Inside the Nscale IPO Gamble Nobody Is Pricing In

Inside the Nscale IPO Gamble Nobody Is Pricing In

Nscale wants public markets to value it at up to $35 billion while racing to build a massive AI compute campus in Mason County, West Virginia, backed by an Anthropic contract worth up to $44.6 billion over six years. Yet the London-based neocloud operator has already walked back its earlier twelve-month funding assurance in an amended SEC filing, admitting it needs fresh financing for 2027 against non-cancellable hardware orders.

That single admission exposes the central tension inside Nscale's New York Stock Exchange listing push under the ticker NSCL. A headline contract book topping $103.4 billion sounds invincible on a roadshow slide. Turning giant multi-year compute promises into energized megawatts, delivered GPUs, and positive cash flow is another matter entirely.

I have watched enough infrastructure cycles to spot the difference between contracted backlog and bankable cash flow. Back in the late 1990s telecom build-out, carriers signed sprawling capacity plans and equipment vendors eagerly financed the expansion, right up until debt service and utilization math collided with reality. AI compute is not telecom fiber, and Nvidia is not Lucent, yet the structural trap rhymes uncomfortably well.

How a Bitcoin Spinoff Became a Hundred-Billion-Dollar Backlog

Nscale started in 2024 as a spinoff from Australian cryptocurrency and Bitcoin mining operator Arkon Energy, pivoting from mining rigs to AI data centers. Moves like that used to be dismissed as opportunistic rebranding. Here, the pivot coincided with an acute shortage of powered AI capacity and a wave of sovereign and hyperscale deals.

By March 2026, Nscale raised $2 billion in a Series C round at a $14.6 billion valuation led by Norwegian energy group Aker and 8090 Industries, with participation from Nvidia, Dell, and Nokia. Six months later, in September 2026, it filed its Form S-1 with the SEC aiming for an NYSE float that media reports placed as high as $35 billion, alongside a $3.36 billion pre-IPO convertible note raise involving Nvidia and funds managed by Apollo.

Look closely at the numbers behind that rapid ascent and the picture gets far more complicated. As of August 31, 2026, Nscale reported roughly $103.4 billion in total contracted value, up from $38 billion at the end of 2025. Only about $2.6 billion of that contract value tied to active capacity already in place.

Roughly 25,000 GPUs were active across its footprint, while total commitments spanned 461,000 GPUs across five operational and twelve contracted data center sites. In the first half of 2026, revenue jumped more than 1,200% year over year from $10.4 million to $140.6 million. During the same six-month stretch, net losses reached roughly $1 billion, and nine-month figures showed nearly $1.7 billion in losses against $307 million in revenue.

Rapid top-line growth from a tiny base does not automatically validate a $35 billion valuation. It proves demand exists for scarce compute. It does not prove Nscale can finance the capex wall before hardware obligations come due.

Why Monarch Is Both the Prize and the Pressure Point

Everything around Nscale's listing circles back to the Monarch Compute Campus on more than 2,250 acres in Mason County, West Virginia. Nscale pitches Monarch's first phase at roughly 1.35 to 1.38 gigawatts of AI compute capacity, with potential expansion above 8 gigawatts over time and more than $20 billion in planned capital investment.

In August 2026, Nscale entered GPU services agreements with Anthropic covering up to approximately $44.6 billion over six years tied to Monarch, with delivery targeted around late 2027 or early 2028. Reports also emerged in October 2026 that Nscale was in talks with Anthropic for a second deal of roughly $45 billion for an additional 460 megawatts at the site.

On paper, landing Anthropic at that scale turns Nscale from a regional European operator into a premier U.S. AI infrastructure contender. In practice, it creates a brutal timing mismatch.

In its November 12, 2026 Form S-1 amendment, Nscale noted that after its year-end 2025 assessment it took on significant new capital purchase orders for GPUs, networking, and related equipment tied to its Anthropic agreements, and some of those commitments cannot be deferred, reduced, or cancelled. At the same time, the prospectus stated Nscale had not yet obtained binding commitments for all financings needed to perform under the Anthropic contracts, projecting roughly $10 billion in data center and power capex for Building 1 alone.

A non-cancellable chip order does not care whether your project debt closed on schedule. Hardware vendors bill on delivery milestones, while equity markets punish sudden capital calls.

If a hypothetical developer signs a massive multi-year customer contract, buys non-cancellable server racks early to lock in supply, and then hits a six-month window to line up site financing, the risk shifts immediately onto the developer's balance sheet. That is the exact squeeze investors are trying to price. Preliminary figures in the November amendment put cash and cash equivalents at $3,575.9 million as of September 30, 2026, while earlier disclosures pointed to billions in debt and tens of billions in outstanding equipment and construction commitments.

Off-Grid Power Solves One Bottleneck and Creates Another

Nscale's pitch for Monarch emphasizes generating power on-site through a microgrid developed with Caterpillar, avoiding draws on local utility grids or municipal drinking water via closed-loop cooling. That positioning is deliberate. Power queues and substation delays have become the single biggest killer of data center schedules.

Yet going behind-the-meter with generation engines—reported in local permitting coverage as including nearly 1,000 natural-gas and diesel engines in the initial phase—trades utility queue risk for fuel logistics, permitting scrutiny, and operating cost exposure. Air-quality approvals, sound mitigation, and ongoing fuel supply matter just as much as nameplate gigawatts.

Even when a site bypasses the local utility grid, it still has to clear environmental rules and stay economical once live. A power plant that works on a blueprint still has to run cleanly and cheaply for six years.

Customer Concentration and the Myth of Equal Contracts

One habit in AI infrastructure coverage is treating every dollar of backlog as equal. It isn't.

Roughly 85% of Nscale's $103.4 billion contract book comes from just two customers, Anthropic and Microsoft, with Microsoft's agreements recorded at up to $43.8 billion through December 2033 in the September filing. Meanwhile, Nscale's largest customer accounted for 52% of revenue during the first half of 2026.

When two buyers hold that much of your backlog, concentration risk is not a footnote. If one customer slows a rollout, disputes a delivery schedule, or renegotiates terms, your revenue visibility shrinks overnight.

Consider what happened in Essex. Microsoft withdrew from Nscale's planned data center project in Loughton, in the UK, where grid power faced multi-year delays. That exit underscored a hard truth: a signed customer interest or planned expansion means little until power arrives on schedule.

Then look at payment quality across the sector. Reports have noted structure variations such as Figure paying half of a $3.5 billion Texas deal in shares rather than pure cash. Equity-settled or milestone-contingent contracts behave very differently from steady, investment-grade cash inflows when you are servicing heavy debt and paying for GPUs.

Risk Dimension What Bulls Point To What Skeptics Watch Closely
Order Backlog Over $103.4 billion in total contracted value as of late August 2026 Only ~$2.6 billion tied to active capacity today, plus non-cancellable chip commitments
Customer Mix Multi-year scale with Anthropic and Microsoft Two customers represent ~85% of backlog; Microsoft exited Loughton over grid timing
Power Strategy Off-grid microgrid with Caterpillar and closed-loop cooling at Monarch Permitting, fuel supply, and operating costs for large-scale on-site generation
Capital & Listing $3.36 billion pre-IPO convertible financing and NYSE filing Dropped 12-month funding assurance for 2027; public AI compute stocks under pressure

Circular Capital and Public Market Skepticism

You cannot analyze Nscale's $35 billion ambition in isolation from the ecosystem around it. Nvidia is an investor, a convertible financing participant, a GPU supplier, and a commercial partner across Nscale's build-out.

There is nothing illegal about strategic suppliers backing infrastructure builders. In fact, tight GPU supply makes close coordination almost mandatory when deploying hundreds of thousands of accelerators.

Still, seasoned institutional buyers ask tough questions when the same chipmaker supplies the hardware, helps fund the capital stack, and anchors the growth story. If public investors start questioning AI capex returns—or if rival neocloud names like CoreWeave and Nebius face share-price pressure—valuation multiples compress fast.

That is why reports in November 2026 suggested Nscale was weighing whether to delay its $35 billion IPO until 2027 amid a broader AI stock pullback and post-midterm policy questions around U.S. data center expansion. A company valued at $14.6 billion in March 2026 has a lot to prove to convince public investors to pay more than double that figure months later while losses widen and 2027 funding remains open.

The counter-argument from Nscale's supporters is straightforward. Compute demand from frontier model labs is growing at historic scale, and locked-in power plus committed capacity gives Nscale scarce inventory that latecomers cannot easily replicate. Even if the listing slips or prices below the top end, long-term contracts with major buyers can still anchor the business.

What Actually Determines Whether Nscale Wins

Forget the press-release hype around record backlogs and target valuations. Three concrete tests will decide whether Nscale justifies a premier listing or faces a painful repricing.

First, watch binding project and 2027 financing. Non-binding intent or future hopes do not pay for turbines, steel, or GPU shipments. Investors need clear visibility into how much extra capital is required, at what cost of debt or equity, and on what timeline.

Second, track energization and delivery milestones at Monarch and across Nscale's European and U.S. sites. A gigawatt on paper earns zero dollars until power flows to live racks and customer acceptance clocks start ticking.

Third, look at unit economics once scale arrives. High depreciation, heavy interest costs, and steep power expenses can eat up seemingly generous top-line contracts if utilization slips or hardware generations turn over faster than expected.

Demand for AI compute is real, yet capital discipline is what separates durable infrastructure giants from cautionary tales.

KM

Kenji Mitchell

Kenji Mitchell has built a reputation for clear, engaging writing that transforms complex subjects into stories readers can connect with and understand.