At the IDI Forum 2026 in Paris, Huawei presented a comprehensive full-stack data infrastructure solution for AI data centers. The announcement covers data lakes, knowledge platforms, agent frameworks, and data security.
The OceanStor Pacific Scale-Out Storage offers 11 PB of storage capacity in a 2U chassis. Additionally, the company is introducing DME Omni-Dataverse, a unified data space solution that supports multimodal and cross-site data import and enables real-time search across hundreds of billions of vectors.
For ultra-scale inference clusters, Huawei is introducing Context Memory Storage (CMS), which the company describes as the first in the market to support heterogeneous computing power. CMS can scale to a petabyte-scale shared KV cache and reduces time to first token (TTFT) by 90 percent. The 3+1 AI data platform combines KV-cache acceleration, a knowledge base with over 95 percent retrieval accuracy, and a memory system. The Unified Cache Manager (UCM) handles scheduling and improves inference accuracy by 30 percent.
ModelEngine and agent framework
ModelEngine enables zero-code adaptation to new models and one-click deployment. Through fine-grained resource partitioning, a single xPU achieves a 1:10 ratio, allowing a single processor to handle multiple tasks simultaneously.
The Nexent agent platform generates agents through natural language interaction, reducing deployment time by 80 percent. Through automatic optimization of skills, prompts, and memory, agents continuously become smarter. Huawei thus describes agents as digital employees that form the core of new digital productivity.
Finally, Huawei focuses on data resilience: end-to-end protection against tool misuse, data poisoning, ransomware, and tampering must fully secure AI data assets.
Tip: Huawei lays the foundation for an intelligent and sustainable European era