Today, Anyscale announced a partnership with Microsoft that will deliver a new fully managed first-party AI computing service on Azure, powered by Anyscale's open-source Ray framework. More precisely, the companies are introducing a first-party Azure service that brings Ray's distributed computing capabilities directly to Microsoft's cloud platform. Co-developed by Anyscale and Microsoft, the service is designed to streamline how enterprises build, run, and scale AI workloads by reducing the operational overhead of managing distributed systems.

The service enters private preview today, with general availability expected in 2026. Azure customers will be able to set up and manage the service directly from the Azure Portal and run their workloads on Azure Kubernetes Service (AKS). Moreover, to simplify operational costs, enterprises will be able to leverage unified billing through Azure.

About Ray

Ray is an open-source distributed computing framework designed for AI workloads that demand coordination across thousands of GPUs and CPUs. It handles everything from data processing to model training and serving, and is particularly suited for multimodal processing that mixes different accelerators in the same pipeline.

Created by the team that founded Anyscale, Ray replaces layers of custom orchestration with a unified framework that can efficiently scale workload processing from one to thousands of machines. Ray was built to handle modern AI's unique demands, where, in addition to scaling in size, workloads often combine CPUs and specialized accelerators, like GPUs, within the same pipeline. These are scenarios that traditional computing architectures struggle to manage efficiently, and that eventually fall victim to hurdles that slow down innovation.

In a recent interview with Data Phoenix, Anyscale CEO Keerti Melkote described Ray as the unit of scale for the current computing evolution, which is taking us from cloud computing to AI: "When we went from internet to cloud computing, we went from virtual machines to containers and microservices, and scaled that way. Similarly, for AI, Ray is that next unit of scale. Basically, we take a container, and run microtasks and services and microactors that allow it to scale up AI workloads very, very quickly. They are almost like rebuilding a supercomputer in a disaggregated way."

The Ray framework has over 27 million monthly downloads and 39,000 GitHub stars, with existing deployments at Uber, Spotify, Canva, and Coinbase.

Key Features

Developer Experience: Teams can deploy clusters directly from Azure Portal, use cluster-backed IDEs for interactive development, and access advanced debugging dashboards.

Performance: Compared to self-managed Ray, the Anyscale Runtime performs 10x more efficiently and does not require code changes.

Security and data governance: The data, compute and models always remain within customer Azure accounts, ensuring customers are always in control of their intellectual property. As a first-party Microsoft product, the service maintains full compliance with Azure security policies.

Production readiness: Fully managed, fault-tolerant clusters eliminate the operational complexity of managing distributed systems, supporting both batch processing and low-latency serving workloads.

Industry Perspective

"AI is redefining every industry, but scaling it remains one of the hardest challenges," said Keerti Melkote, CEO of Anyscale, in the official press release. "Together with Microsoft, we're making it dramatically easier for enterprises to build and run AI at scale."

Keerti doubled down on this sentiment during the interview: "As AI makes its way from Silicon Valley AI-native startups to digital-native companies to mainstream enterprises, they are looking for a partner that can help them build their own AI. They want to build their own core intellectual property using their data within their security domain. And for that, a tool like Anyscale and Ray becomes extremely important. And so, we are super excited, because we now can take Microsoft's power, the ability for them to take this to market and take Anyscale to their channels. It's a signature partnership for us, and we are really looking forward to it."

Brendan Burns, Corporate Vice President of Cloud-Native Compute at Microsoft and co-creator of Kubernetes, emphasized the value of bringing Ray to AKS: "We're giving developers the ability to use familiar cloud-native patterns to build and scale AI applications."

Coming soon: Our interview with Anyscale CEO Keerti Melkote

At Ray Summit 2025, Data Phoenix founder Dmytro Spodarets had an opportunity to sit down with Keerti Melkote, CEO of Anyscale, and discuss a variety of topics, from Ray's role in the current computing landscape, and important announcements, including Anyscale contributing Ray to the PyTorch Foundation and Anyscale's partnership with Microsoft, to the lessons learned while serving some of the biggest names in the industry, and much more. Check back soon to catch the full interview, which will be available as part of the Voices of AI series here at Data Phoenix.