AMD Unveils 2nm AI Chip at CES 2026
At CES 2026, AMD Chairman and CEO Lisa Su officially unveiled the company's latest technological advancements on January 5, Pacific Time, emphasizing that artificial intelligence (AI) is evolving from a value-added tool into the next generation of computing infrastructure. AMD's latest generation of data center chips has successfully entered the 2nm era. Leveraging the co-designed Helios rack-level platform, it meets the extreme demands of AI training and inference.
Lisa Su noted that AI is AMD's core strategy. Global computing power demand is projected to surge from the current Zetta level to 10 YottaFLOPS over the next five years—a ten-thousandfold increase from 2022 levels. This generational leap relies not only on system architecture innovation but also on pushing manufacturing process boundaries to their limits.
AMD's next-generation AI accelerator, the Instinct MI455, utilizes TSMC's most advanced manufacturing and packaging technologies—such as CoWoS and 3D Chiplet—enabling a single chip to contain 32 billion transistors. Simultaneously, the next-generation EPYC server CPU, codenamed Venice, marks AMD's industry-first adoption of 2nm process technology.
Based on the latest Zen 6 architecture, the Venice CPU integrates up to 256 cores per processor. Expected to complete tape-out in April 2025 and already in mass production, it signifies TSMC's official deployment of GAA nanowire technology into the core computing domain of AI servers.
AMD's Helios rack-scale platform, specifically designed for AI hyperscale data centers, integrates MI555 GPUs, Venice CPUs, and Pensando networking chips per rack. Through high-speed Ethernet and the HyperAccelerator interconnect protocol, it enables 72 GPUs to operate as a single computational unit. Lisa Su revealed that Helios is an open-standard AI reference platform that has evolved into a fully liquid-cooled architecture.
In the consumer segment, AMD has launched multiple product lines to compete with NVIDIA, including Ryzen AI Halo, which directly challenges NVIDIA's DGX Spark for workstation market share. For AI PCs, AMD introduced the new Ryzen AI 400 series processors, featuring up to 12-core Zen 5 CPUs, RDNA GPUs, and XDNA 2 NPUs, delivering AI compute power of up to 60 TOPS.
Additionally, Lisa Su partnered with multiple AI startups, including Fei-Fei Li's World Labs, to showcase the 3D generative model “Marble,” which rapidly converts sparse imagery into three-dimensional worlds.