Ray has become the engine for AI-native computing for 1000s of AI- and digital-natives. Now Anyscale and Microsoft are partnering to bring the power of Ray to every enterprise.

Every company today is racing to become an AI company. From automating code to deploying custom models that enhance every customer interaction, the ambition is universal: make AI core to the business. But realizing that vision is anything but straightforward.

The hard reality is that AI workloads have outgrown the computing architecture we’ve relied on for the last decade. What starts as a notebook experiment quickly turns into a complex web of computing challenges to support multimodal data pipelines, post-training, reinforcement learning, and distributed inference across heterogenous hardware. Managing all of that — efficiently, reliably, and at scale — has become one of the hardest engineering challenges in AI.

This is the problem Anyscale was founded to solve. And today, we’re taking the next step forward: announcing our first-party partnership with Microsoft to launch Anyscale on Azure as an AI compute service co-engineered by Anyscale and Microsoft.

LinkA New Era of Computing

At the heart of this partnership is Ray, the open-source distributed compute framework that now powers AI workloads at 1000s of companies like Uber, xAI, Cursor, Perplexity, and Coinbase.

Ray began not in a company, but in academia. Born in the UC Berkeley RISELab — the same lab and under the same professor, Ion Stoica, that created Apache Spark — Ray started as a way to tackle a hard problem in reinforcement learning: how to scale research that needed massive compute, across heterogeneous hardware, without drowning in complexity or cost.

The Berkeley team wasn’t thinking about redefining computing — they were researchers exploring how a more flexible architecture could help them build AI and iterate faster. But in the search for flexibility, redefining computing is exactly what happened.

Ray with Anyscale has evolved into the backbone of what we now call **AI-native computing** — the computing paradigm designed for the dynamic, stateful, and accelerator-heavy workloads that define the AI era. With Ray, engineers express distributed AI workloads in simple, familiar Python — scaling seamlessly from a laptop to thousands of nodes.

What once took huge teams and endless tech debt (and of course compute debt) building and maintaining distributed systems for AI can now be done in a fraction of time and money. AI teams can develop, train, and scale models across any framework, on any infrastructure, without dealing with orchestration or utilization.

LinkWhy Microsoft Partnered — and Why Now

A shift has been unfolding. A new pattern has emerged in AI.

Many companies consume AI via LLM API endpoints. But the leading companies aren’t just *consuming* AI — they’re *creating* it, building AI platforms that let their engineers and researchers train, fine-tune, and deploy custom models on their own data — using frameworks like PyTorch, TensorFlow, vLLM, and SGLang.

It’s a maturity from API-first to code-first development. We are seeing it; Microsoft is too.

Hosted APIs remain ideal for teams consuming pretrained models, but as enterprises integrate proprietary data and build fine-tuned or domain-specific models, they need full control of compute and orchestration — exactly what Anyscale and Azure are delivering together.

Several hundred enterprises have already been running open-source Ray on Azure Kubernetes Service (AKS) — combining Ray’s flexible computing architecture with Kubernetes’ orchestration to power their internal AI platforms. That trend is more than adoption; it’s a clear signal that AI computing is converging on a unified architecture for building and running AI at scale: Ray+Kubernetes.

That realization was the spark of this partnership.

With Anyscale on Azure, co-engineered as a first-party Azure service, Microsoft and Anyscale are giving every enterprise the power to build highly differentiated AI — backed by the flexibility of Ray, the simplicity and efficiency of running Ray with Anyscale, and the trust of Azure’s cloud scale and security.

LinkWhat This Partnership Means for Enterprises

This is one of a handful (literally counting on one hand) of first-party partnerships Microsoft has.

At its core, this partnership is about enabling a new generation of AI builders. Together we’re forming the foundation for AI-native computing — with a compute fabric that will underpin the next decade of AI innovation.

Together with Microsoft, we’re building ****Anyscale on Azure as a first-party Azure service, co-engineered by Anyscale and Microsoft, that brings the power of Ray directly into the Azure ecosystem. It’s delivered as a native extension of Azure Kubernetes Service (AKS) — so companies can build, run, and scale AI workloads with Anyscale from the Azure Portal, with unified billing, enterprise security, and full integration across Azure’s cloud stack.

For enterprise AI teams, this means less operational overhead and more time for innovation. The control plane that powers Anyscale will run natively on Azure, providing a single, secure environment that unifies development, training, and deployment. Teams can scale training, fine-tuning, and inference seamlessly across CPUs and GPUs — without worrying about distributed systems, cluster management, or resource utilization.

What makes this partnership special is the depth of collaboration. Microsoft’s AKS team under Brendan’s leadership has been an incredible partner in co-designing this integration, ensuring that every part of the service feels native to Azure customers. It’s a co-engineered service that brings AI-native computing to every Azure customer.

For developers, this means the same programmable power of Ray — now delivered with the scale, security, and reliability of Azure. For enterprises, it’s a way to build and run AI like the AI-natives — with the flexibility to build game-changing AI platforms, and the confidence of a first-party cloud service backed by Microsoft.

LinkBuilding the AI Future

At Anyscale, our mission is to make AI-native computing accessible to everyone. With Microsoft, we’re accelerating that mission. We’re taking the architecture that powers the AI-natives — and bringing it to every enterprise ready to build and scale.

We’re thrilled to be partnering with Microsoft.

Check out our Anyscale on Azure announcement blog to learn more.