AMD Launches Instinct MI350P AI Accelerator with 144 GB HBM3E and PCIe 5.0 Support

• May 8, 2026

AMD has announced the Instinct MI350P, a new graphics processing unit designed specifically to accelerate artificial intelligence workloads. Unlike some server accelerators that require specialized Open Accelerator Module (OAM) platforms, the MI350P is built as a traditional PCIe expansion card compatible with standard server configurations.

Innovations in AI Hardware for Air-Cooled Servers

The Instinct MI350P targets AI inference applications, enabling the deployment of machine learning models without the need for complex or proprietary server architectures. Its form factor and interface support PCIe 5.0 x16, allowing for high bandwidth connectivity within data center environments that utilize air cooling solutions.

Key to the MI350P’s performance is its inclusion of 144 gigabytes of HBM3E (High Bandwidth Memory), offering substantial memory capacity and bandwidth for demanding AI computations. The accelerator’s power consumption is rated at 600 watts, reflecting its high-performance capabilities tailored for enterprise-grade workloads.

This new addition to AMD’s Instinct lineup is designed to meet the needs of organizations seeking powerful AI inference acceleration in existing server infrastructures. By providing a PCIe-based solution, AMD facilitates easier integration into current data center setups, potentially reducing deployment complexity compared to OAM-dependent alternatives.

The MI350P represents AMD’s continuing push to expand its presence in the AI hardware sector, emphasizing scalability, versatility, and compatibility with prevailing server environments. While detailed specifications beyond memory capacity, power consumption, and interface were not disclosed, the MI350P aligns with broader industry trends toward PCIe 5.0 adoption and enhanced memory architectures like HBM3E.

As AI workloads become increasingly central to enterprise computing, the availability of potent, easily deployable accelerators like the Instinct MI350P could play a vital role in supporting next-generation inference demands.

AMD introduces the Instinct MI350P, a PCIe AI accelerator featuring 144 GB of HBM3E memory and designed for air-cooled server deployments.