AMD Launches AI Rack System Helios Deployed by Microsoft
AMD has officially launched its first AI-focused rack system, Helios, and announced that Microsoft will deploy the system in its Azure data centers, becoming an important customer following Meta, OpenAI, and Oracle.
Helios competes with NVIDIA's Grace Blackwell and Vera Rubin systems, integrating self-developed GPUs, CPUs, network chips, and software ecosystems. It uses the MI400 series GPUs, with each compute tray equipped with 4 Instinct GPUs and 1 EPYC CPU, expected to be delivered later this year, offering lower costs per token.
Capital for AI infrastructure is accelerating towards diversification among multiple suppliers, with hyperscalers like Microsoft becoming buyers of AMD Helios. NVIDIA's reliance on a single supplier is under pressure, while Meta's deployment plan of up to 1GW and the fact that 8 out of the top 10 global AI companies use AMD Instinct are driving AMD's market share expansion in AI hardware.
Source: Public Information
ABAB AI Insight
AMD has been continuously challenging NVIDIA's position in the AI accelerator market through its Instinct GPU series. The launch of the Helios rack system continues its historical path of evolving from a single GPU to a full-stack AI infrastructure.
On the capital front, AMD integrates self-developed hardware and software resources to attract major clients like Microsoft and Meta, motivated by the goal of competing for AI computing orders through lower costs per token, forming a parallel supply chain option to NVIDIA.
Similar to Intel's multi-supplier strategy in the data center CPU market, or the past GPU market's shift from a single dominance to competition between AMD and NVIDIA, AMD is currently in a transition phase in AI hardware from NVIDIA's monopoly to a dual oligopoly or multi-competitive landscape.
Structural Judgment Essentially belongs to the restructuring of the industry chain: Full-stack solutions like Helios break NVIDIA's end-to-end control over AI rack systems, promoting diversification in the hyperscaler supply chain, driven by cost pressures and supply chain security needs, accelerating capital concentration towards more cost-effective alternatives.
ABAB News · Cognitive Laws
- The greater the risk of a single supplier, the more urgent the need for multi-source procurement.
- Cost per token determines the winner in the AI computing market.
- Full-stack integration is the true weapon to challenge monopolies.