Huawei unveils OceanStor M900 to accelerate AI inference

At HUAWEI CONNECT 2026, Huawei Deputy Chairman David Wang introduced the OceanStor M900 Context Memory Storage, a product designed to accelerate AI inference within hyperscale data centers.

As AI models scale toward 10 trillion parameters and support context windows exceeding one million tokens, traditional on-chip memory and DRAM face capacity and cost-effectiveness limitations. The OceanStor M900 addresses these bottlenecks by providing a fully shared memory space for SuperPoDs, offering PB-scale capacity and TB/s-level performance.

The system utilizes the UnifiedBus network to create a multi-tier KV cache with one-hop connections, coordinating on-chip memory, DRAM, and SSDs. This approach represents a shift in AI infrastructure from a compute-centric model toward a collaborative architecture involving compute, network, and storage to support the growing demands of agentic AI.