Much of the value of the current AI boom has accrued disproportionately to one company: Nvidia. But the biggest challenge to its flagship product has now emerged.
AMD Helios has arrived to take on Nvidia's Vera Rubin, the most powerful system in the world for both training the most advanced AI models and running them at scale. While nearly all the major AI and tech companies are trying to build their own AI chips now, Vera Rubin — and now Helios — represent the top of the food chain.
These are the rack-scale systems that combine CPU, GPU, high-bandwidth memory, networking, and software into one platform. And they are the kinds of chips that will power the AI factories, those massive data centers custom-designed to efficiently run the most advanced cutting-edge systems.
"Helios is truly a game-changer," AMD CEO Lisa Su said last year when the company first announced the product. "For the first time we architected every part of the rack as a unified system. That's combining our CPUs, our GPUs, our Pensando NICs, and our ROCm software all together in one platform. And it's really purpose-built for the most demanding AI workloads from training the largest frontier models to scaling inference across thousands of nodes."
It's not an exaggeration to say that AMD bet the company on this product. AMD's data center GM Forrest Norrod told CNBC that it has a town hall every quarter for everyone working on Helios, and attendance has swelled to 8,000 employees.
On Monday, AMD Helios got a major vote of confidence from the world's second-largest hyperscaler. Microsoft officially signed on as a partner to deploy the full-stack version of Helios on Azure to allow its customers to train frontier models, deploy AI inference at scale, and offer the most advanced AI agent capabilities.
AMD claims that its first-generation AI rack system will rival Nvidia's second-gen system in virtually every aspect, and will exceed it in some. Agents are expected to be one of its strengths because agents rely on CPU as much as GPU, unlike chatbots, and AMD has a much longer history in building CPUs than Nvidia. AMD has also signed up Meta, OpenAI, Oracle, and Tata as partners planning to deploy Helios.
Our Deeper View #
AMD's rival to Vera Rubin doesn't mean Nvidia is in trouble. This is not a zero-sum market, but one that is growing rapidly and doesn't have enough chips or servers to grow as fast as it would like. Nvidia has reportedly already sold its next 12 months of products before they are even manufactured. Demand is so strong for the infrastructure to power the growth of AI that there's plenty of room for AMD to ease some of that burden, if its products turn out to be in the same ballpark as Vera Rubin. Helios will cost more, around $5 to $5.5 million compared to $3.5 to $4 million for Vera Rubin, according to the Futurum Group. But it's also promising greater efficiencies and potentially cost savings, so it could be less expensive to operate over time. I'll be at AMD's Advancing AI 2026 event this week in San Francisco and will have more to report on the latest details of Helios and the future of AI infrastructure.
Disclaimer: Jason Hiner's travel to AMD's Advancing AI 2026 event was paid for by AMD. The Deep View's coverage is editorially independent from the companies we cover.