By Stuart Kerr, Technology Correspondent, LiveAIWire

The Nscale Anyscale acquisition landed on Thursday at a reported 1.65 billion dollars, and it captures a shift in the AI infrastructure business that has been building for months: owning GPUs is no longer enough. Nscale, the London-based “neocloud” that generates its own power and builds its own data centres, has agreed to buy Anyscale, the company behind the widely used open-source Ray framework, according to Bloomberg reporting cited by TechCrunch and The Next Web. Neither company disclosed a price. Bloomberg’s figure, from a source familiar with the deal, is the only number in public circulation.

Anyscale’s roughly 200 staff, spread across the US, Europe and India, will join Nscale when the deal closes in the second half of 2026. The company will keep operating under its own brand and continue serving existing customers, which include Coinbase, Runway and Bedrock Robotics.

Why the Nscale Anyscale Acquisition Is About Software, Not Chips

Nscale’s pitch since its founding has been that it owns every physical layer beneath the chip: the electricity generation, the data centre buildings, the racks. The Nscale Anyscale acquisition moves the company up into the layer that decides how much useful work each expensive GPU actually does. Anyscale’s platform, built on Ray, spreads data preparation, training, fine-tuning and inference across an entire fleet of machines, and the company says that cuts total cost of ownership by as much as 90 percent compared with stitching together separate tools.

“Most infrastructure providers just buy GPUs and rent them,” Nscale chief executive Josh Payne told reporters, describing his company’s approach as building and owning every layer itself instead. Nscale’s product chief, Dan Bathurst, put the commercial logic to Bloomberg more bluntly: renting raw GPU capacity by the hour is a commodity business fought against every rival buying the same Nvidia hardware, and the money increasingly sits a layer up, in the software that decides how a job gets split across thousands of chips at once.

The Open-Source Project at the Centre of the Deal Is Not for Sale

Ray itself is not part of the transaction. Governance of the open-source framework passed to the PyTorch Foundation in 2025, and Nscale is committing to join that foundation as part of the acquisition rather than taking control of the project. Ray has logged roughly 237 million downloads and runs inside tools as varied as Cursor and xAI’s own infrastructure, and it will remain free to run anywhere regardless of who owns Anyscale’s commercial platform. What Nscale is actually buying is the commercial layer built on top: the paid managed service, the engineering team, and Anyscale’s existing customer base.

Anyscale arrives with real momentum behind it. The company reported 70 percent revenue growth in its most recent quarter, and was valued at 1.38 billion dollars in a 2022 funding round, making the reported 1.65 billion dollar price a solid return for its earlier investors. Anyscale’s own statement framed the logic as mutual: pairing its software with Nscale’s infrastructure lets the two “co-design the software layer and infrastructure beneath it,” something the company said neither side could do as effectively working alone.

A Land Grab Playing Out Across the Whole Neocloud Sector

Nscale is one of a wave of AI-focused “neocloud” providers that have emerged to compete with the traditional hyperscalers, and the Anyscale purchase is not an isolated move. Nebius spent 643 million dollars in May acquiring Eigen AI, a similar workload-efficiency company, and chipmaker Qualcomm bought the compiler startup Modular this month for the same underlying reason. “Everyone’s wanting to move up the stack,” Bathurst told Bloomberg, describing a scramble among infrastructure providers to become the single vendor a customer turns to for training, fine-tuning and inference rather than one link in a longer chain.

Nscale itself has moved fast since emerging from a cryptocurrency-mining business in early 2024. It raised 2 billion dollars in a March 2026 Series C round that valued the company at 14.6 billion dollars, with investors including Nvidia, Nokia, Blue Owl, Dell and the Norwegian industrial group Aker. Sheryl Sandberg and Nick Clegg, both former Meta executives, sit on Nscale’s board. The company has since closed a 900 million dollar revolving credit facility, committed 2.5 billion dollars to UK data centres, and is reportedly considering a public listing later this year.

The Access Question Sitting Underneath the Deal

The Nscale Anyscale acquisition fits a pattern LiveAIWire has tracked closely across AI infrastructure more broadly: control over compute is increasingly concentrated among a small number of well-capitalised players, whether that control comes through direct ownership or through financing structures. LiveAIWire’s coverage of Nvidia’s revenue-sharing model for AI startups found the same underlying dynamic from a different angle, with Nvidia extending its dominance in hardware into a recurring claim on the usage economics of the compute it enables, rather than simply selling chips outright.

LiveAIWire’s reporting on SpaceX’s Colossus commercial compute platform documented a parallel dynamic, in which a handful of companies with the capital and physical infrastructure scale to build frontier AI compute are positioning themselves as gatekeepers for the rest of the industry. The Nscale Anyscale acquisition is the same story told through vertical integration rather than platform access: whoever controls the software layer that decides how efficiently a GPU fleet runs captures value that pure hardware ownership alone no longer guarantees.

Why the Software Layer Nscale Is Buying Connects to How Models Actually Get Trained

LiveAIWire’s earlier coverage of the funding boom around RL environments used to train AI agents found that reinforcement learning infrastructure has become one of the most contested layers in AI, precisely because a well-built training stack lets a company train models directly on its own workflows rather than renting a general-purpose system. Ray sits underneath a meaningful share of that infrastructure already, since Anyscale’s platform is built specifically to orchestrate the kind of large-scale, multi-step training runs that reinforcement learning environments depend on. Owning the commercial layer on top of that orchestration gives Nscale a direct stake in exactly the training workloads that RL environment vendors are racing to serve.

What This Means for Anyone Buying AI Compute

For an AI startup or enterprise deciding where to run training and inference workloads, the practical question the Nscale Anyscale acquisition raises is whether a single vertically integrated vendor, offering power, data centres, GPUs and orchestration software all from one company, is a better bet than assembling a stack from specialists. Nscale’s argument is that co-designing the software and the physical infrastructure together produces efficiency gains neither layer could achieve alone.

The counterargument, implicit in why Ray remained open source even as its commercial arm changed hands, is that customers value the ability to run the same orchestration layer on infrastructure from any vendor, and lock-in risk is exactly what a fully vertically integrated provider is asking customers to accept in exchange for that efficiency.

Whether that trade-off pays off is partly a timing question. Anyscale sells against compute capacity that exists today, while Nscale’s largest single deployment, a 1.35 gigawatt site being built for Microsoft in West Virginia, does not begin coming online until 2027. The deal bets that demand for integrated, full-stack AI compute will still be growing by the time that capacity arrives, a bet every major neocloud provider making similar acquisitions this year is currently making in parallel.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.