Huawei Launches Unified AI Data Platform

The platform's core is a "3+1" architecture, featuring three key technologies and one managing software component to tackle common AI bottlenecks like slow response times and model hallucinations. This design specifically targets the inference stage of AI, which is critical for model adoption but often overlooked in favor of training. The system integrates a knowledge base for high-accuracy retrieval, a KV cache for faster inference, and a memory bank to allow models to learn from experience. A key software component, the Unified Cache Manager (UCM), intelligently schedules data across different memory and storage tiers. Huawei claims UCM can reduce AI inference latency by up to 90% and increase system throughput significantly by optimizing the use of various memory types, from ultra-fast HBM to standard SSDs. This software-based approach is also positioned as a way to mitigate reliance on foreign-made high-bandwidth memory (HBM) chips. For handling the large datasets required by AI, especially multimodal resources like video, the platform uses a technique of multimodal lossless parsing and token-level encoding. This is designed to convert raw data, including video, into a high-accuracy knowledge base, with Huawei claiming a retrieval accuracy of over 95%. This capability is crucial for newsrooms dealing with vast archives of video content. The hardware foundation for the platform's appliance mode is the OceanStor A800, a high-performance all-flash storage system. A single cluster can scale out to 512 controllers, reaching exabyte-level capacity to handle massive data volumes. Each controller enclosure supports 1,024 GB of memory and network speeds up to 200 Gb/s, providing the high bandwidth needed for loading large training sets and rapid checkpoint resumption. This launch places Huawei in direct competition with the integrated data and AI platforms from major cloud providers like Amazon Web Services, Microsoft Azure, and Google Cloud. While AWS SageMaker, Azure AI, and Google's Vertex AI offer comprehensive toolsets for the entire machine learning lifecycle, Huawei's strategy focuses on a tightly integrated hardware and software solution to optimize the data foundation specifically for AI inference performance and efficiency. For newsrooms, the challenges of adopting AI often revolve around sourcing, accuracy, and the risk of "AI hallucinations". A platform that ensures high-accuracy knowledge retrieval from an organization's own verified data, including video archives, can directly address concerns about publishing content based on flawed or non-traceable sources. As journalists increasingly use AI for research and content generation, the underlying data infrastructure's ability to provide reliable and fast information becomes a critical component of maintaining editorial standards.