Huawei Vice President Yuan Yuan: 70% of Computing Power Idle, Data Development Rate 28.4%, AI Development Requires Enhanced Storage Support

Smart East West reported on October 30 that yesterday, at the 2026 Data Storage Industry Conference, the Data Storage Special Committee released the "Data Storage Industry Development Research Report (2026)" and the "Data Storage Industry and Technology Panorama (2026 Edition)".

The report predicts that by the end of 2026, the global storage industry will exceed US$1.7 trillion (about RMB 11.3 trillion), and the growth focus of the storage industry will shift from simply expanding capacity to enterprise SSDs, AI data centers, and data infrastructure for inference.

At the conference, Yuan Yuan, Vice President of Huawei and President of the Data Storage Product Line, argued that GPU scale does not equal effective computing power. AI development needs to improve storage's ability to support computing power, strengthen data governance and reuse, enable limited computing power to deliver higher output, and at the same time turn industry data into an advantage for AI applications.

He said: "Data is the fuel of AI. Data determines the height of models and also determines the accuracy of inference." In his view, storage must, on the one hand, improve computing power utilization and reduce repeated computation; on the other hand, it must clean and organize raw data into corpora, knowledge, and memory that can be directly invoked, and truly use them for model training, inference, and agent operation.

1. Global storage industry scale expected to exceed RMB 11.3 trillion, storage accelerating its shift toward AI infrastructure

The "Data Storage Industry Development Research Report" analyzes the development trends and technology paths of the storage industry in the AI era. At the conference, Sun Gang, Secretary-General of the Data Storage Professional Committee of the China Electronics Standardization Association, interpreted the core content of the report.

The report predicts that by the end of 2026, the global storage industry will exceed US$1.7 trillion (about RMB 11.3 trillion). As AI inference and agent applications accelerate deployment, the growth focus of the storage industry is shifting from simply expanding capacity to enterprise SSDs, AI data centers, and data infrastructure serving model inference.

The report proposes that AI is driving storage systems from simply preserving data to simultaneously managing data, knowledge, and memory. In terms of industry models, the report judges that storage is moving from "building data lakes" to "using data lakes." In the past, enterprises focused more on centrally storing data; next, whether data can be retrieved, analyzed, and reused, and truly serve the business, will become a more important measure. At the same time, storage delivery will gradually shift from one-time equipment procurement to on-demand service provision, and data protection must also shift from simple backup to ensuring rapid recovery after system failures, minimizing impact on business operations.

2. 70% of computing power has not yet created effective value; storage needs to support computing efficiency and data governance

Yuan Yuan said that the average GPU utilization rate of current national intelligent computing centers is about 30%, and about 70% of computing power has not yet created effective value; the development rate of data resources that enterprises can truly use is about 28.4%, and Chinese corpora account for about 4.8% of the global total.

In Yuan Yuan's view, storage must first solve the problems of data supply, low-cost retention, and reuse and sharing, reducing repeated computation and ineffective consumption of computing power. In the inference stage, KV Cache requires storage support with low latency, high bandwidth, and lower cost; in the training stage, as model parameter scale continues to expand, Checkpoint efficiency and training cluster reliability will become key bottlenecks.

Data governance corresponds to three types of needs: corpora, knowledge, and memory. Data must be cleaned, transformed, and labeled before it can become high-quality corpora; industry experience and experimental data need to be further transformed into knowledge; and the context and experience generated during agent operation need to be turned into manageable long-term memory through storage.

Around the different stages of AI applications, Yuan Yuan proposed two paths: "All in One" and "One for All." "All in One" targets the AI user side in industries such as finance, manufacturing, and healthcare, using a unified AI data platform to carry knowledge accumulation, memory management, KV Cache acceleration, and resource scheduling; "One for All" targets the AI production side, such as model vendors, internet companies, and national-level AI training bases, building dedicated data infrastructure for ultra-large-scale training and inference.

As AI applications expand from model training to inference and agent operation, the competitive focus of storage vendors will also shift from single hardware performance to software-hardware collaboration, data governance capabilities, and scenario adaptation capabilities. Whoever can reduce the cost of data processing and computing power invocation will have a better chance of truly bringing AI capabilities into industry applications.