Chipmaker AMD is stepping up its challenge to market leader Nvidia with the introduction of its Helios AI rack-scale system, a powerful new hardware solution designed for the demanding computational needs of the world's largest artificial intelligence laboratories. The system, which combines multiple processors into a single high-performance unit, is slated to begin shipping to customers later this year, marking a significant push into a sector previously dominated by Nvidia's offerings.
Helios was unveiled by AMD Chair and CEO Dr. Lisa Su at the company's recent Advancing AI conference. Dr. Su highlighted Helios as the tech industry's "highest performance AI rack," built to train and run the most complex "frontier models" at an immense scale. This system is crucial for data centres, where it will facilitate the training and operation of advanced AI models and other compute-intensive workloads. The company has already secured commitments from several prominent AI developers and tech giants, including Microsoft, OpenAI, Meta, Oracle, and Anthropic, all planning to deploy the system.
The introduction of Helios signals a direct confrontation with Nvidia, which has historically held a strong position in this market with its Vera Rubin and Grace Blackwell rack-scale systems. Reports suggest that Helios's performance metrics are highly competitive, even surpassing Vera Rubin in several key areas. This intensified competition is expected to accelerate innovation within the AI hardware space, potentially leading to more advanced and efficient solutions for businesses and researchers globally.
Beyond the Helios system, AMD also introduced its Venice-X CPU, a new central processing unit specifically designed for data centres and high-computing workloads, expected to launch in 2027. These developments underscore AMD's strategic focus on the burgeoning AI industry. Dr. Su projected that by 2030, chips powering AI will constitute a substantial portion of the overall computing market, driven by a "step change in compute demand" fuelled by the rise of 'agentic AI'.
Dr. Su elaborated on the computational demands of agentic AI, explaining that tasks requiring agents involve numerous steps, reasoning, tool calls, and data access, all executed repeatedly until a problem is solved. This intricate process necessitates significant GPU power. She forecasted that the AI accelerator market alone could reach approximately £1.1 trillion by 2030, approaching the current size of the entire semiconductor market. GPUs are expected to form the vast majority of this market, given the nascent stage of AI algorithms and the evolving nature of workloads that favour programmability in the silicon ecosystem.