AMD unveiled Helios, a 72-GPU rack-scale AI system built to rival NVIDIA’s top infrastructure. Here is why open standards and real hardware competition benefit the AI industry.
NVIDIA has dominated AI hardware for years. AMD just launched a direct challenge. At its Advancing AI event, CEO Lisa Su unveiled Helios- a fully integrated rack-scale system built to train and run giant AI models.
Helios packs 72 new Instinct MI455X GPUs, 6th-Gen Epyc CPUs, and high-speed Pensando networking into a single liquid-cooled frame. AMD claims the system delivers up to 30% more tokens per dollar than NVIDIA’s competing Vera Rubin NVL72 platform.
More importantly, major players like Meta, OpenAI, Microsoft, and Anthropic have already committed to test and deploy Helios at gigawatt scale.
Building a fast chip is no longer enough. Modern AI labs need entire data centers operating as unified supercomputers. NVIDIA mastered this end-to-end integration early. AMD is now matching that hardware scale while taking a distinct strategic path: open standards.
Helios relies on open frameworks like UALink and Ultra Ethernet rather than locking customers into proprietary networks. AMD is also aggressive on software. Anthropic even agreed to use its Claude model to help optimize AMD’s open ROCm platform.
This open strategy provides vital market nuance:
- Dual-sourcing hardware reduces single-vendor supply risks and cuts training costs for AI labs.
- Open standards prevent ecosystem lock-in to drive faster innovation across hardware tiers for developers.
AMD still faces a steep climb.
NVIDIA’s CUDA software ecosystem remains deeply entrenched among developers- translating new hardware into raw developer adoption takes time.
But Helios has proved that genuine competition has finally arrived in top-tier AI infrastructure. And that shift creates a more sustainable market for everyone building the future of AI.
