AMD Helios Rack-Scale AI Platform Enters Mass Production with Performance Claims

At the Advancing AI 2026 event, AMD CEO Lisa Su announced that the next-generation Helios rack-scale AI platform has entered mass production, shipping this quarter with continued capacity ramp-up next quarter. Officially positioned as the world's fastest AI rack, AMD claims it delivers 15% more computing power and 50% higher HBM4 memory capacity and bandwidth compared to competing products. Additionally, AMD and on-stage guests stated that Helios can produce up to 30% more tokens per dollar than Nvidia Vera Rubin NVL72.

已确认

According to multiple attendees at the event, the following information was officially announced:

- **量产与交付状态**: The Helios platform is in full mass production, shipping this quarter, with capacity continuing to ramp up next quarter.

- **硬件规格优势**: Compared to competitors, Helios offers 15% more computing power, and HBM4 memory capacity and bandwidth are both 50% higher.

- **性价比指标**: Helios can achieve up to 30% more token output per dollar compared to Nvidia Vera Rubin NVL72.

为什么重要

The mass production of Helios marks AMD's direct challenge to Nvidia in the high-end rack-scale AI infrastructure market. The various comparative data points released by AMD—especially the cost-effectiveness metrics directly benchmarking Vera Rubin NVL72—indicate that AMD is attempting to prove the competitiveness of its full-stack hardware in large-scale AI training and inference through compute power, memory bandwidth, and token yield per dollar.

2026-07-24 ~ 2026-07-24 · 5 related posts

Full story(3 episodes)→

Primary sources