# Anyscale Newsletter: Run AI in your own Azure tenant + Ray Summit… Canonical: https://brew.new/browse/templates/email/pt1_k97j57e53vgxz70grt350tx03x8e5vxb Brand: anyscale.com Category: newsletter ![Preview of Anyscale Newsletter: Run AI in your own Azure tenant + Ray Summit…](https://cdn.brew.new/email-preview-930adab6d9673754-tracking_r57prfqq8ykpfj89e0j8z32djh8dv5dv-1789000177464.png) ## Email content Hi Thomas, Enterprises are moving past calling hosted model APIs and toward building AI systems on their own data, inside infrastructure they control. That shift was the headline of our biggest announcement this month, and it runs through everything below: a major platform launch, the Ray Summit speaker lineup, and new stories from teams scaling AI in construction, drug discovery, payments, and autonomous driving. Anyscale on Azure enters public preview Anyscale_Azure Announced at Microsoft Build: any Azure customer can now provision Anyscale inside their own Azure tenant. Co-engineered with Microsoft and delivered as an Azure Native integration, Anyscale on Azure inherits the governance and operational model of first-party Azure services. Deploy through the Azure Portal or ARM templates, govern access with Microsoft Entra ID and Azure RBAC, apply Azure Policy without modification, and draw down consumption against your existing MACC. Teams already building on Anyscale on Azure include Wayve, training and deploying autonomous driving AI at the scale needed for real-world deployment, and Xoople, running distributed AI workloads over planetary-scale satellite imagery. Read the announcement blog → Get started on Azure Webinar: Anyscale on Azure, build and deploy AI at scale in your own tenant Azure webinar Want to see the public preview in action? Join us on June 16 for a session with Daniel Arrizza (Anyscale) and Paul Yu (Microsoft) covering where Anyscale fits in your AI stack on Azure, how it integrates with Microsoft Entra ID, Azure RBAC, Azure Policy, Azure Monitor, and Microsoft Cost Management. The session includes a live demo of building, training, and serving a real AI workload in an Azure environment. Register for the June 16 webinar → Ray Summit returns to San Francisco, Aug 24-26 Ray Summit 2026 speaker lineup The first wave of speakers is live: Felix Heide (Torc Robotics), Ioannis Antonoglou (Reflection AI), Kevin Peterson (Bedrock Robotics), Liam Fedus (Periodic Labs, co-creator of ChatGPT), and Anyscale co-founders Robert Nishihara and Ion Stoica. Three days of technical talks, hands-on training, and community across four frontiers: foundation model training, multimodal data curation, physical AI, and LLM reinforcement learning. Running AI in production on Ray? Submit a talk now before the CFP closes on June 20. Early bird pricing runs through July 10, with Summit passes at $200 (regular $400). Register for Ray Summit → Topping off Ray on the Road: how Adyen trained a foundation model on 51 trillion tokens Ray Day NYC Ray Day London closed out our eight-city Ray on the Road series. Adyen shared how they trained a Transaction Foundation Model on roughly 51 trillion tokens of ray payment sequences, more than twice the 22 trillion text tokens behind Llama 4, after replacing a legacy PySpark pipeline with a single streaming Python pipeline on Ray. Xoople ran pixel-level predictions over 500,000 km² of satellite imagery in under 5 minutes on 12 A10 GPUs, Criteo took its retail media relevancy model multimodal and sped up its offline inference pipeline 50x, and BMW walked through self-hosting models behind its AI Gateway with Ray Serve and vLLM. Read the London recap → And from the NYC stop: Torc Robotics lifted average GPU utilization from 30-40% to around 90% by consolidating five fragmented systems into one Ray-based compute engine, plus talks from Discord, Cubist, and Coinbase. Read the NYC recap → How Bedrock Robotics builds autonomous construction systems on Anyscale bedrock-excavator Bedrock Robotics is bringing autonomy to heavy machinery, starting with excavators on construction sites. Their pipelines mix GPU video decoding, CPU preprocessing, and GPU inference in the same workload, a pattern existing processing engines weren’t built for. Running everything from raw sensor ingestion to model deployment on Ray on Anyscale, Bedrock scaled from 20K to 1.7M compute hours in under 12 months, an 85x increase, while cutting costs 40% by running 80% of their fleet on spot instances. Read the Bedrock case study → TMLS Kickoff Social: Toronto, June 16 tmls kickoff social Kicking off Toronto Machine Learning Summit week? Join us for an evening on the patio with Georgian and Google Cloud for Startups. RSVP for the June 16 social → Anyscale at PlatformCon, June 22-26 PlatformCon Kubernetes and its ecosystem are the foundation of modern platforms, but AI workloads bring requirements around distributed computing, resource and dependency management, and production reliability that existing platform tooling wasn't designed for. Join Christian Stano, Field CTO at Anyscale, for a virtual session during PlatformCon (June 22 to 26) on why platform teams at companies like Cursor, Uber, Netflix, and NVIDIA choose Ray, and how AI platform architecture is evolving to meet today's scaling needs. Register for PlatformCon → Anyscale_logo_grey_blue Advance your AI Platform with Anyscale. LinkedIn X Facebook Anyscale, Inc., 600 Harrison St, San Francisco, CA 94107, USA Unsubscribe Manage preferences [Open and remix this design](https://brew.new/browse/templates/email/pt1_k97j57e53vgxz70grt350tx03x8e5vxb) [Browse email designs](https://brew.new/browse/templates)