The British company Nscale has finalized an agreement to acquire Anyscale, the creator of the commercial platform based on Ray, with the aim of offering an AI infrastructure that spans from data centers and GPUs to training, inference, and deployment of agents. Nscale has not disclosed the price, although financial media estimate the deal at around $1.65 billion.
The key points of the Anyscale acquisition in 30 seconds
- Nscale will incorporate Anyscale’s platform to run distributed AI workloads across thousands of GPUs.
- The official price has not been published, but the deal is valued at approximately $1.65 billion.
- Anyscale will retain its brand, its customers, and the ability to work with infrastructures from various providers.
- Its approximately 200 employees will join Nscale.
- Ray will continue as an open project under the PyTorch Foundation.
The acquisition is still subject to regulatory approvals and usual closing conditions. Both companies expect to complete it during the second half of 2026. Until then, they will continue operating as independent organizations.
This move reflects a clear trend in the AI infrastructure market. Renting servers with GPUs is no longer enough to differentiate. Providers now aim to control also the software that distributes work, prepares data, trains models, and maintains applications in production.
Nscale Aims to Go Beyond Selling GPU Capacity
Nscale positions itself as a vertically integrated AI cloud platform. Its business combines data centers, power access, accelerator systems, and services for training and inference workloads.
The acquisition of Anyscale adds a software layer that sits atop this physical infrastructure. Its platform enables distributing Python applications and AI workloads across different machines, clusters, and GPUs without requiring teams to manage each resource individually.
This combination aims to offer more than just renting compute hours. The company wants to provide a complete environment where a client can process data, train or fine-tune models, deploy for inference, and run AI agents—all on the same platform.
This distinction is significant in a market where multiple providers can purchase the same Nvidia accelerators. Hardware remains scarce and expensive, but having GPU access doesn’t guarantee efficient usage. Software is needed to allocate tasks, recover resources after failures, scale applications, and coordinate thousands of processes.
Nscale was valued at $14.6 billion after announcing a $2 billion Series C round in March 2026. The company has also used various financing methods to deploy clusters in Europe and expand internationally.
Integrating Anyscale increases the complexity of this expansion. Nscale will need to develop data centers, secure power and chips, manage large contracts, and simultaneously incorporate a software company born in the US with its own technical community.
What Anyscale Brings and Why Ray Is So Valuable
Anyscale was founded by several creators of Ray, an open framework for deploying Python applications and AI workloads.
Ray enables a program to transition from running on a single machine to distributing across hundreds or thousands of processors without developers having to build distributed computing logic from scratch. On top of Ray, processes like data prep, training, reinforcement learning, inference, and simulation can be executed.
Anyscale’s commercial platform adds managed services, management tools, observability, and automation around Ray. Its customers include companies such as Coinbase, Runway, and Bedrock Robotics, according to the company.<
This gives Nscale access to a strategically important technology layer. Anyscale doesn’t manufacture chips or build data centers, but its software decides how to utilize available resources and how to scale applications across them.
The purchase follows a vertical integration strategy. Nscale provides the physical infrastructure, while Anyscale helps transform it into a platform that development teams can use without directly managing each server.
For clients, the promise is simplification of projects that today often depend on multiple vendors: one for data storage and prep, another to lease GPUs, various open-source tools, and an internal layer for maintaining models in production.
Anyscale Will Keep Its Brand and Support Multi-Cloud Use
Nscale has indicated that Anyscale will continue to operate under its own brand and serve its current customers. It will also retain the ability to run workloads across infrastructure from different providers.
This is significant because part of Anyscale’s value lies in its software not being tied to a single cloud provider. A company can use Ray and the Anyscale platform on multiple environments, including its own data centers.
Over time, these customers will have an additional option: running Anyscale’s layer on Nscale’s infrastructure. The acquiring company expects this integration to facilitate joint sales of physical capacity and managed services.
The approximately 200 employees of Anyscale in the US, Europe, and India will join Nscale upon deal closing. Both companies have not announced layoffs or immediate leadership changes.
Anyscale will also remain as an independent product, at least during the initial stage. Nscale seems interested in maintaining its identity, customer base, and relationship with the developer community.
Ray Will Continue As an Open Project Under PyTorch Foundation
A key question concerns Ray’s future.
Anyscale moved the project to the PyTorch Foundation in 2025. This organization is part of the Linux Foundation and hosts other AI infrastructure projects like PyTorch, vLLM, and DeepSpeed.
This means Nscale does not gain exclusive control over Ray’s open source code. The project remains community-governed and can be used by developers and companies in other infrastructures.
Nscale has announced that it will join the PyTorch Foundation and support the project’s continuity. This reduces the risk that Ray will immediately become a closed or solely data center-oriented technology under the new ownership.
However, open licensing does not eliminate all doubts. The new owner employing many of Ray’s top developers can influence technical priorities, optimizations, and development pace.
Nscale has a clear incentive for Ray to perform well on its own infrastructure. Such collaboration can improve performance for its customers, but the community must ensure portability across providers remains a true priority.
The $1.65 Billion Price Is Not Officially Confirmed by Companies
Nscale’s statement does not include financial details. The $1.65 billion figure comes from Bloomberg reports and other media sources citing an informed person familiar with the deal. Nscale declined to publicly disclose the amount.
Therefore, the price should be viewed as an estimate based on near sources, not a confirmed figure from either the buyer or the seller.
Anyscale had raised over $60 million after its Series B in 2020 and later expanded its funding rounds. Its valuation depends not only on revenue but also on its team, customers, and position regarding Ray.
For Nscale, paying such a substantial amount implies betting on its software layer’s ability to increase revenue per GPU. Instead of competing solely on price, availability, or data center locations, it aims to offer a comprehensive platform and charge for higher-value services.
The outcome will depend on how well it can integrate both businesses without compromising Ray’s adoption-friendly neutrality. It will also need to demonstrate sufficient enterprise demand to justify investments in capital-intensive infrastructure combined with ongoing software development.
This acquisition illustrates how competition among the so-called neoclouds is evolving. The next battle isn’t just about installing more GPUs but about controlling the software that determines their use.
Frequently Asked Questions
How much will Nscale pay for Anyscale?
Nscale has not disclosed the financial terms. Bloomberg and other outlets estimate the price around $1.65 billion, but this has not been officially confirmed.
What is Anyscale?
Anyscale develops a platform for running and scaling distributed AI workloads. It was created by the developers of Ray, one of the most widely used open frameworks for scaling Python and AI applications.
Will Ray become exclusively owned by Nscale?
No. Ray is an open project hosted by the PyTorch Foundation. The acquisition affects Anyscale and its commercial platform, but the open-source Ray framework remains community-governed and available for use in other infrastructures.
When will the acquisition be finalized?
Both companies expect to close the deal during the second half of 2026, once regulatory approvals are obtained and other closing conditions are met.
