AMD CES 2026 keynote unveiled a comprehensive “AI Everywhere, for Everyone” vision—introducing the Helios rack-scale platform delivering 3 exaflops per rack, next-gen MI500 GPUs promising 1,000x performance gains, and Ryzen AI platforms with 60 TOPS NPUs alongside a $150 million education commitment.
Helios: Blueprint for Yotta-Scale AI Infrastructure
As global compute capacity races from 100 zettaflops to 10+ yottaflops within five years, AMD revealed the Helios rack-scale platform—a modular design delivering up to 3 AI exaflops per rack. Powered by Instinct MI455X accelerators, EPYC “Venice” CPUs, and Pensando “Vulcano” NICs, Helios connects thousands of accelerators through open ROCm software for trillion-parameter training.
CEO Lisa Su emphasized the shift: “We are entering the era of yotta-scale computing, driven by unprecedented growth in both training and inference.” Partners including OpenAI, Luma AI, Liquid AI, and World Labs demonstrated real-world AMD-powered breakthroughs across generative AI, drug discovery, and space exploration.
AMD CES 2026 AI Portfolio Overview
| Product Category | Key Innovation | Target Application |
|---|---|---|
| Helios Platform | 3 exaflops/rack, modular design | Yotta-scale AI infrastructure |
| Instinct MI440X | 8-GPU enterprise form factor | On-premises AI training/inference |
| Instinct MI500 | 1,000x MI300X performance (2027) | Next-gen AI compute leadership |
| Ryzen AI 400 | 60 TOPS NPU, ROCm support | Cloud-to-client AI scaling |
| Ryzen AI Max+ 392/388 | 128GB unified memory, 128B params | Local inference & creation |
MI400/MI500 Series: Performance Roadmap
The Instinct MI440X joins the recently announced MI430X (powering Oak Ridge’s Discovery supercomputer and France’s Alice Recoque exascale system) to complete the MI400 lineup. MI440X targets enterprise deployments with compact eight-GPU configurations integrating seamlessly into existing infrastructure for scalable training and fine-tuning.
Looking ahead, the MI500 Series (2027 launch) built on CDNA 6 architecture, 2nm process technology, and HBM4E memory promises up to 1,000x performance improvement over 2023’s MI300X—positioning AMD for sustained AI compute leadership.
Ryzen AI Expands Cloud-to-Edge Portfolio
The Ryzen AI 400 and PRO 400 Series deliver 60 TOPS NPU performance with full ROCm platform support, enabling seamless cloud-to-client AI scaling. First systems ship January 2026, with broader OEM availability in Q1.
For premium applications, Ryzen AI Max+ 392 and 388 processors support models up to 128-billion-parameters using 128GB unified memory—enabling advanced local inference, content creation workflows, and exceptional gaming in thin-and-light notebooks and small form factor desktops.
The Ryzen AI Halo Developer Platform brings these capabilities to a compact SFF desktop, delivering leadership tokens-per-second-per-dollar for AI development (Q2 2026 availability).
Physical AI & Embedded Solutions
AMD introduced Ryzen AI Embedded processors—the P100 and X100 Series designed for edge AI applications from automotive digital cockpits to humanoid robotics. These x86 processors deliver high-performance, efficient compute for constrained embedded systems powering physical AI.
The announcement reflects AMD’s comprehensive ecosystem approach, spanning data center yotta-scale infrastructure to edge devices requiring single-digit watt budgets.
Genesis Mission & Education Commitment
Lisa Su and White House Office of Science and Technology Policy Director Michael Kratsios discussed AMD’s role in the Genesis Mission—a public-private initiative securing U.S. AI leadership. Two AMD-powered supercomputers at Oak Ridge National Laboratory (Lux and Discovery) anchor this effort.
AMD pledged $150 million to expand AI education access, joining the White House’s initiative to bring hands-on AI opportunities to students. The commitment builds on AMD’s AI Robotics Hackathon partnership with Hack Club, which engaged over 15,000 student innovators.
As AI adoption accelerates across industries, AMD’s CES 2026 showcase demonstrates comprehensive technology leadership—from rack-scale infrastructure engineering to grassroots education investment, positioning AI as accessible infrastructure rather than exclusive capability.
