AMD announces MI350P PCIe AI accelerator card with 144GB of HBM3E — roughly 40% faster in FP16 and FP8 theoretical compute compared to Nvidia's H200 NVL competitor

AMD has unveiled its latest addition to the MI350 series — the MI350P PCIe AI accelerator card, designed to offer a drop-in upgrade solution for existing air-cooled servers. Built on TSMC's 3nm and 6n... Read original →

AMD’s MI350P isn’t just a new accelerator—it’s a strategic wedge into NVIDIA’s enterprise stronghold. By combining TSMC’s 3nm/6nm nodes with 144GB HBM3E, it forces memory suppliers to accelerate HBM output and compels server OEMs to redesign PCIe thermal/power specs. Amid tightening U.S. export controls, its 450W compatibility mode slashes customer migration costs and mitigates supply-chain risks tied to NVIDIA’s restricted SKUs. NVIDIA may now rush a PCIe B200 variant, even at the cost of Blackwell’s integration density, to retain data-center loyalty. Over the next 18 months, as LLM inference shifts toward FP8/MXFP4, CDNA4’s precision-optimized architecture will dominate edge AI deployments—making HBM3E capacity the next flashpoint in U.S.-Japan-South Korea tech diplomacy.