Big bet! OpenAI plans to cooperate with AMD to develop MI500 series AI chips

Kuai Technology reported on July 24 that according to media reports, OpenAI is accelerating the deployment of high-performance AI computing power infrastructure. The person in charge of OpenAI's infrastructure recently revealed that OpenAI began deploying AMD Helios GPU racks three months ago and plans to cooperate with AMD to develop the next generation MI500 series AI chips and their follow-up products. It is reported that the MI500 series is expected to be officially released in 2027. It will use TSMC’s more advanced 2nm process technology and be equipped with the new CDNA 6 architecture and HBM4E memory. Its memory bandwidth will exceed the 19.6 TB/s achieved by the MI400 series based on the HBM4 solution, further improving the efficiency of large-scale AI training and inference. AMD CEO Su Zifeng said at this year's CES 2026 that from MI300X in 2023 to MI500 in 2027, AMD aims to achieve a thousand-fold jump in AI performance within four years. This roadmap demonstrates AMD’s continued accelerating technological ambitions in the field of AI accelerators. At the same time, AMD is highly optimistic about the market prospects. By 2030, the AI ​​accelerator market will reach US$1.4 trillion, the data center CPU market will reach US$220 billion, and the overall computing market is expected to exceed US$2 trillion.