AMD just locked in a full-stack deal with Azure touching GPUs, processors, network gear, and the software layer, and Microsoft is going all-in on the whole package. The centerpiece is Helios Rackscale, bundling Instinct MI455X accelerators, EPYC Venice CPUs, Pensando DPUs, and ROCm into one open platform for frontier-model inference, with shipments rolling out later this year.
Beyond Helios, Azure is picking up two fresh VM series running sixth-generation Venice EPYC chips. HDv2 covers agentic AI alongside data pipeline work, while HXv2 handles semiconductor design. On the networking side, Microsoft is weaving Pensando DPUs deeper into Azure Boost to squeeze better throughput and connection handling across cloud-scale deployments, which is where the real bottlenecks live.
Lisa Su framed this as the payoff from years of joint infrastructure work, while Nadella pitched it as giving builders the range to train, serve, and iterate without being locked into one vendor's silicon. Enterprise teams can tap the whole stack via Azure Foundry Managed Compute, and frontier labs get direct access to AMD-powered clusters for scaling their models.
Beyond Helios, Azure is picking up two fresh VM series running sixth-generation Venice EPYC chips. HDv2 covers agentic AI alongside data pipeline work, while HXv2 handles semiconductor design. On the networking side, Microsoft is weaving Pensando DPUs deeper into Azure Boost to squeeze better throughput and connection handling across cloud-scale deployments, which is where the real bottlenecks live.
Lisa Su framed this as the payoff from years of joint infrastructure work, while Nadella pitched it as giving builders the range to train, serve, and iterate without being locked into one vendor's silicon. Enterprise teams can tap the whole stack via Azure Foundry Managed Compute, and frontier labs get direct access to AMD-powered clusters for scaling their models.
