Nscale Adds Ray Platform With Anyscale Deal At About $1.65bn

AI cloud competition is moving from renting GPU capacity to controlling the software that schedules work across it.

The Next Web reported that Nscale has agreed to acquire Anyscale, the commercial company behind Ray, with Bloomberg putting the deal value at about $1.65bn.

The acquisition would bring Anyscale's roughly 200 employees into Nscale and is expected to close in the second half of 2026.

Nscale did not publish a price for the transaction, so the valuation remains tied to Bloomberg's sourcing rather than the company's announcement.

Anyscale Gives Nscale A Software Layer Above GPUs

Nscale already presents itself as an AI-cloud operator with power, data-centre, rack and compute assets.

Anyscale changes the operating layer above that hardware because Ray is used to distribute data preparation, model training, fine-tuning and inference across large fleets of chips.

The product argument extends beyond more capacity.

Anyscale's commercial platform sits around the open-source Ray framework, while Ray itself remains with the PyTorch Foundation.

The Next Web put Ray's download count at about 237 million after the foundation took over the project in 2025, keeping the open-source system available outside Nscale.

The commercial assets being acquired include the platform, engineering team and customers named for the company: Coinbase, Runway and Bedrock Robotics.

That gives Nscale customer and workflow evidence for software orchestration, rather than only a plan to rent accelerators by the hour.

Capacity Plans Still Arrive On Different Timelines

The timing contrast is central to the deal.

Anyscale sells software against infrastructure customers can use now, while Nscale's largest disclosed deployment is a 1.35GW Microsoft site in West Virginia that starts arriving in 2027.

The transaction therefore puts software revenue and customer relationships in front of a major capacity ramp that is still dated for the future.

Nscale's buildout is capital-heavy.

The source record says the company has passed 1GW of capacity, closed a $900m revolving credit facility in July and has a $2.5bn commitment for UK data centres.

The company also plans to float, possibly later this year, while its board includes former Meta executives Sheryl Sandberg and Nick Clegg.

Cost Claims Need Customer-Level Proof

The strongest performance figure remains vendor-owned.

Anyscale claims its platform can reduce total cost of ownership by up to 90% compared with using separate tools, but the article did not include an independent benchmark methodology for that number.

Competitors are also buying software around the AI infrastructure stack.

Nebius spent $643m in May on Eigen AI, and Qualcomm bought compiler startup Modular this month.

Those transactions put Nscale's acquisition inside a wider push by AI infrastructure companies to own more of the scheduling, compiler and orchestration software around installed chips.

Anyscale revenue, customer contract values and a Microsoft delivery schedule after the 2027 start window remain outside the transaction record.