AMD

Advanced Micro Devices (AMD) is a US chip designer, incorporated in Delaware on May 1, 1969, with its executive offices in Santa Clara, California.¹ It describes its beginnings as a Silicon Valley start-up with dozens of employees.² Besides its EPYC server CPUs, it makes the Instinct line of data-center GPUs, which its 2025 annual report lists as the MI200, MI300, MI325 and MI350 series, built on its CDNA architecture for AI training, inference and scientific computing.¹ The same report says large hyperscale customers, OEMs and ODMs deployed its MI350 Series GPUs in 2025, the year it launched them.¹

In AI Accelerators With the Most HBM per Chip (October 2026), AMD holds three of the top six places: the Instinct MI350X and MI355X share #1 with NVIDIA’s Rubin at 288 GB of HBM3E, and the MI325X is #6 at 256 GB. AMD’s MI355X page lists 288 GB of HBM3E at 8 TB/s, a June 12, 2025 launch date, a typical board power of 1,400 W, and 5 PFLOPS of OCP-FP8 dense, or 10.1 PFLOPS with structured sparsity.³ The MI325X, launched October 10, 2024, has 256 GB at 6 TB/s.⁴ It is also the AMD chip that the January 15, 2026 US export rule names, next to NVIDIA’s H200, as an example of what can now get case-by-case license review for China and Macau, as How US AI Chip Export Controls Work explains.⁵

AMD’s next generation sits on that ranking’s announced-but-not-shipping list. It launched the Instinct MI455X on July 23, 2026, with 432 GB of HBM4 in 12 stacks at 23.3 TB/s on its CDNA 5 architecture.⁶ AMD says its Helios racks, each built on 72 MI455X GPUs, are now in production, and that OpenAI expects to bring Helios online beginning in the fourth quarter of 2026.⁷ Its 2025 annual report also records a multi-generation agreement with OpenAI to deploy six gigawatts of Instinct GPUs.¹