AMD has unveiled a comprehensive portfolio of next-generation AI infrastructure to address the era of inference and agentic AI. Spanning the entire spectrum from RackScale solutions to robotics platforms, this announcement also includes adoption plans from major big tech companies.

AMD announced this on the 24th at the 'Advancing AI 2026' event held in San Francisco on the 23rd (U.S. local time).

Lisa Su (Dr. CEO Lisa Su stated, “The next stage of AI is expanding to frontier models, agents, and physical AI,” adding, “We will support our customers’ AI expansion from the data center to the edge through an open platform.”

AMD projected that the Total Addressable Market (TAM) will reach approximately $2 trillion by 2030 due to the expansion of AI demand.

The AMD Helios rackscale solution, the centerpiece of this announcement, has now entered mass production.

It is a configuration that integrates 72 AMD Instinct MI455X GPUs, 18 6th generation AMD EPYC 'Venice' CPUs, AMD Pensando frontend, scale-up and scale-out networking, and AMD ROCm software.

AMD explained that it processes up to 30% more AI inference tokens per dollar compared to competing solutions.

OpenAI, Anthropic, Meta, Microsoft, Oracle, and others are pursuing adoption, and it is scheduled to be supplied through major OEMs such as Bull, HPE, Lenovo, and Supermicro.

The AMD Instinct MI400 series was also unveiled.

It was announced that the MI455X GPU provides up to 34 times higher token throughput compared to the previous generation MI355X.

The AMD Instinct MI430X, designed for high-precision computing environments, supports FP64 performance of up to 288 TFLOPS and is scheduled to be used in the construction of next-generation exascale supercomputers in the U.S. and Europe.

The AMD Instinct MI350P GPU is designed to be easily added to existing AI infrastructure, offering up to 4.2 times higher payouts per second compared to competitors.AMD stated that it provides high processing performance.

In the field of robotics, AMD introduced the new Kria AI solution.

It consists of a CREA AI System-on-Module (SOM) equipped with AMD Ryzen AI Embedded X100 series processors and an AMD CREA AI Robotics Development Platform that integrates CPU, GPU, NPU, and FPGA into one.

In terms of software, they unveiled ROCm.ai, an AI-based development environment.

AI coding agents such as Claude, Codex, and Cursor support the use of AMD platforms and ROCm at a native level, and major open source frameworks such as PyTorch, Hugging Face, vLLM, and SGLang already support MI455X.

AMD also presented a roadmap extending to 2030.

The plan is to launch the following in order: next-generation EPYC CPUs based on 'Zen 7' (2028), 'Ravenna' CPUs based on 'Zen 8' (2030), AMD Instinct MI500 series (2027), AMD Instinct MI600 series (2028), and AMD Helios 500 and 600 rackscale solutions.