TL;DR
Nscale, the British cloud infrastructure operator backed by Nvidia, is telling prospective investors it holds around £81bn ($103bn) in contracted revenue, according to documents seen by The Information. A stock market debut could follow within weeks. One person close to the process describes the numbers as illustrative rather than guidance.
What the figure actually represents
The deals behind it run 5.7 years on average, which works out at something near £14bn ($18bn) a year. That is a backlog, not turnover: money contracted for capacity that in large part has yet to be built, delivered or paid for. The caveat attached to it matters as much as the total. Illustrative figures shared privately ahead of a listing carry none of the obligations that attach to a prospectus.
Much of the book traces to a single counterparty. Anthropic agreed in late August to spend roughly £35bn ($45bn) renting compute at the company’s West Virginia campus, where Nvidia’s forthcoming Vera Rubin parts are due to run the workloads. Nscale did not respond to a request for comment.
Why the concentration is the story
Britain has wanted a domestic operator at this scale for years, and on these numbers it has one. The awkward part is what the numbers rest on. A backlog dominated by one AI lab is a bet on that lab’s demand holding for the better part of a decade — and Anthropic is itself spending faster than it earns, having added a £28bn ($35bn) Lambda commitment days after signing with Nscale, taking its disclosed compute obligations past £63bn ($80bn) inside a week.
Each party is underwriting the next. Nvidia backs the operator and supplies the silicon; the operator books revenue from a customer buying capacity for products still scaling; the customer’s own listing plans assume that demand compounds.
Looking forward
UK investors will get a rare direct route into AI infrastructure if this lists, and rather less clarity than the headline suggests. The questions worth asking are how much of the backlog is take-or-pay, what happens if a major tenant renegotiates, and whether capacity exists to serve it. A sceptical read of the case that AI will need far fewer GPUs than assumed is the other side of this trade.