Huawei Launches OceanStor M900 to Accelerate AI Inference

Huawei introduced the OceanStor M900 Context Memory Storage at HUAWEI CONNECT 2026 to accelerate AI inference within hyperscale data centers. Deputy Chairman David Wang announced the product as a solution to memory capacity bottlenecks caused by ultra-long context windows and models with 10 trillion parameters.

The system utilizes a UnifiedBus network to establish a PB-scale, global multi-tier KV cache, which expands storage capacity from on-chip memory and DRAM to SSDs. By integrating the CPU, network controller unit, and NAND controller unit, the architecture reduces access latency from milliseconds to 60 microseconds and delivers an aggregate access bandwidth of 40 TB/s.

To address hardware longevity, the OceanStor M900 employs KV-aware adaptive storage technology. This feature extends SSD endurance by 16 times, which the company intends will lower the long-term operational costs associated with large-scale AI inference infrastructure.