AMD on Thursday unveiled Helios, a rack-scale AI system designed to compete directly with Nvidia’s dominant lineup of data center hardware. The announcement came at the sold-out Advancing AI conference in San Francisco, where AMD Chair and CEO Dr. Lisa Su promoted the system as the company’s answer to Nvidia’s Vera Rubin and Grace Blackwell platforms.
Helios combines many processors into a single high-powered unit for data centers. It is designed to train and run AI models and other compute-intensive workloads. Su called Helios the “highest-performance AI rack” and said it was “built to train and run the most demanding frontier models in the world at massive scale.” The system represents AMD’s most ambitious push yet into the market for large-scale AI infrastructure, a space long dominated by Nvidia.
Helios already has major customers lined up
AMD revealed Helios in 2025 and showed it onstage at CES 2026. The system already has a roster of customers: OpenAI, Meta, Oracle, Anthropic, and Microsoft. Microsoft CEO Satya Nadella said Monday that Azure infrastructure would expand with Helios. On Wednesday, Anthropic and AMD announced a strategic partnership to deploy up to two gigawatts of GPUs via Helios. Helios is expected to ship later this year.
The system beats Nvidia’s Vera Rubin by several metrics, according to The Register. Nvidia has historically dominated the rack-scale market with Vera Rubin and Grace Blackwell systems, but AMD is positioning Helios as a serious challenger. The Register reported that Helios outperforms Vera Rubin on key benchmarks, though specific figures were not disclosed at the event. AMD executives emphasized that the system’s modular design allows data center operators to scale from a single rack to thousands of interconnected units.
New Venice-X CPU targets data centers
Alongside Helios, AMD introduced the Venice-X CPU on Thursday. Venice-X is designed for data centers and high-computing workloads and is expected to launch in 2027. The chip adds to AMD’s growing portfolio of hardware aimed at the AI and cloud computing markets. AMD said Venice-X will offer improved memory bandwidth and core counts compared to its predecessor, though exact specifications were not released. The chip is intended to complement Helios by handling general-purpose computing tasks while GPUs focus on AI training and inference.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
AI accelerator market could hit $1.4 trillion by 2030
Su painted a picture of explosive growth in the AI hardware market. “We’re now expecting that by 2030, the AI accelerator market is going to reach about $1.4 trillion,” she said. “What that means is, by the end of the decade, the AI accelerator market is going to approach the size of the entire semiconductor market today.”
She added that GPUs will dominate that market. “We do expect that GPUs are going to make up the vast majority of that market because the algorithms are still very much in their infancy, and we’re still continuing to see the workloads change, and that favors programmability in the overall silicon ecosystem,” Su said. The $1.4 trillion figure represents a significant upward revision from earlier AMD forecasts, reflecting the company’s confidence that demand for AI compute will continue to accelerate.
Agentic AI drives demand for more GPUs
Su cited the rise of agentic AI as a key driver of demand. “When you ask the agent to do something, it actually has dozens of steps, and it has to reason, and it has to call tools, and it has to access data, and it has to keep doing it over and over until it solves the problem, and so you need lots of GPUs to do all that,” she said. She described the shift as “a step change in compute demand.” Su explained that traditional AI models process a single query and return a result, but agentic AI systems perform multiple iterative steps, each requiring significant GPU compute. This new paradigm, she argued, will require data centers to invest in far more hardware than previously planned.