AMD unveils Helios server to challenge Nvidia’s AI dominance

AMD has revealed its upcoming AI server, Helios, which is set to launch in 2026 using the new MI400 series chips.

The announcement, made by CEO Lisa Su at the ‘Advancing AI’ event in San Jose, marks a direct challenge to Nvidia’s hold on the AI hardware market.

OpenAI’s Sam Altman joined the presentation, confirming that his firm is already collaborating with AMD on future chip development, specifically the MI450 line.

The Helios server will house 72 MI400 chips, mirroring Nvidia’s NVL72 server design. Unlike Nvidia, AMD plans to make Helios’ networking standards openly available, encouraging broader collaboration across the sector instead of restricting competitors through proprietary tech like NVLink.

AMD executives stressed that fostering an open ecosystem is essential for the future of AI, a position that appears to resonate with partners such as Meta, xAI and Oracle, all of whom expressed support during the event.

While AMD’s ROCm software still lags behind Nvidia’s CUDA in industry adoption, the company is investing heavily to close the gap. AMD recently acquired server builder ZT Systems and has made 25 strategic investments tied to AI in the past year.

Despite regulatory limits on exports to China, AMD still expects strong AI chip growth. However, its stock dipped 2.2 per cent following the announcement, as analysts noted the new chips alone are unlikely to shift market dynamics in the short term.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!