Microsoft expands Azure AI chip supply with AMD Helios

Microsoft plans to deploy more AMD AI chips in Azure through AMD's forthcoming Helios platform, giving cloud customers another option for large-model infrastructure as Nvidia continues to dominate the market. The companies did not disclose purchase volumes, pricing, capacity targets or financial terms, which limits any read-through on how much Nvidia demand this displaces.

AMD said Helios is scheduled to ship in the second half of 2026 and is built for running large AI models and high-performance computing workloads. Microsoft framed the deal as part of an effort to broaden customer choice in Azure AI infrastructure. AMD CEO Lisa Su described the arrangement as a major milestone, while Microsoft CEO Satya Nadella said customers need more options for AI systems.

The announcement is a concrete win for AMD's data center AI push, but it does not by itself prove a broad shift away from Nvidia. Nvidia still has substantial pricing power in accelerator systems, according to the reporting, and its GPUs remain the default reference point for many AI training and inference deployments. What is changing is the willingness of major buyers to fund and test alternatives, including AMD GPUs and custom TPUs developed by cloud providers and other large technology companies.

Anthropic may be testing AMD systems

AMD could also be in line for another high-profile AI customer. A public GitHub profile associated with an AMD director indicates that Anthropic, the company behind Claude, is testing AMD hardware. SemiAnalysis reported that AMD assigned Anthropic its highest priority rating, a level comparable to major customers such as Meta.

That potential relationship remains unconfirmed. SemiAnalysis said Anthropic is still evaluating the hardware and that AMD may announce the partnership at its upcoming Advancing AI conference if engineering teams address remaining concerns around software quality. AMD and Anthropic have not publicly disclosed a completed deal, deployment scale, timing or commercial terms in the material available.

Software remains one of the central issues for any Nvidia challenger. Buyers can procure non-Nvidia accelerators, but AI teams also need compilers, libraries, model support and operational tooling that work reliably at scale. The reference to unresolved software quality concerns around Anthropic's testing suggests the competitive gap is not only about silicon availability or performance claims.

AMD gains logos, Nvidia keeps the lead

The Microsoft expansion follows another disclosed AMD customer win: OpenAI has a long-term partnership with AMD for multiple generations of Instinct GPUs. Taken together, Microsoft, OpenAI and a possible Anthropic relationship would give AMD more credibility with frontier AI developers and hyperscale buyers.

The harder question is how much workload share AMD can win. The companies did not say how many Helios systems Azure will deploy, which Azure regions will receive them, or whether the platform will be used mainly for internal Microsoft workloads, customer-facing cloud rentals, or both. Without those details, the announcement is best read as a supply diversification move by Microsoft and a validation point for AMD rather than evidence that Nvidia's position has been materially reset.

This story draws on original reporting from The Decoder.