Our Anyscale on Azure webinar is now available on demand. Daniel Arrizza (Anyscale) and Paul Yu (@Microsoft) on running production AI inside your own Azure tenant, plus a live build-train-serve demo. Watch now 👉 na2.hubs.ly/H0699F80

Some intuition about PD disaggregation from the blog - PD doesn't speed up prefill and can actually hurt TTFT - PD's real benefit is flat, stable TPOT under load - TPOT savings compound over output sequence length The optimal P:D ratio is dependent in particular on input lengths, output lengths, and cache hits. Meaningful optimizations are possible, but tuning can be sensitive. Benchmarks performed with @raydistributed @vllm_project on AMD MI325X. anyscale.com/blog/ray-vllm-p…

We usually divide AI workloads into two buckets: "training" and "inference". However, data processing is quickly emerging as a third major AI workload. It was previously confined to CPUs, but going forward, most data processing will happen on GPUs.

Data processing has become a GPU workload and is dominated by inference. A new architecture is required, and it is not Spark on GPUs. anyscale.com/blog/data-proce…

Save 67% with prefill-decode disaggregation using Ray vLLM on AMD GPUs. anyscale.com/blog/ray-vllm-p…

One pattern we keep seeing with customers serving LLMs at scale: Prefill-decode disaggregation is often treated like a magic wand. But the reality is more nuanced. So we wrote down the core insights for when PD helps, when it does not, and validated them on AMD vLLM — where the PD path has been much less paved. 🧵

Last chance to sign up for tomorrow's webinar! Don't miss the opportunity to learn: - Where Anyscale on Azure fits in your AI stack - How it integrates with the Azure services you already use: Microsoft Entra ID, Azure RBAC, Azure Policy, Azure Monitor, and Microsoft Cost Management - What you need to get started, from your Azure tenant to your first Ray cluster - Plus a live demo: building, training, and serving a real AI workload on Anyscale in an Azure environment na2.hubs.ly/H061tcr0

Thanks to our Ray Day: London speakers: Marcell Ferencz (@Xoople), Martin Iglesias & Maxime Battello (@Adyen), Paul Coursaux (@Criteo), and Thomas Riedl (@BMWGroup), plus a keynote from @pcmoritz. Recap 👉 na2.hubs.ly/H065bcv0 The road leads to Ray Summit SF, Aug 24–26 → na2.hubs.ly/H0659wN0

Thank you to our Ray Day: NYC speakers 🔆 Serrana Aguirregaray (@discord), Neil Wadhvana (@torc_robotics ), Todd Gaugler (Cubist), and Aman Choudhary (@coinbase) brought four different takes on Ray in production, from ML platforms to quant finance. Highlights from all 4 talks 👉 na2.hubs.ly/H064bc-0

My vLLM pipeline wouldn't start. KV cache came out negative on the L4. One prompt to /anyscale-platform-fix and the agent takes it from there. Four minutes to debug instead of a full afternoon. Full walkthrough: na2.hubs.ly/H062c-V0

Ray Summit 2026 keynote lineup is coming together. @LiamFedus (@periodiclabs, prev. co-creator of ChatGPT), @_FelixHeide_ (@torc_robotics), @real_ioannis (@reflection_ai), @kevinmpeterson1 (@BedrockRobotics), @robertnishihara & @istoica05 (Anyscale). ⏰ CFP closes June 20: na2.hubs.ly/H062cPc0

Thanks for joining Ray Day: NYC! Two tracks — distributed training with Ray PyTorch and VLA fine-tuning — plus talks from @discord, @torc_robotics, @coinbase & Cubist. Recap → na2.hubs.ly/H062d0F0  Next: Ray Summit SF, Aug 24–26 →na2.hubs.ly/H062gYk0

Last chance to sign up for tomorrow's webinar! Neil Wadhvana, Staff ML Engineer at @torc_robotics, will walk through how Torc consolidated its autonomy data processing stack to support multimodal AI at scale with Ray on Anyscale. Don't miss the opportunity to learn: - The trends driving growth in autonomous driving developments, - An overview of Torc’s data loop from production to consumption, - The internal trends in multimodal AI that drove need for consolidation, - The before and after Ray was adopted as common compute framework. na2.hubs.ly/H05Dgrf0

Anyscale on Azure is now in public preview, and we're going deep on how it works. Join Daniel Arrizza (Field Engineer, Anyscale) and Paul Yu (Senior Cloud Advocate, Microsoft) for a working session on running production AI inside your own Azure tenant – where your data stays within your existing governance. You will learn: - Where Anyscale on Azure fits in your AI stack - How it integrates with the Azure services you already use: Microsoft Entra ID, Azure RBAC, Azure Policy, Azure Monitor, and Microsoft Cost Management - What you need to get started, from your Azure tenant to your first Ray cluster - Plus a live demo: building, training, and serving a real AI workload on Anyscale in an Azure environment na2.hubs.ly/H061t3b0

Impressive work from the Cosmos team. Check out the Cosmos 3 technical report: arxiv.org/pdf/2606.02800

Congratulations to @nvidia on the release! Super exciting to see two models trained with Ray back to back (MAI-Thinking-1 and Nemotron 3 Ultra).

Nemotron 3 Ultra is an impressive 550B parameter (55B active) MoE model. It was trained with Ray / Megatron / vLLM (via NeMo RL).

GPUs in Mumbai, training data in Iowa? Cross-region reads tax every epoch. We put @Alluxio NVMe caching in front of the bucket with Ray Data on Anyscale: 1TB warm reads went 20x faster. na2.hubs.ly/H05YMGW0

The bottleneck in drug discovery isn't designing molecules, it's making them. onepot combines robotic synthesis with large-scale ML inference on Anyscale to predict which reactions will work before they run, achieving 3B compounds and 10B reactions scored. Case study: na2.hubs.ly/H05PcCG0"