AMD (NASDAQ: AMD) unveiled a broad set of new AI computing products at its Advancing AI 2026 event in San Francisco, including its first rack-scale AI solution, the AMD Helios, which the company says is now in production.

AMD Helios integrates 72 Instinct MI455X GPUs and 18 sixth-generation EPYC "Venice" CPUs connected by AMD Pensando networking. According to AMD internal estimates from July 2026, the system delivers up to 30% more inference tokens per dollar than what the company describes as the leading competitive solution, based on comparison with an Nvidia Vera Rubin NVL72 rack.

Companies including OpenAI, Anthropic, Meta, Microsoft, and Oracle are listed as Helios customers. Anthropic announced a partnership to deploy up to two gigawatts of MI455X GPUs in Helios racks, alongside a multiyear engineering collaboration using Anthropic's Claude model to assist AMD software development. OpenAI expects to bring Helios online in the fourth quarter of 2026, with broader deployment planned through 2027.

AMD also launched sixth-generation EPYC server CPUs and the Instinct MI400 Series GPUs, including the MI455X and the MI430X, the latter aimed at high-performance computing and sovereign AI applications. The company additionally introduced the Instinct MI350P GPU, which AMD claims delivers up to 4.2 times more tokens per second per dollar than a comparable Nvidia H200 NVL GPU, based on internal testing.

For edge and embedded applications, AMD introduced the Ryzen AI Embedded X100 Series processors, the Kria AI system-on-modules, and the Kria AI Robotics Developer Platform.

AMD outlined a product roadmap extending to 2030, including Instinct MI500 Series GPUs in 2027, MI600 Series in 2028, and EPYC server CPUs based on the "Zen 7" architecture in 2028 and "Zen 8" in 2030. The company said it estimates its total addressable market will reach approximately $2 trillion by 2030, though it noted this figure involves forward-looking assumptions subject to risks.