AMD is preparing to ship Helios, its first rack-scale AI system, later this year, with Microsoft announcing it will deploy the system in Azure data centers1,3. Microsoft joins Meta, OpenAI, Oracle, and Anthropic as early customers for the system, which AMD positions as the first direct rival to Nvidia's Grace Blackwell and Vera Rubin rack-scale platforms2.
AMD Chair and CEO Dr. Lisa Su unveiled the system at the company's sold-out Advancing AI conference in San Francisco, calling Helios "the tech industry's highest performance AI rack" built to "train and run the most demanding frontier models in the world at massive scale". Su said the system will be deployed by leading AI companies at gigawatt scale.
Each Helios rack contains 72 GPUs and 18 CPUs, weighs 7,000 pounds, and carries a price tag of approximately $5 million — at least $1 million more than Nvidia's Vera Rubin4,5. AMD describes Helios as a reference design that manufacturing partners can use to build data center racks, and says it comes in four customizable configurations depending on customer needs.
Su said in an interview that Helios systems deliver 10% to 15% more performance than competing systems and offer up to a 30% benefit in tokens per dollar for customers. She cited multi-gigawatt-scale MI350 deployments with OpenAI and Meta, and said Anthropic will deploy up to 2 gigawatts of MI350 in Helios starting in the first half of 2027.
Meta announced it will use up to six gigawatts of AMD GPUs, starting with one gigawatt deployed on Helios racks later this year. Microsoft announced the partnership alongside three new Azure instance families. "We are expanding the Azure infrastructure portfolio with AMD Helios," a Microsoft representative said. Financial terms and compute capacity details were not disclosed.
AMD's data center revenue is up 57% year-over-year, and the company hit a record 46% revenue share in the x86 server market in Q1 2026. AMD says eight of the top ten AI companies are running workloads on its Instinct GPU line. The company expects tens of billions of dollars in data center AI revenue starting next year, with the majority coming from Helios.
The system's development drew on AMD's 2025 acquisition of server builder ZT Systems for $4.9 billion and its 2022 acquisition of Pensando, which AMD says made its networking technology a differentiator. AMD also committed $10 billion in May to Taiwanese advanced packaging companies such as ASE.
One analyst quoted in CNBC's coverage framed the competitive question directly: "Is AMD winning because they're technologically superior? Or does AMD win because there's just such a constraint on capacity that if they can build it, someone will buy it".
ANALYSIS The customer roster — Microsoft, Meta, OpenAI, Oracle, and Anthropic — represents the largest buyers of AI compute infrastructure. The gigawatt-scale deployment commitments cited by Su indicate that these customers are treating Helios as a primary compute platform rather than a secondary or experimental option. The $5 million per-rack price point, at least $1 million above Vera Rubin, suggests AMD is competing on performance and efficiency metrics rather than undercutting on cost. AMD's ability to convert its CPU market share gains and GPU adoption among top AI companies into rack-scale system sales marks a structural shift in how the company competes with Nvidia — moving from component supplier to integrated system vendor.