AMD takes on Nvidia with its Helios AI rack-scale system
TechCrunch Lucas Ropek ● Covered by 8 sources
AMD unveiled Helios, a rack-scale AI system meant to rival Nvidia's top hardware. OpenAI, Meta, Microsoft and Anthropic are already lined up to buy it.
Based on reporting by TechCrunch, Lucas Ropek — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AMD spent Thursday's Advancing AI conference in San Francisco doing something it rarely gets to do: playing offense against Nvidia. CEO Lisa Su stood on stage and called Helios, the company's new rack-scale AI system, the industry's highest-performance rack, built to train frontier models at gigawatt scale. That's not a small claim, and according to a report from The Register, Helios actually beats Nvidia's Vera Rubin system on several performance metrics.
The customer list is the real tell here. OpenAI, Meta, Oracle, Anthropic, and Microsoft have all signed on to deploy Helios once it ships later this year. Satya Nadella confirmed Monday that Microsoft will build out Azure using the system, and just a day before AMD's event, Anthropic announced a partnership to deploy up to two gigawatts of AMD GPUs through Helios. Nvidia has owned this rack-scale market for years with Grace Blackwell and now Vera Rubin — AMD landing this many marquee names in one week is not nothing.
AMD didn't stop at Helios. It also introduced the Venice-X, a data-center CPU aimed at heavy compute workloads, though that one won't arrive until 2027. Su used the moment to lay out a much bigger forecast: by 2030, she said, the AI accelerator market alone will hit roughly $1.4 trillion, nearly matching the entire size of today's semiconductor industry. GPUs, she argued, will keep dominating that spend because AI algorithms and workloads are still shifting fast, and flexible silicon wins when nobody knows what next year's model architecture looks like.
Su tied the demand surge directly to agentic AI — systems that don't just answer a prompt but chain together dozens of reasoning steps, tool calls, and data lookups to finish a task. Each of those steps eats compute, and multiplied across millions of users, that's the workload AMD is betting Helios gets built for. Whether AMD can turn early enthusiasm into sustained market share against Nvidia's entrenched ecosystem is still an open question, but for the first time in a while, it's a fair fight.
My take — AI-written commentary, not fact-checked reporting
I've been skeptical of every 'Nvidia killer' announcement for years, but locking in OpenAI, Meta, Microsoft, and Anthropic in the same week is a different kind of signal than a benchmark slide. Big labs don't diversify suppliers out of loyalty — they do it because single-vendor dependency on Nvidia has become a real business risk, and that's the crack AMD is wedging itself into. I'd rather see three credible hardware makers fighting over gigawatt-scale AI infrastructure than one company setting the price for the entire industry.
Read more about this at: TechCrunch