Nscale Anyscale acquisition values AI software startup at reported $1.65B
Nscale agreed to buy Anyscale, with Bloomberg reporting a $1.65B price, adding Ray cluster software to its AI data center stack.
By Wei-Lin Zhao · AI Correspondent
· 3 min read
Nscale’s Anyscale acquisition would bring a venture-backed AI infrastructure software company into a data center builder that is trying to sell more than raw compute. Nscale Global Holdings Ltd. announced the planned purchase of Anyscale Inc.; deal terms were not disclosed, but Bloomberg reported, citing a person familiar with the transaction, that the price is $1.65 billion.
The transaction comes months after Nscale raised $2 billion in March from a group of backers that included Nvidia Corp. and other large technology companies. Nscale did not disclose Anyscale’s revenue, headcount, customer count or ARR, leaving the reported price hard to benchmark against operating metrics.
London-based Nscale is building a network of AI data centers across six countries. Its largest disclosed project is a 2,250-acre site in Mason County, West Virginia, which includes its own small power grid. Nscale has said the campus could theoretically support more than eight gigawatts of computing capacity, a figure that signals the scale of its ambitions but does not indicate how much capacity is already live.
Why is Nscale buying Anyscale?
Nscale is adding software that helps AI teams run distributed workloads across clusters, a problem that becomes more expensive as GPU fleets get larger. The company said it plans to offer Anyscale’s products alongside its own infrastructure optimization tools, including managed versions of Kubernetes and Slurm.
Anyscale was founded in 2019 by computer scientists who created Ray, an open-source framework used to run and optimize large AI workloads across many servers. The company sells a managed cloud version of Ray, giving developers access to clusters that Anyscale says can be started in less than a minute, along with dashboards for monitoring and debugging.
Ray addresses a practical issue in AI infrastructure: spreading a training or inference workload across many machines creates reliability, utilization and networking problems. If a server fails during a training job, Ray can replace it with another machine, reducing the need for teams to build their own fault-tolerance systems. It can also place models and datasets on the same server to cut unnecessary network traffic, according to Anyscale’s product positioning.
What does this signal for AI infrastructure?
The deal points to more vertical integration among companies selling AI capacity. Nscale already presents itself as controlling the physical inputs of AI compute, including power, data centers and hardware. By adding Anyscale, it is trying to move higher in the stack, where customers may care less about which building houses the GPUs and more about whether large jobs run predictably.
Nscale founder and Chief Executive Josh Payne said in the company’s announcement that Anyscale adds managed services used by AI teams to scale workloads, supporting Nscale’s pitch of a vertically integrated AI cloud platform. That is a company claim, and the announcement did not say how Anyscale will be integrated operationally or whether its existing cloud service, pricing or roadmap will change.
Nscale said it expects the acquisition to close by the end of the year. Until then, the main confirmed facts are the strategic direction and the reported price: Nscale wants software control over the AI clusters it is building, and Bloomberg’s reporting puts that bet at $1.65 billion.
This story draws on original reporting from SiliconANGLE.
