AMD introduces Instinct MI350P for drop-in enterprise AI hardware

AMD launched the Instinct MI350P PCIe accelerator for standard servers and existing racks, aimed at on-premises inference and RAG workloads.

The card supports multiple precision formats, up to eight cards per system, and AMD claims up to 2,299 TFLOPS, 144 GB of HBM3E, and 4 TB/s bandwidth.

AMD is emphasizing interoperability and software support, including an open-source reference stack, to simplify deployment and cut licensing costs.

The launch offers enterprises a lower-friction way to add AI capacity on premises while improving data control and potentially lowering infrastructure costs.