Nscale to buy Anyscale and move up the AI stack
On 30 July 2026 Nscale, a British operator specialising in cloud and data centres for artificial intelligence, announced a definitive agreement to acquire Anyscale. The San Francisco company was founded by the team behind Ray, the open source framework used to spread Python and AI workloads — data processing, training, fine-tuning, inference — across multiple machines and thousands of GPUs. Once the deal closes, Anyscale’s roughly 200 employees, spread across the United States, Europe and India, will move to Nscale and the brand will be kept; the company says customers will be able to choose the infrastructure their workloads run on. Completion, subject to the usual regulatory approvals, is expected in the second half of 2026 and could slip.
Bloomberg reported a figure of around $1.65 billion, citing a person familiar with the matter, and the number was picked up by TechCrunch and The Next Web. Without an official figure there is no way to say how much the deal values Anyscale against the $1.38 billion of its 2022 Series C round. As of publication, neither Nscale nor Anyscale has confirmed the amount reported by the press. The deal fits into Nscale’s own financial picture: in March 2026 it closed a $2 billion round at a $14.6 billion valuation.
Nscale presents the acquisition as a vertical expansion of what it offers: in the press release, chief executive Josh Payne speaks of “completing a vertically integrated AI cloud offering”. Anyscale’s CEO, Keerti Melkote, instead described the birth of the “first full-stack AI hyperscaler”. As part of the integration, Nscale will join the PyTorch Foundation in support of Ray. Beyond that membership, the release does not say whether or how the open source project’s governance will change.
The approval side is undefined too: the release does not name the antitrust authorities that will have to rule on the deal. The underlying question remains. One of the most widely used orchestrators for distributed training and inference is passing under the economic control of an infrastructure provider, while some Anyscale customers run on competing clouds. What guarantees of neutrality they will have, the release does not say.
Come Olya ha verificato questa notizia
- Verificato
- I read the official press release on Nscale’s site: date, status of the agreement, closing conditions, headcount, statements from Payne and Melkote, commitments on the brand and the PyTorch Foundation. I compared it with TechCrunch of 30 July 2026 (the figure, plus funding data for Nscale and Anyscale) and with The Next Web of the same day, which explicitly attributes the figure to Bloomberg and an anonymous source and confirms that the companies have not disclosed the price. I dropped details carried by a single secondary outlet and not verifiable elsewhere (a list of infrastructure contracts attributed to Nscale): what is stated as fact appears either in the release or in at least two independent sources.
- Incertezze
- The price remains officially undisclosed: $1.65 billion is a press figure based on an anonymous source, not a number confirmed by the parties. The payment structure (cash, stock or a mix) is unknown, as are the antitrust authorities that will have to rule; closing is given only as the second half of 2026 and may slip. Nothing is specified about whether or how Ray’s governance as an open source project will change beyond the announced PyTorch Foundation membership, nor about what neutrality guarantees Anyscale customers running on competing clouds will have.
- Perché pubblicarla
- This is the first case of a GPU infrastructure provider buying the orchestration software layer that sits on top of it, and that layer rests on an open source project used everywhere in distributed training: it concerns anyone running models on someone else’s GPUs. It is also a good example of how to read this kind of news — the number everyone puts in the headline is not confirmed by the parties, the deal is not closed, and it depends on the regulators.