Nscale, the British AI neocloud that raised $2 billion at a $14.6 billion valuation in March from investors including Nvidia, Nokia and Dell, has signed a definitive agreement to acquire Anyscale — the company built around Ray, the distributed-compute framework that sits under a large share of production AI training and serving.

What was announced

Around 200 Anyscale staff across the US, Europe and India move to Nscale. Anyscale retains its brand and its customers, named in the release as Coinbase, Bedrock Robotics and Runway. Nscale is joining the PyTorch Foundation, and the release is careful to note that Ray "was donated to the PyTorch Foundation and remains open source and community governed". Anyscale's revenue is described as up 70% quarter on quarter.

The number nobody officially gave

The press release says exactly this: "Financial terms of the transaction were not disclosed." $1.65 billion is TechCrunch's reporting, not a company figure, and it is already being written as though the parties announced it. The comparison being drawn against Anyscale's $1.38 billion Series C mark is also weaker than it looks — that valuation is from 2022, four years and an entirely different rate environment ago, so framing it as a modest step-up misreads the interval.

Signed is not closed

This is a definitive agreement pending regulatory clearance, with closing expected in the second half of 2026. Expect "Nscale has acquired" phrasing that drops the condition.

Why a landlord buys the software layer

Neoclouds rent capacity, and capacity is the least defensible product in the stack — the hardware is the same hardware everyone else is buying, and the margin compresses toward whoever has the cheapest power. Buying the orchestration layer is an admission that renting bare GPUs is not a business you can hold. It also raises a genuine governance question about a widely used open framework passing into an infrastructure provider's hands, which is presumably why the foundation commitment is in the first paragraph of the release rather than the last.