Under CEO Lisa Su’s leadership, AMD has transformed from a struggling player in the datacenter market into a major force, driven by the success of its Zen microarchitecture and Epyc server processors. Now, as the AI revolution accelerates, AMD is aiming to challenge Nvidia’s dominance in the GPU and AI infrastructure space. At the 2026 Consumer Electronics Show in Las Vegas, Su delivered a keynote that laid out AMD’s vision for the future of AI compute—specifically, systems capable of operating at yottascale. Su emphasized that keeping pace with AI’s exponential growth requires more than raw performance. It demands an integrated ecosystem of CPUs, GPUs, NPUs, networking, and scalable system architecture. To meet this challenge, AMD unveiled Helios, its next-generation rack-scale platform designed for the era of massive AI workloads. Helios is a double-wide system based on the Open Compute Project’s Open Rack Wide standard, weighing nearly 7,000 pounds—comparable to more than two compact cars. The platform will be powered by AMD’s latest Instinct MI455X AI GPU and the upcoming Epyc “Venice” server CPUs, both of which Su showcased. The MI455X, built on a mix of 2nm and 3nm processes, features 320 billion transistors—70% more than its predecessor, the MI355X—and 432 GB of HBM4 memory, enabled by AMD’s 3D chip-stacking technology. It delivers up to 10 times the inference performance of the MI355X, a significant leap in efficiency and speed. Looking ahead, AMD’s 2027 MI500 series, built on the next-generation CDNA 6 architecture and manufactured on 2nm process technology, will use HBM4E memory. Su announced that this will result in a 1,000-fold increase in AI performance over four years—demonstrating AMD’s aggressive roadmap. Helios is designed for massive scale. Each liquid-cooled tray contains four MI455X GPUs, a Venice CPU, and AMD’s Pensando “Vulcano” and “Salina” networking chips, which provide 800 Gb/sec Ethernet with ultra-low latency. A single Helios rack delivers up to 2.9 exaflops of performance, over 18,000 CDNA5 GPU compute units, and more than 4,600 Zen 6 CPU cores. It also includes 31 terabytes of HBM4 memory and a staggering 260 terabytes-per-second of scale-up bandwidth, along with 43 terabytes-per-second of scale-out bandwidth. This infrastructure is built for seamless expansion. Su noted that tens of thousands of Helios racks can be deployed across datacenters, forming a unified, high-performance AI platform. AMD’s push comes amid fierce competition. Nvidia is also unveiling new hardware at CES, including its Rubin GPU accelerators, Vera Arm-based CPUs, NVLink memory fabric, and Spectrum Ethernet interconnects. The race to deliver the most powerful, efficient, and scalable AI systems is heating up. With its bold vision, Helios, and a clear multi-year roadmap, AMD is positioning itself not just as a contender, but as a key architect of the yottascale AI future. The battle for AI supremacy is now in full swing.