Back
AI News

Anthropic Committed $45 Billion to Nscale. The Data Center Doesn't Exist Yet.

A little-known data center operator with Norwegian hydroelectric campuses has reportedly secured a six-year, $45 billion contract from Anthropic — for computing power that, in large part, has not been built.

AIMag.no
AIMag.no
August 31, 2026 · 4 min
Illustration: A vast empty concrete foundation slab in a wet Norwegian valley, marked with survey stakes, a single power cable running out of frame toward distant hydroelectric turbines under an overcast sky.

$45 billion over six years. That is the amount Anthropic has reportedly committed, per Bloomberg reporting, to pay to rent AI computing power from Nscale — a company almost nobody had heard of a year ago. And most of that computing power does not exist yet. Phase one of the Monarch campus in West Virginia, engineered for 1.35 gigawatts, targets delivery in early 2028, according to Nscale's own materials.

As long as the contract is reported rather than confirmed by either company, the figure should be read with caution. But the outline is clear: an operator with a modest track record has become a counterparty in one of the largest capacity deals the AI industry has seen.

From Fjord to Gigawatt

Nscale operates and builds AI data centers across twelve locations: Oslo, Stavanger, Narvik, and Glomfjord in Norway; Hayes and Loughton in the UK; Keflavík and Blönduós in Iceland; Sines in Portugal; and West Virginia, North Carolina, and Texas in the US. The company markets itself on renewable power — Arctic hydro in Glomfjord, where 30 MW is now operational and expandable to 60 MW — and on a modular, pre-fabricated, chip-agnostic architecture that it claims enables faster capacity buildout.

So the party behind the $45 billion contract is not a hyperscaler empire. It is a company with one operating campus in the double-digit megawatts and a portfolio of projects still on the drawing board, including Monarch in West Virginia — the first state-certified AI microgrid in the US, with a stated expansion path to more than 8 GW.

The Loop That Finances the Build

Here lies the mechanism of the story. Anthropic reportedly has its motivation spelled out: the company is expanding computing power aggressively to meet the demand it expects for products like AI coding. For Nscale, the six-year commitment functions as revenue credibility — a document it can take to lenders and the equity market. According to Yahoo Finance reporting, the company is now working toward raising as much as $3 billion in a US IPO.

Last month, Nscale also bought the software company Anyscale for $1.65 billion — a deal SiliconAngle describes as an expansion of its software portfolio, a step up the stack from raw GPU rental toward better margins.

The pattern is plain: the lab commits to rent. The operator commits to build. The investors commit to fund. Each link in the chain is betting on the others' solvency — and each link has a deadline.

The math is demanding: an offering of up to $3 billion is meant to help finance delivery obligations reported at $45 billion, plus a $1.65 billion acquisition. S&P Global Ratings estimates that AI infrastructure investment will exceed $1.3 trillion by 2027 — Nscale is a test of whether a second-tier operator can raise capital at this scale.

When the Gap Between Signature and Concrete Must Be Financed

The consequences reach beyond one company's balance sheet. Labs are now locking up capacity years before it is delivered. Operators are monetizing commitments before the ground is dug. And politics comes along for the ride: West Virginia Governor Patrick Morrisey has publicly praised the reported $45 billion investment in his state — a bet whose jobs and power consumption arrive on a multi-year lag.

The counterforce deserves to be stated at full strength. Every key figure is reported, not confirmed. Capacity deals can be renegotiated. Anthropic is stretching toward its own reported IPO run while committing $45 billion in future rent — obligations that now exceed what the company earns. If demand for model training and inference flattens, the loop reverses direction, and the newest, most heavily indebted operator absorbs the shock first. It is worth remembering that operational capacity and engineered capacity are two different things: Glomfjord's 30 MW delivers now. Monarch's 1.35 GW does not. Not yet.

Everything Comes Due in 2028

Early 2028 is the date on which every party's story converges. Anthropic's demand, Nscale's valuation, the software margin thesis of the Anyscale acquisition, and West Virginia's microgrid bet all fall due at the same switch: Monarch phase one delivers, or it does not.

Until then, the honest state of the AI infrastructure boom is not an answer. It is a deadline.

AIMag.no
AIMag.no
The AIMag.no editorial team covers artificial intelligence, tools, research, and regulation.