale of the ambition is visible in the hardware. Nscale says it has about 25,000 GPUs active today, with roughly 289,000 active and contracted as of the second quarter of 2026, including approximately 194,000 of Nvidia's next-generation Vera Rubin accelerators. Its power footprint is the binding constraint and the prize: around 831 megawatts of active and contracted capacity, with plans to add roughly 10 to 11 gigawatts across more than a dozen sites spanning Norway, the UK, the US and Portugal. Building a gigawatt-scale AI campus takes years and billions of dollars; Nscale's bet is that the market will value the contracted backlog as de facto revenue visibility.
Revenue, by contrast, is still small. Executives told investors revenue topped $100 million in the second quarter of 2026, up from about $37 million in the first quarter and roughly $33 million for all of 2025. That gap — a quarterly revenue run-rate just above $100 million against a nine-figure contract book — is the central tension of the IPO. Nscale is asking investors to underwrite the infrastructure build-out today on the promise that the contracted revenue converts to cash over the next half-decade.
The company is not alone in the window. CoreWeave, the New Jersey neocloud that listed in March 2025, raised $1.5 billion at its IPO; Nscale is seeking to raise as much as $3 billion, roughly double, according to people familiar with the matter. Goldman Sachs and JPMorgan are running the deal. The timing follows a wave of pre-IPO financing: a $900 million revolving credit facility in July, a $1.4 billion delayed-draw term loan in February, and a $3.5 billion pre-IPO round that includes as much as $1.5 billion of convertible notes carrying a double-digit discount to the IPO price with a conversion cap at a $30 billion valuation, plus roughly $2 billion of financing from Nvidia. That cap sits well above the $14.6 billion valuation set by Nscale's $2 billion Series C in March 2026.
So what should investors actually make of a $100 billion backlog? The answer depends less on the headline number than on what kind of asset Nscale really is.
The Backlog Is Real — But It Is Not Revenue
The first discipline is definitional. Nscale's "$100 billion" is the gross value of multi-year compute contracts signed with customers such as Microsoft, Anthropic and Figure. Microsoft is the anchor tenant: in October 2025 the company announced an expanded deal for approximately 200,000 Nvidia GB300 GPUs across Europe and the US, including a roughly 240-megawatt hyperscale campus in Texas supporting phased delivery from the third quarter of 2026, and the Aker-Nscale joint venture in Narvik, Norway, is delivering approximately 52,000 GB300 GPUs. In September, reports emerged that Anthropic would spend $45 billion to rent AI cloud computing power from Nscale's West Virginia campus.
But signed contracts in capital-intensive infrastructure are not the same as recognized revenue. They are contingent on the company actually delivering power, cooling, and working GPU clusters on schedule. Every month of construction delay pushes revenue recognition further out while interest expense on the debt used to finance the build keeps accruing. The 5.7-year average contract length cuts both ways: it locks in a long revenue tail, but it also means the annualized yield — roughly $18 billion — is only realized if execution holds across a half-decade of permitting, grid interconnection, and hardware delivery.
This is the mechanism that separates the neocloud model from software. A software company signs a contract and recognizes revenue as service is delivered with near-zero marginal cost. A neocloud must spend billions before the first dollar is recognized. The backlog is therefore best read as a measure of demand visibility, not profitability. Investors pricing Nscale against CoreWeave's roughly $104.2 billion backlog should note that CoreWeave's backlog-to-revenue conversion has been the market's yardstick — and that yardstick has been volatile.
The Second-Order Question: Who Owns the Pricing Power?
The first-order read of the filing is simple: AI compute demand is insatiable, so whoever builds data centers wins. The second-order question is harder: in a value chain of chip supplier, data-center builder, and hyperscaler customer, who captures the margin?
Nscale's answer is vertical integration plus location arbitrage. By building above the Arctic Circle in Norway, on the Portuguese coastline, and along the Ohio River — places where power is cheap and available — it aims to offer compute at a lower cost base than hyperscalers can achieve in power-constrained Northern Virginia. Josh Payne, the company's chief executive, has argued that AI-native companies will dominate the next era of technology.
"The largest businesses on earth in the next five years are going to be AI-native," Payne said.
But the counter-pressure is real. Nvidia is both Nscale's backer and its supplier, and it has invested heavily across the neocloud ecosystem — including $2 billion in CoreWeave in January 2026 — which gives the chipmaker leverage over allocation of its most advanced accelerators. If Nvidia tightens supply to protect its own cloud partnerships or to favor one neocloud over another, Nscale's delivery schedule is at risk. And the customers are not passive: Microsoft and Anthropic are themselves building massive in-house compute fleets. The moment hyperscalers decide that owning capacity is cheaper than renting it — or that renting from a competitor is cheaper — Nscale's pricing power compresses.
The competitive pressure is no longer theoretical. In July, shares of CoreWeave and peer Nebius fell sharply after reports that Meta Platforms is developing an internal initiative to rent out its excess cloud computing capacity to third-party customers — turning one of the neoclouds' largest customers into a potential direct rival. CoreWeave's own shares, which more than doubled from their $40 IPO price to an all-time closing high of $183.58, have since given back much of the gain and trade around $81, down more than 50% from that peak.
Cyclical Wave, Structural Shift — or Both?
The cleanest way to frame the investment is to separate the cyclical leg from the structural one.
The cyclical leg is the GPU capacity crunch. Supply of advanced accelerators and grid interconnection slots is tight today; anyone who holds a contract and a power connection can charge a premium. Cyclical claims require evidence of mean reversion, and here the evidence is emerging: the Meta news in July sent neocloud shares down roughly 15% in a single session, and CoreWeave's stock has fallen more than 50% from its peak as investors reassessed how much of the capacity premium is durable. Capacity that was scarce in 2025 becomes commoditized in 2027. If the cycle turns, contracted pricing gets reset at renewal, and the backlog's present value falls.
The structural leg is different and stronger. AI inference demand is not a single boom-bust cycle; it is a step-change in the global demand for compute, underpinned by a physical constraint — electricity and interconnection — that cannot be solved by writing code. Permitting and grid build-out take a decade, not a quarter. That is why Nscale's real asset is not its GPUs but its power rights and sites. A company that controls gigawatts of permitted, connected power in a power-constrained world holds a structural scarcity asset, regardless of which chip generation sits in the racks.
The verdict: the near-term valuation is cyclical and vulnerable to a capacity glut; the long-term asset base is structural if, and only if, Nscale converts sites into connected power faster than competitors. The backlog is the claim on that future; execution is the proof.
The Adversarial Case — and the Signal That Would Break It
The strongest counter-thesis is not that AI demand will fade. It is that Nscale's contract figures are illustrative rather than firm guidance — a caution one source familiar with the discussions raised about the $103 billion figure — and that the company is raising capital at a valuation that prices flawless execution. At a $30 billion conversion cap on the pre-IPO notes, and a potential IPO valuation reported near $25 billion, the market is already assigning a large premium to revenue that has not yet been recognized. If even a fraction of the $100 billion backlog is delayed, renegotiated, or canceled, the equity has little margin of safety.
There is also a customer-concentration risk that the filing will lay bare. Microsoft alone has commissioned 1.35 gigawatts of capacity from Nscale — a single customer relationship large enough that any change in Microsoft's capital-spending posture would move the stock. CoreWeave has faced the same critique, and it has not stopped that stock from rising sharply since its IPO; but it remains the single biggest underwriting risk in the sector.
The falsifying signal is specific and observable: watch Nscale's quarterly revenue recognition against its contracted backlog over the next four quarters. If revenue grows at less than half of the annualized contracted run-rate — that is, if recognized revenue stays far below the roughly $18 billion annualized figure implied by the contract book — the backlog-conversion thesis is broken, and the premium valuation cannot be sustained. A second falsifier: any disclosure that a named anchor customer has delayed or reduced a gigawatt-scale commitment.
What Comes Next: Three Horizons
What happens next is a test of public-market appetite for infrastructure-duration assets. In the short term, the stock will trade on sentiment around the AI capital-spending cycle and on comparisons to CoreWeave's post-IPO path — a path that has included both a more-than-doubling from the IPO price and a drawdown of more than 50% from the peak. Liquidity and lock-up dynamics, not fundamentals, will dominate the first months.
Over the medium term, the fundamental question is conversion: can Nscale turn $100 billion of signed contracts into recognized revenue at a pace that justifies a $25 billion to $30 billion valuation? The base case is that the company delivers steady quarterly revenue beats as its Texas, West Virginia, and Norway campuses come online through 2027, supported by the $900 million revolver and the pre-IPO convertible notes. The upside case is that power scarcity becomes the binding constraint across the industry, turning Nscale's permitted sites into repricing assets and pushing the valuation toward the note conversion cap and beyond. The downside case is that construction delays, grid interconnection bottlenecks, or a hyperscaler pullback leave the backlog stranded on the balance sheet while interest expense compounds.
Who benefits and who is exposed: a successful Nscale listing validates the "compute landlord" thesis and lifts the entire neocloud complex — CoreWeave, Nebius, and private peers such as Lambda. It also benefits Nvidia, whose accelerator allocation underpins the whole model. The exposed are investors who treat the backlog as revenue; the contract book is an option on execution, not a bank balance.
The split verdict, by horizon: short term, trade the sentiment and the CoreWeave correlation; medium term, watch revenue recognition versus backlog; long term, bet on power, not chips. The GPUs will be obsolete in three years. The gigawatts, if permitted and connected, will compound for a decade.
Nscale's IPO is not really a bet on artificial intelligence — it is a bet that electricity, permitting, and interconnection are the true scarce resources of the AI era, and that a two-year-old company can convert a $100 billion promise into a power grid before the cycle turns.
Explore more exclusive insights at nextfin.ai.
