CES 2026: AMD Reveals yotta-scale AI platform and Next-Gen GPU roadmap
At CES 2026, AMD CEO Dr. Lisa Su announced the company's vision for " yotta-scale computing "—a massive leap beyond today's data center capabilities. The centerpiece is "Helios," a rack-scale platform that packs 3 AI exaflops into a single rack using new Instinct MI455X GPUs and EPYC "Venice" CPUs. AMD says this infrastructure is built specifically for training trillion-parameter AI models , with partners like OpenAI and Blue Origin already testing the technology.
The company also introduced the Instinct MI440X GPU for enterprise customers. The MI440X GPU is targeted towards customers that want on-premises AI deployments without massive infrastructure investments. AMD says its compact eight-GPU design fits existing setups while handling training, fine-tuning, and inference workloads. Furthermore, AMD also previewed its MI500 Series GPUs launching in 2027 with 2nm process technology and HBM4E memory—claiming up to 1,000x better performance than the MI300X chips from 2023.
Consumer AI gets more accessible
On the PC side, AMD's new Ryzen AI 400 Series processors ship this month with 60 TOPS neural processing units and full ROCm software support. The more powerful Ryzen AI Max+ variants stand out with 128GB of unified memory, letting users run 128-billion-parameter models directly on their laptops or small desktops—no cloud connection required. These chips target creators and developers who need serious local AI horsepower in portable machines.
AMD's Ryzen AI Halo Developer Platform arrives in Q2 2026, giving programmers a dedicated desktop tool for AI development work at what the company calls "leadership tokens-per-second-per-dollar."
Edge devices go intelligent
The new Ryzen AI Embedded processors bring AI processing to cars, medical devices, and robots. Split into P100 and X100 Series, these chips are designed for tight spaces and power constraints—think automotive dashboards, healthcare monitors, and autonomous systems that need to make split-second decisions locally.
The company also introduced the Instinct MI440X GPU for enterprise customers. The MI440X GPU is targeted towards customers that want on-premises AI deployments without massive infrastructure investments. AMD says its compact eight-GPU design fits existing setups while handling training, fine-tuning, and inference workloads. Furthermore, AMD also previewed its MI500 Series GPUs launching in 2027 with 2nm process technology and HBM4E memory—claiming up to 1,000x better performance than the MI300X chips from 2023.
Consumer AI gets more accessible
On the PC side, AMD's new Ryzen AI 400 Series processors ship this month with 60 TOPS neural processing units and full ROCm software support. The more powerful Ryzen AI Max+ variants stand out with 128GB of unified memory, letting users run 128-billion-parameter models directly on their laptops or small desktops—no cloud connection required. These chips target creators and developers who need serious local AI horsepower in portable machines.
AMD's Ryzen AI Halo Developer Platform arrives in Q2 2026, giving programmers a dedicated desktop tool for AI development work at what the company calls "leadership tokens-per-second-per-dollar."
Edge devices go intelligent
The new Ryzen AI Embedded processors bring AI processing to cars, medical devices, and robots. Split into P100 and X100 Series, these chips are designed for tight spaces and power constraints—think automotive dashboards, healthcare monitors, and autonomous systems that need to make split-second decisions locally.
Next Story