AMD unveils yotta-scale AI roadmap & USD $150m pledge

AMD has outlined an aggressive AI roadmap at CES 2026, unveiling a rack-scale platform for what it calls "yotta-scale" computing, new data centre and PC chips, and a USD $150 million commitment for AI education and community projects.

Chair and CEO Dr Lisa Su used the show's opening keynote in Las Vegas to position AMD's expanding AI portfolio across data centres, PCs and embedded systems, as well as in public sector supercomputing initiatives.

The company highlighted collaborations with a range of AI and science partners, including OpenAI, Luma AI, Liquid AI, World Labs, Blue Origin, Generative Bionics, AstraZeneca, Absci and Illumina, which are deploying AMD hardware in training and inference workloads.

"At CES, our partners joined us to show what's possible when the industry comes together to bring AI everywhere, for everyone," said Dr. Lisa Su, chair and CEO of AMD. "As AI adoption accelerates, we are entering the era of yotta-scale computing, driven by unprecedented growth in both training and inference. AMD is building the compute foundation for this next phase of AI through end-to-end technology leadership, open platforms, and deep co-innovation with partners across the ecosystem."

The 'Helios' blueprint

AMD set out its vision for what it describes as yotta-scale compute infrastructure. It cited projections of global compute capacity growing from about 100 zettaflops today to more than 10 yottaflops within five years.

The company introduced "Helios", a rack-scale platform that it pitches as a blueprint for future AI infrastructure. AMD said a single Helios rack will deliver up to 3 AI exaflops of performance.

The system uses AMD Instinct MI455X accelerators as its GPUs. It combines these with AMD EPYC "Venice" CPUs and AMD Pensando "Vulcano" network interface cards for scale-out networking.

AMD said Helios uses an open, modular rack design that can evolve across product generations. It uses the AMD ROCm software stack across the platform.

The company gave an early demonstration of Helios at CES. It also disclosed the full AMD Instinct MI400 Series accelerator line-up and provided a preview of its next-generation Instinct MI500 Series GPUs.

New Instinct accelerators

The latest product in the MI400 family is the AMD Instinct MI440X GPU. AMD is positioning MI440X for on-premises enterprise AI installations.

The MI440X targets training, fine-tuning and inference workloads. It uses an eight-GPU form factor that AMD said fits into existing data centre infrastructure.

The new chip builds on the recently launched AMD Instinct MI430X GPUs. AMD has designed MI430X for high-precision scientific and high-performance computing tasks, as well as what it labels sovereign AI workloads.

MI430X systems will underpin AI factory supercomputers at sites including Oak Ridge National Labouratory's Discovery machine and the Alice Recoque system, which AMD described as France's first exascale supercomputer.

AMD shared early specifications for its MI500 Series GPUs, which it plans to launch in 2027. The company said the family is on track to deliver up to a 1,000-times increase in AI performance compared with the AMD Instinct MI300X GPUs introduced in 2023, based on internal engineering projections.

The MI500 line will use the next generation of AMD's CDNA 6 GPU architecture. It will also use 2nm process technology and HBM4E memory.

AI PCs and developers

AMD moved to extend its presence in the fast-developing AI PC segment with new Ryzen AI platforms. The company framed the PC as a key interface for billions of users who will run AI applications locally and in the cloud.

The AMD Ryzen AI 400 Series and Ryzen AI PRO 400 Series platforms will target laptops and other client systems. AMD said both product lines deliver a 60 TOPS NPU and support the ROCm software environment for what it described as cloud-to-client AI scaling.

First devices using the new chips will ship in January 2026. Broader OEM availability is scheduled for the first quarter of the year.

AMD also introduced the Ryzen AI Max+ 392 and Ryzen AI Max+ 388 products. These systems support models with up to 128 billion parameters.

The platforms offer up to 128GB of unified memory. AMD is pitching them at advanced local inference, media creation workflows and gaming in premium thin-and-light notebooks and small form factor desktops.

For software developers, AMD announced the Ryzen AI Halo Developer Platform. This is a compact small form factor desktop system built on Ryzen AI Max+ processors.

AMD said Ryzen AI Halo will deliver strong tokens-per-second-per-dollar metrics for AI development, with availability expected in the second quarter of 2026.

AI at the edge

AMD extended its AI portfolio into embedded markets with Ryzen AI Embedded processors. These x86 parts will sit in edge systems where thermal and power constraints can be tight.

The initial Ryzen AI Embedded range includes the P100 and X100 Series processors. AMD said they target automotive digital cockpits, smart healthcare deployments and physical AI systems, including autonomous platforms and humanoid robotics.

The company said the new embedded products provide efficient AI compute in constrained designs.

Genesis Mission role

Su shared the stage with Michael Kratsios, Director of the White House Office of Science and Technology Policy, who outlined AMD's role in the US government's Genesis Mission.

The initiative brings together public and private organisations around AI research infrastructure, scientific discovery and economic competitiveness. Genesis includes two AMD-based AI supercomputers at Oak Ridge National Labouratory, named Lux and Discovery.

Kratsios described efforts by the White House to encourage industry pledges that expand access to AI education. These pledges focus on more hands-on opportunities for students.

In support of that effort, AMD announced a commitment of USD $150 million that it will direct into classrooms and communities. The company said the funding will expand AI education and practical learning programmes. This has already included 15,000 student innovators who participated in the AMD AI Robotics Hackathon in partnership with Hack Club.