Analysis: The growth rate of AI data has outpaced projections for computing power, making storage a new bottleneck for AI infrastructure.

Western Digital’s latest analysis notes that as AI applications scale rapidly, AI data center construction is shifting from a pure GPU computing power race to a competition in data storage capacity, with storage planning becoming a core component of AI infrastructure. Citing IDC forecasts, global annual new data volume will reach 718 ZB by 2030. Data generated by AI systems does not disappear once computing tasks end; training data, model checkpoints, embedding vectors, inference logs, prompts, output results, and evaluation data will all accumulate continuously. Western Digital points out that many current AI infrastructure plans overemphasize GPU utilization while neglecting data accumulation throughout the AI lifecycle. Data generated during training and inference will become critical assets for model iteration, quality assessment, and compliance audits, with storage costs directly impacting the long-term operational efficiency of AI systems. As data scales to PB and even EB levels, a single storage architecture can no longer meet demands. Enterprises need to adopt a tiered storage strategy: using high-performance flash memory for training and real-time inference, and high-capacity HDDs and object storage for long-term data archiving, historical records, and low-frequency access scenarios. The analysis concludes that the key metrics for future AI infrastructure competition will not only be the number of GPUs, but also per-PB data storage cost, energy consumption, recovery efficiency, and data lifecycle management capabilities. If enterprises still treat storage as an afterthought to computing, they may face issues such as out-of-control data costs and reduced model iteration efficiency.