Nscale eyes a US IPO in September on a $51B order backlog

Nscale, the London company building data centres and GPU capacity for AI, is preparing for a US stock market listing that could come as soon as September. In briefings to potential backers, reported by Bloomberg, it put its total contracted revenue at around $51B and pointed to fast growth: revenue of more than $100M in the second quarter of 2026, up from roughly $37M in the first quarter and about $33M in all of 2025. Goldman Sachs and JPMorgan are advising on the potential float.

A British AI-infrastructure company heading for a US IPO at this scale is a notable moment for the UK’s compute ambitions. Nscale is one of a handful of domestic firms racing to supply the raw computing power that training and running AI depends on. It is adding some 10 gigawatts of power for its facilities, on top of the 831 megawatts already active or contracted, and reported roughly 25,000 active chips, with close to 289,000 active and contracted, as of the second quarter.

The $51B number deserves a careful read, though, and this is where our house scepticism kicks in. It is a contracted backlog, not booked revenue: a single multi-year compute agreement is counted as billions in future revenue the moment it is signed, long before a chip is switched on. Much of that money is years away and depends on Nscale actually building the capacity to deliver it. Impressive momentum, yes, but the figure measures promises more than performance.

It also sits inside a bigger UK story we keep returning to. Alongside FluidStack and Ori, Nscale is part of a cluster of British firms trying to keep more of the compute that AI needs, for both training and inference, built and owned at home. That ambition has drawn government backing and a large share of the country’s venture funding. A successful listing would be the clearest test yet of whether investors believe the neocloud model can pay off.