Anyscale, the company behind the open-source Ray framework, has launched its AI scaling platform on Microsoft Azure in public preview. The service lets enterprises distribute and scale machine learning workloads across clusters, with built-in hooks into Azure's native services.
What the platform does
At its core, Anyscale is a managed version of Ray, an open-source framework for distributed computing. Companies use Ray to split large AI jobs—training models, running simulations, processing data—across many machines. Anyscale handles the orchestration, monitoring, and infrastructure so teams don't have to build their own distributed systems.
On Azure, that means customers can spin up Anyscale clusters directly from their Azure account. The platform connects to Azure Blob Storage, Azure Machine Learning, and other services without extra configuration. For enterprises already running on Azure, it's a way to add distributed computing without managing separate hardware or Kubernetes clusters.
Why Azure matters
Azure is one of the three big cloud providers, and Anyscale's native integration eliminates the friction of stitching together different tools. Users can launch a Ray cluster from the Azure portal, pull data from Azure storage, and push results back into Azure pipelines. The public preview lets early adopters test the integration before a full general availability release.
Anyscale has been available on AWS and Google Cloud for some time. Adding Azure means the platform now covers all three major clouds, making it easier for multi-cloud enterprises to standardize on Ray for AI workloads.
What's still missing
Public previews come with limitations. Microsoft and Anyscale haven't disclosed pricing for the Azure edition—users pay only for underlying compute resources during preview. The final pricing model, including any premium features, won't be known until the service reaches general availability. Azure users interested in testing can sign up through the Anyscale website or Azure Marketplace.
The preview also restricts some advanced features. For example, autoscaling and spot instance support may behave differently on Azure compared to other clouds. Anyscale says documentation and support channels are active, but users should expect occasional hiccups as the integration matures.
For enterprises wrestling with AI at scale, the preview offers a chance to try a managed Ray service without building it from scratch. Whether it replaces custom Kubernetes setups or other distributed computing tools will depend on how well it performs in real production environments—and on the final price tag.
