CES 2026: AMD unveils Helios, a rack designed to power AI content

At CES 2026 AMD introduced Helios, an AI rack the company says delivers 2.9 exaflops of compute and includes 31 TB of HBM4 memory.

AMD's CEO Lisa Su unveiled Helios during a Monday evening keynote at CES 2026, presenting it as the company's next-generation AI rack. The presentation emphasized Helios's scale and technical specifications and included remarks from representatives of OpenAI and Luma AI about how larger pools of compute support their work. The announcement comes amid wider public debate over AI-generated content and concerns about the local impacts of building and operating large data centers. AMD also announced new consumer-focused Ryzen AI 400 series chips during the same keynote. Key details: - Helios is described as a full AI rack that weighs nearly 7,000 pounds and integrates multiple AMD components, including the Instinct MI455X GPU, EPYC "Veince" CPU, Pensando "Vulcano" 800 AI NIC, and Pensando "Salina" 400 DPU. - AMD reported Helios can provide 2.9 exaflops of AI compute and comes with 31 TB of HBM4 memory. - The rack is said to offer 43 TB per second of scale-out bandwidth and is developed using 2nm and 3nm process nodes. - AMD stated Helios contains about 4,600 "Zen 6" CPU cores and roughly 18,000 GPU compute units. - During the keynote Su noted a rapid rise in global AI compute, from about 1 zettaflop in 2022 to 100 zettaflops in 2025, as context for the product. - AMD also announced the Ryzen AI 400 series, described as offering 12 Zen 5 cores/24 threads, 16 RDNA 3.5 GPU cores, a 60 TOPS XDNA 2 NPU, and 8,533 MT/s memory speeds, and said these chips will ship soon in a range of major PC brands. Summary: The Helios reveal highlights how hardware makers are responding to sharply increased demand for AI compute while the broader conversation around AI content and data center impacts continues. AMD shared detailed technical specifications and previewed consumer AI chips, but broader deployment timelines for Helios are undetermined at this time.