AMD takes on Nvidia with its Helios AI rack scale system

Chipmaker AMD is taking purpose astatine rival Nvidia with its latest hardware release: a rack-scale strategy designed to powerfulness computing needs of the world’s largest AI labs.

At the company’s sold-out Advancing AI league successful San Francisco connected Thursday, AMD Chair and CEO Dr. Lisa Su promoted the caller AI rack strategy known arsenic Helios — on with its increasing database of customers, including Microsoft — arsenic the institution prepares to vessel it aboriginal this year. Su besides pitched the company’s newest chips that are designed to provender the compute-hungry dragon that is the AI industry.

Rack systems harvester galore processors into a single high-powered unit. They are built for information centers, wherever they bid and tally AI models and different compute-intensive workloads.

Su called Helios the tech industry’s “highest show AI rack,” adding that it was “built to bid and tally the astir demanding frontier models successful the satellite astatine monolithic scale.” The strategy volition beryllium deployed by starring AI companies astatine gigawatt-scale, the institution said.

Nvidia has historically dominated this marketplace with its Vera Rubin and Grace Blackwell rack-scale systems. AMD is intelligibly looking to get successful connected the action. And Helios’ show metrics look to springiness it a existent chance, beating retired Vera Rubin by a fig of metrics, the Register reported.

Helios, which was archetypal revealed successful 2025 and shown onstage successful January astatine CES 2026, already has respective well-known customers, including OpenAI, Meta, Oracle, Anthropic, and Microsoft, each of which have plans to deploy the system. Microsoft CEO Satya Nadella said Monday that the institution would grow its Azure infrastructure with Helios. Meanwhile, Anthropic and AMD announced a strategical partnership Wednesday for deploy up to 2 gigawatts of GPUs via the caller rack system.

AMD also introduced Thursday its Venice-X CPU, which is designed for information centers and to grip high-computing workloads. The Venice-X is expected to motorboat successful 2027.

During her remarks, Su commented connected the trajectory of the spot industry, claiming that, by the twelvemonth 2030, chips that powerfulness AI volition go a monolithic portion of the wide computing market. This’s due to the fact that the manufacture is “seeing a measurement alteration successful compute demand” driven mostly by the emergence of agentic AI, she said.

“When you inquire the cause to bash something, it really has dozens of steps, and it has to reason, and it has to telephone tools, and it has to entree data, and it has to support doing it implicit and implicit until it solves the problem, and truthful you request tons of GPUs to bash each that,” the enforcement said.

“We’re present expecting that by 2030, the AI accelerator marketplace is going to scope astir $1.4 trillion,” Su said. “What that means is, by the extremity of the decade, the AI accelerator marketplace is going to attack the size of the full semiconductor marketplace today.”

“We bash expect that GPUs are going to marque up the immense bulk of that marketplace due to the fact that the algorithms are inactive precise overmuch successful their infancy, and we’re inactive continuing to spot the workloads change, and that favors programmability successful the wide silicon ecosystem,” she added.