Analysis
AMD used its Advancing AI 2026 event in San Francisco on July 22-23 to unveil its most significant infrastructure push yet: EPYC Venice, a Zen 6 server CPU built on TSMC's 2nm process with up to 256 cores and 512 threads and the first major adoption of PCIe 6.0, alongside the Helios rack-scale AI system, which CEO Lisa Su confirmed is now in full production.
Helios pairs 72 Instinct MI455X GPUs with EPYC Venice CPUs and Pensando networking in a single rack carrying 31 terabytes of HBM4 memory, with partner shipments beginning by the end of Q3 2026. The more consequential news, however, was commercial: Meta and OpenAI each signed multi-generation agreements for up to 6 gigawatts of combined AMD compute capacity, with initial 1-gigawatt deployments starting in the second half of 2026. Oracle separately committed to a 50,000-GPU public cloud cluster in Q3, and Microsoft will deploy Helios for Azure AI inference workloads.
“Oracle separately committed to a 50,000-GPU public cloud cluster in Q3, and Microsoft will deploy Helios for Azure AI inference workloads.”
The deals mark a genuine shift in the competitive landscape: Nvidia has held an overwhelming share of frontier AI training and inference compute for years, but Meta and OpenAI -- two of the largest AI infrastructure spenders on earth -- formally diversifying their chip supply at gigawatt scale is a different kind of validation than AMD has received before. It follows AMD's own $5 billion investment commitment into Anthropic disclosed days earlier, deepening the compute relationship from multiple angles.
For infrastructure and semiconductor investors, hyperscaler-scale commitments of this size are a stronger signal of AMD's competitive viability than any benchmark comparison against Nvidia's chips -- Meta and OpenAI are underwriting real, multi-year capacity commitments with their own capital, not just running comparative tests. The risk is execution: AMD has a mixed history of hitting aggressive shipment timelines at the volumes hyperscalers require, and any delay in Helios's Q3 shipment target would hand Nvidia more time to extend its lead.
Watch whether AMD hits its Q3 2026 Helios shipment target, and whether the initial 1-gigawatt Meta and OpenAI deployments scale toward the full 6-gigawatt commitment on schedule.
