Nscale to Acquire Anyscale, Maker of the Open-Source Ray Framework

Nscale, a GPU cloud infrastructure provider, has agreed to acquire Anyscale, the AI infrastructure startup founded by the creators of Ray, the open-source framework many companies use to run distributed AI workloads. The two companies announced the deal on July 30, 2026, without disclosing terms; Bloomberg and other outlets have reported the price at roughly $1.65 billion, a figure neither company has confirmed.

What each company brings

Nscale supplies the physical layer of AI computing — GPUs, data centers and power contracts — that has made it one of a new class of infrastructure providers known as neoclouds. Anyscale adds the software layer that engineers use to schedule and scale AI compute across thousands of GPUs at once, covering data processing, model training, inference and reinforcement learning.

“We’re creating the first full-stack AI hyperscaler: one that runs any AI workload at greater scale and lower cost,” Anyscale co-founder and CEO Keerti Melkote said in the companies’ joint statement. Nscale CEO Josh Payne said Anyscale’s managed services extend Nscale’s push to turn raw GPU capacity into a complete AI cloud.

Ray stays open source

Anyscale’s roughly 200 employees across the United States, Europe and India are expected to join Nscale once the deal closes, which the companies expect in the second half of 2026 pending regulatory approval. Anyscale says it will keep operating under its own brand and serving existing customers — among them Coinbase, Bedrock Robotics and Runway — who will remain free to choose which data centers or cloud run their workloads.

Ray itself, which Anyscale donated to the PyTorch Foundation in 2025, will remain open-source; Nscale says it will also join the foundation as part of the deal.

Part of a wider consolidation

The acquisition follows a pattern among GPU cloud providers racing to move beyond selling raw compute and into the software and orchestration tools that determine how efficiently that compute gets used, as AI labs and enterprises compete for a limited supply of chips.