Yuan Yuan, Vice President of Huawei and President of the Huawei Data Storage Product Line, delivered a keynote titled “Data Awakening, Infra Evolving”, unveiling the full-stack data infrastructure solution of AI data centers (DCs) to accelerate industry intelligence.
Today, AI is transforming the way enterprises operate. Agents are becoming a key driver of new digital productivity and are evolving into digital employees of enterprises. The rapid expansion of AI applications is driving a surge in enterprise token consumption at an unprecedented scale. Yuan pointed out that enterprises need to rapidly evolve their existing IT architecture into AI DC data infrastructure to accelerate AI adoption. Such infrastructure must be systematically planned and built around the following pillars: data lakes, AI data platforms, compute power, models, agent frameworks, and data resilience.
Against this backdrop, Huawei unveiled the full-stack data infrastructure solution of AI DCs to accelerate AI DC construction and large-scale AI adoption for enterprises.
AI data lake
High-density OceanStor Pacific Scale-Out Storage delivers 11 PB capacity in a 2 U space, enabling massive data storage at optimal total cost of ownership (TCO). DME Omni-Dataverse, the company’s unified data space solution, supports multimodal, cross-site, and real-time data import, global data visibility and manageability, and retrieval from hundreds of billions of 1,000-dimension vectors in seconds, achieving high-quality data aggregation and supply.
Knowledge and memory platform
- For ultra-scale inference clusters, Huawei introduces the industry’s first Context Memory Storage (CMS) that supports heterogeneous computing power. CMS supports key-value (KV) semantic direct pass or uses the dedicated data processing unit (DPU) for semantic offloading. It can expand into a PB-scale shared KV cache pool and reduce the time to first token (TTFT) by 90%.
- For enterprises’ AI inference scenarios, the industry-first 3+1 AI data platform developed by Huawei integrates KV cache acceleration, a knowledge base with over 95% retrieval accuracy, and an ever-evolving memory bank. In addition, Huawei’s Unified Cache Manager (UCM) enables seamless scheduling and management, improving inference accuracy by 30%.
Model engineering and resource scheduling
Huawei’s ModelEngine delivers out-of-the-box model usability and model gateway capabilities, enabling zero-code adaptation to new models and one-click model deployment. Furthermore, powered by fine-grained compute resource partitioning and intelligent scheduling, ModelEngine achieves an up to 1:10 ratio of xPU partitioning to make one xPU serve multiple purposes, improving resource efficiency.
Agent framework
Huawei’s ModelEngine Nexent agent platform directly generates agents via natural language–based interaction, simplifying development and cutting rollout time by 80%. Through automatic optimization of skills, prompts, and memory, Nexent ensures that agents grow smarter through continuous evolution.
Data resilience platform
To address potential data security risks across agents, models, platforms, and infrastructure, enterprises must build an end-to-end data protection solution that prevents tool misuse, data poisoning, tampering, and ransomware attacks, achieving all-round protection of AI data assets.
“AI is unlocking new opportunities for the IT industry”, Yuan said, “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.”
