Huawei stacks five infrastructure layers to handle the token loads that enterprise AI agents are generating
Huawei stacks five infrastructure layers to handle the token loads that enterprise AI agents are generating

At the Huawei Innovative Data Infrastructure Forum 2026, Yuan Yuan, Vice President of Huawei and President of the Huawei Data Storage Product Line, unveiled a full-stack AI data centre solution built around five pillars: data lakes, an AI data platform, compute and model infrastructure, an agent framework, and data resilience.

The headline storage number is 11 petabytes in a 2U chassis, delivered by the OceanStor Pacific Scale-Out Storage. Alongside it, Huawei's DME Omni-Dataverse handles multimodal and cross-site data ingestion and claims retrieval from hundreds of billions of thousand-dimension vectors in seconds.

The more operationally significant announcement is Context Memory Storage, which Huawei describes as the industry's first KV cache system supporting heterogeneous computing. CMS offloads KV semantics to a dedicated data processing unit and can scale to a petabyte-class shared cache pool; the company says it cuts time-to-first-token by 90%. A separate 3+1 AI data platform — integrating KV cache acceleration, a knowledge base and an evolving memory store — is said to improve inference accuracy by 30%, with retrieval accuracy above 95%.

On the compute side, Huawei's ModelEngine supports zero-code adaptation to new models and one-click deployment. Through fine-grained resource partitioning, it achieves a 1:10 xPU partitioning ratio, letting a single accelerator serve multiple workloads simultaneously. The Nexent agent platform generates agents from natural language input and cuts rollout time by 80%, according to the company.

Rounding out the stack is an end-to-end data resilience layer addressing threats specific to agentic AI: tool misuse, data poisoning, tampering and ransomware targeting model weights and training data.

"AI is unlocking new opportunities for the IT industry," Yuan said at the Paris forum. "The next chapter of AI is data. Committed to technological innovation in data storage, Huawei will accumulate the experience of industrial AI adoption, and work closely with the entire industry to help customers accelerate their journey into the intelligent era."

The next chapter of AI is data. Committed to technological innovation in data storage, Huawei will accumulate the experience of industrial AI adoption, and work closely with the entire industry to help customers accelerate their journey into the intelligent era.

Yuan Yuan (Vice President, Huawei; President, Huawei Data Storage Product Line)

The forum took place a week after Microsoft Build 2026 and coincides with a broader push by infrastructure vendors to reframe their storage and compute portfolios around agent workloads. Huawei did not announce specific customer deployments or pricing at the event.

More News