The focus of debate over AI infrastructure is shifting from computing to data. WD on Wednesday released a white paper titled "Design for Scale: The Enduring Role of HDDs in the AI Era" that includes results from an IDC global study it sponsored. It said data continues to accumulate after computing tasks end, becoming a new variable in companies' AI expansion.

WD said 94.7 percent of surveyed companies and institutions stored more data in the 12 months after adopting AI and generative AI. Some 61 percent said data increased by at least 25 percent, and 74 percent expected it to rise by at least 25 percent within the next 3 years. Another 85.4 percent said their data lakes had grown. Some 59.4 percent cited AI-generated data such as synthetic data, inference results and model logs as factors driving the increase.

How companies handle data is also changing. About 95 percent said the value of their data had risen, and 74.3 percent said they kept data longer than before. Some 75.9 percent said more cases were emerging in which cold data that had not been used for some time is moved back online for AI workloads. Another 96 percent said faster access to archived data is needed to support AI inference and retrieval-augmented generation (RAG) applications.

The survey found that 74.6 percent of all enterprise data is stored across warm, cool and cold storage tiers. More than 60 percent of data lakes are filled with cold data or data accessed infrequently. Some 98.2 percent cited total cost of ownership per terabyte as a key factor in storage decision-making.

Irving Tan (어빙 탄), WD CEO, said discussions on AI infrastructure have centred on computing in recent years, but AI ultimately rests on data. "Companies are generating more data and preserving it for longer, while also looking for ways to create new value from the data they already have," he said.

"Computing demand will change over time, but demand will continue to grow for storing and managing data at scale and accessing it when needed," he added. "This data foundation will be a key factor in determining AI scalability."