Nscale: $45 Billion in Contracts, 30 Megawatts in Service

In under a month this summer, three things happened to a company most people hadn't heard of a year earlier: Nscale bought the software firm Anyscale for $1.65 billion, agreed — according to Bloomberg's reporting — to a six-year deal with Anthropic worth $45 billion, and, per the same reporting, filed toward a US IPO of up to $3 billion.

The compression in that comparison is the point. The Glomfjord facility in Nordland — the only Nscale-operated site with published figures on its own website — delivers 30 MW of operational compute today, expandable to 60 MW. The Anthropic contract alone amounts, if the reporting is right, to roughly $7.5 billion a year for six years. The company is selling six years of compute it has partly not built yet — and going public to close the financing gap in the meantime.

From Fjord to West Virginia

Nscale describes itself as a company that designs, builds, and operates AI data centers, with a portfolio that — per its own website — spans Norway (Oslo, Stavanger, Narvik, and Glomfjord) through the UK, Iceland, and Portugal to Texas, North Carolina, and West Virginia. Some locations are partner-owned rather than Nscale-operated — a nuance worth holding in mind when reading "global footprint."

Glomfjord is the concept in miniature: Arctic hydropower, 100 percent renewable energy, 30 MW today and a 60 MW ceiling. Those are real, delivered megawatts — but a fraction of what the company has now committed to.

The big card is the Monarch Compute Campus in West Virginia, which Nscale presents as the first state-certified AI microgrid in the United States. Phase one is engineered for 1.35 GW of compute capacity, with delivery targeted for early 2028, and an expansion path toward more than 8 GW. West Virginia's governor, Patrick Morrisey, has publicly praised the reportedly $45 billion commitment to the state. But for now, Monarch is a construction project with a date, not a facility with power on the terminals.

The Contract Is Both Asset and Liability

This is how the new financing model for so-called neoclouds — cloud providers built purely for AI — works: a multi-year customer contract from a company like Anthropic functions as collateral. With a six-year revenue stream on paper, Nscale can borrow, build, and hire against future revenue long before the capacity exists. The IPO is the mechanism that makes closing the remaining gap possible: it converts 2027 and 2028 obligations into capital today.

That is why the $45 billion figure is simultaneously Nscale's greatest resource and its greatest constraint. The contract is, according to the reporting, a commitment from Anthropic — not cash in the account. And Anthropic itself, per the rationale in its own coverage, is building out compute because it expects growth in products like AI coding assistance and agent services. The demand underwriting the entire structure is an expectation of future use — not historical, booked revenue.

The scale of this capital need is not Nscale's problem alone. S&P Global Ratings estimates that AI infrastructure investment will exceed $1.3 trillion by 2027, and the question the agency itself raises is how all of it will be financed. Nscale is the test case: a company not yet public, taking on billions in obligations on the strength of long-dated contracts, and heading to market to pay the bill.

The Anyscale acquisition, as SiliconANGLE describes it, extended the software portfolio — an understandable hedge against ending up as a pure commodity supplier of megawatts, and also one more line against the cash box ahead of the listing.

What Could Break the Model

The counterforce is easy to describe. The structure only works if AI demand keeps compounding through the full contract term. If the demand curve bends before 2028, Nscale is left holding gigawatts under construction and contracted payments that no longer correspond to what the market actually needs. Companies like Anthropic are customers with capital pressure of their own — and coverage points to Anthropic itself weighing an IPO, partly to finance exactly this kind of commitment.

One more caveat belongs in the reader's assessment: most of the key figures here are press reports, not public registrations. The $3 billion IPO is "according to people familiar with the matter." The $45 billion is "reported." None of the numbers are confirmed in primary sources such as prospectuses or public filings. That does not make them untrue — but it means the market has not yet priced them, and every deal in this industry can be renegotiated before the first gigawatt is delivered.

Three Signals to Watch

The earliest honest indicators of whether the loop closes are concrete. First: the price of the IPO. If Nscale secures the full $3 billion at a valuation the market will bear, the capital markets have accepted the premise. If the company is valued below what its commitments imply, that is the first market-based doubt.

Next: delivery at Monarch. Phase one is targeted for early 2028, per Nscale. A slipping timeline is the clearest sign that the construction method, the power supply, or the capital base is tightening.

And finally: whether Anthropic's commitment holds in its original form. Any mention of renegotiated volume, shifted start dates, or reallocated capacity will say more about the model's durability than any announcement ever will.

If everything goes right, the consequence is real: compute capacity decouples from hyperscaler balance sheets, and new geographies — West Virginia's microgrid, Arctic hydropower in Glomfjord — become production sites for AI. If it slips, the consequence is just as concrete: a young, unprofitable infrastructure company with six years of sales on paper and a construction project that has to deliver. Today Nscale stands with 30 MW in operation, 1.35 GW with a date attached, and a contract that presupposes both — and much more.