Anyscale

Distributed Python for AI workloads

Anyscale builds a managed platform around Ray. Ray's task and actor abstractions distribute Python computation across a cluster; higher-level libraries address data processing, training, tuning, reinforcement learning and model serving. Anyscale supplies the operational service around those building blocks, rather than a foundation model trained by the company. [1][2]

Ray 2.0's August 2022 release expanded that AI workflow with an integrated runtime then called Ray AIR, alongside serving and data changes. AIR was a beta integration layer, and the announcement now notes that its namespace was later retired. The historical release should not be read as the name of today's Ray product line. [3]

Company and community

The company's July 2026 Nscale announcement describes an acquisition agreement subject to closing conditions. It also states that Ray had moved to the PyTorch Foundation in 2025 and would remain community-governed. A commercial platform transaction does not transfer ownership of every project that platform uses; the agreement alone also does not establish a completed acquisition of Anyscale. [4]