Nscale Buys Anyscale for $1.65 Billion in a Bid to Own the Full AI Compute Stack

Nscale Buys Anyscale for $1.65 Billion in a Bid to Own the Full AI Compute Stack

British AI cloud upstart Nscale said on July 30 that it has agreed to acquire Anyscale, the San Francisco software company built by the creators of the open-source Ray framework, in a deal that stitches the software layer of modern AI directly onto Nscale's power plants, data centers and GPUs. The price, which neither company disclosed, is about $1.65 billion, according to Bloomberg and Reuters, both citing people familiar with the transaction. It is the boldest move yet by a fast-rising "neocloud" trying to prove it can compete with the hyperscalers not just on cheap compute, but on the entire ladder of tools above it.

The logic is vertical integration. Nscale sells the raw ingredients of AI — electricity, buildings, chips and the low-level software that binds them — while Anyscale sells the layer that machine-learning engineers actually touch: orchestration, observability and the plumbing that lets a training or inference job fan out across thousands of GPUs. Buying Anyscale lets Nscale sell a path "from power to production AI," and capture a larger share of each customer's budget rather than renting bare metal and watching the margin flow to someone else's platform.

"Most infrastructure providers just buy GPUs and rent them. Nscale is doing something unique. We build and own every layer ourselves: the power, the data centers, the compute, and the software that turns them into an AI cloud," said Josh Payne, chief executive and founder of Nscale. "Anyscale extends that offering with managed services that AI teams use to scale any workload, completing a truly vertically integrated AI cloud platform."

The Deal

Under the definitive agreement, Anyscale will continue to operate under its own brand and keep serving existing customers, who remain free to run their workloads on any infrastructure they choose. Over time, those customers will gain the option of running the Anyscale software layer on Nscale's stack. Anyscale's entire team — roughly 200 people spread across the United States, Europe and India — will join Nscale. The companies said the transaction is subject to closing conditions and regulatory approvals and is expected to close in the second half of 2026.

Anyscale grew out of Project Ray, the distributed-computing framework created by researchers at UC Berkeley's RISELab and released as open source. After the debut of GPT-3 pulled large language models into the mainstream, Anyscale pivoted from generic distributed workloads toward AI-specific scaling: training, fine-tuning, inference, data curation and reinforcement learning, all orchestrated through Ray. In 2025 the company donated Ray to the PyTorch Foundation, where it remains community-governed; Nscale said it will join the foundation to signal its commitment to the project and the tens of thousands of developers who depend on it. Anyscale's platform today powers AI at companies including Coinbase, Bedrock Robotics and the video-generation firm Runway.

The financials underscore why Nscale was willing to pay a rich premium. Anyscale was last valued at $1.38 billion in a 2022 Series C round, meaning the reported $1.65 billion price is only a modest markup on a four-year-old valuation — a sign of how uneven the AI software market has been even as demand exploded. The company told TechCrunch that its revenue rose 70% in its most recent quarter compared with the prior one, evidence that the pivot to AI workloads is finally converting Ray's enormous developer base into paying business. Goldman Sachs International served as lead financial adviser to Nscale; Qatalyst Partners advised Anyscale.

"Companies are moving beyond simply using AI to actually building their own. Doing that well requires the software and the infrastructure it runs on to be designed together," said Keerti Melkote, chief executive of Anyscale. "By combining Anyscale's platform, built on Ray — the open standard for scaling AI, with Nscale's data centres, compute and AI cloud services — we're creating the first full-stack AI hyperscaler."

Why It Matters

The acquisition is a wager on where value accrues in the AI economy. For two years, the safest business in AI has been selling GPUs, and a wave of neoclouds — Nscale among them — raised billions to build capacity that Nvidia could not supply fast enough. But renting chips is a commoditizing business with thin, cyclical margins, and every neocloud now faces the same question: what keeps a customer from decamping to the cheapest GPU on the market next quarter? The answer, increasingly, is software. Owning the orchestration layer makes a customer's workloads sticky and lets a provider price the whole stack rather than one interchangeable input.

Nscale has the balance sheet to place that bet. In March the company raised $2 billion in a Series C that valued it at $14.6 billion, drawing in Nvidia, Nokia, Blue Owl, Dell and Norway's Aker, and it has layered on debt raises and compute deals with Microsoft, British Telecom and others. The Anyscale purchase mirrors a broader convergence across the industry, where infrastructure players are absorbing software and software players are chasing compute, all reaching toward the same "full-stack" position the hyperscalers occupy by default. If Nscale can genuinely co-design its software and hardware — optimizing them together in ways a standalone vendor cannot — it could deliver the cost and performance edge Melkote described. The risk is integration: bolting a beloved open-source-adjacent software company onto a power-and-metal operator is a cultural and technical challenge that has humbled larger acquirers.

What to Watch

The immediate test is regulatory. A cross-border deal of this size, closing in the back half of 2026, will draw scrutiny in both the U.K. and the U.S., and any conditions could reshape the timeline. Watch, too, how Nscale handles Ray's open-source community and its stewardship inside the PyTorch Foundation — mishandling that constituency would blunt the very asset it just bought. Finally, keep an eye on Anyscale's existing customers, from Coinbase to Runway, and whether the promise that they can keep running anywhere holds up once Nscale has an incentive to steer them onto its own GPUs. If it does, a neocloud will have shown it can grow into something closer to a hyperscaler; if not, the deal will read as an expensive lesson in the limits of vertical integration.