AMD just drew a line in the sand. The chipmaker will begin shipping its Helios platform to customers in the second half of 2026, with Microsoft deploying it at scale on Azure for AI inference.

Helios isn’t a single chip. It’s an entire rack-scale system integrating AMD’s next-generation Instinct GPUs alongside its 6th-generation EPYC CPUs, codenamed Venice.

What Helios actually delivers

A single Helios rack can push up to 1.4 exaFLOPS of FP8 compute and 2.9 exaFLOPS of FP4 compute. Each rack also comes loaded with 31 terabytes of HBM4 memory.

The GPU side of Helios features the MI455X and MI450 models built on AMD’s CDNA 5 architecture. The platform was first unveiled at the 2025 Open Compute Project Global Summit as an open-source alternative to proprietary AI systems, combining AMD Instinct GPUs, EPYC CPUs, and advanced networking technologies. AMD has also announced strategic partnerships with Celestica and Super Micro to support deployments.

Engineering samples and low-volume production begin in the second half of 2026. Mass production ramps are expected by Q2 2027.

The customer list is already forming

Microsoft is deploying Helios on Azure for AI inference. Meta has plans to begin using custom MI450-based GPUs for AI in the second half of 2026, with initial deployments targeting 1 gigawatt of capacity. Oracle Cloud will initiate a rollout of 50,000 GPUs starting in Q3 2026.

Why this matters for markets

By offering a rack-scale solution rather than individual GPUs, AMD is competing on system-level integration. The open-source nature of the platform could lower switching costs for cloud providers. If AMD’s Helios creates genuine pricing pressure on Nvidia, that could reduce costs across the board for AI inference and other parallel processing workloads.

Investors should watch three things closely: whether the Q2 2027 mass production timeline holds, how quickly Azure inference workloads migrate to Helios hardware, and whether the Oracle and Meta deployments expand beyond initial commitments.