AMD wants a much bigger piece of the AI hardware market, and it is not being subtle about it. At the company’s sold-out Advancing AI conference in San Francisco on Thursday, Chair and CEO Lisa Su announced Helios, a rack-scale AI system built to compete directly with Nvidia’s dominant lineup. The company plans to ship it later this year.
Rack systems bundle many processors into a single high-powered unit built for data centers, where they train and run AI models and other compute-heavy workloads. Su called Helios the tech industry’s “highest-performance AI rack,” saying it was “built to train and run the most demanding frontier models in the world at massive scale.” AMD says the system will be deployed at gigawatt scale by leading AI companies.
The announcement matters because Nvidia has owned this market for years, with its Vera Rubin and Grace Blackwell rack systems setting the standard. AMD is now claiming Helios beats Vera Rubin on several performance metrics, a claim backed up by reporting from The Register. If those numbers hold up under real-world conditions, AMD could give data centers a credible alternative for the first time.
The customer list already attached to Helios is hard to ignore. OpenAI, Meta, Oracle, Anthropic, and Microsoft all have plans to deploy the system. Microsoft CEO Satya Nadella said this week the company will expand its Azure infrastructure using Helios. Anthropic went further, announcing a strategic partnership with AMD on Wednesday to deploy up to two gigawatts of GPUs through the new rack system. These are not small pilot programs. These are major commitments from organizations that spend billions on compute every year.
Helios is not a surprise out of nowhere. AMD first showed it in 2025 and brought it onstage at CES 2026 in January. But Thursday’s conference put it center stage alongside a full customer rollout and a clearer shipping timeline. The company also introduced the Venice-X CPU, a data center chip built for high-compute workloads, though that one is not expected to land until 2027.
Su used her remarks to lay out why the stakes are so high right now. She pointed to agentic AI as the key driver of surging compute demand. When an AI agent tackles a task, it does not just generate a single response. It reasons through a problem, calls tools, accesses data, and repeats that process many times over until it reaches a solution. That kind of workload demands a lot of GPUs running in parallel.
The numbers Su put forward reflect that shift. AMD now expects the AI accelerator market to reach around $1.4 trillion by 2030. To put that in perspective, that would be roughly the size of the entire semiconductor market today. Su also predicted GPUs will make up the vast majority of that market, because AI algorithms are still changing fast and programmable silicon handles that better than fixed-function chips.
For AMD, Helios is a bet that the AI hardware race is still open enough to win share from Nvidia. The customer commitments suggest it has a real shot. Whether performance in production matches the benchmarks will be the real test once deployments begin later this year.
