AI drives enterprise storage demand as data piles up

WD has published IDC research showing that artificial intelligence is increasing enterprise data storage requirements, with most organisations expecting AI-related data growth to continue over the next three years.

The findings point to a shift in how companies manage information created and reused by AI systems, including larger data lakes, longer retention periods and more archived material being brought back into active use.

IDC surveyed 763 IT and business decision-makers across seven countries with direct responsibility for AI infrastructure or data storage decisions. The study found that 94.7% of organisations had stored more data over the past 12 months because of AI and generative AI adoption.

Among those respondents, 61% said AI had driven data growth of 25% or more in the past year. A larger share, 74%, expected data volumes to rise by at least 25% over the next three years.

Data growth

The research found that 85.4% of organisations had seen their data lake volumes increase over the past year. It identified AI-generated data, including synthetic data, inference outputs and model logs, as the main reason for that expansion for 59.4% of respondents.

This suggests AI's impact on storage extends beyond the datasets used to train models. Data produced during the operation of AI systems is also adding to long-term storage demand.

Nearly 95% of respondents said AI and generative AI adoption had increased the value of their organisation's data. At the same time, 74.3% said those technologies had led them to retain data for longer.

The survey also found that 75.9% were bringing back growing volumes of archived cold-tier data to support AI workloads. In another sign of changing storage needs, 96% said they expected to need faster archive retrieval for AI inference and retrieval-augmented generation applications.

Storage mix

IDC's findings suggest much of the information used by businesses sits outside the fastest storage layers. According to the research, 74.6% of enterprise data resides in warm, cool and cold storage tiers, while more than 60% of data lake volume consists of cold or infrequently accessed data.

This matters because companies often build AI systems around assumptions of immediate access to large volumes of information. The figures indicate that a substantial share of potentially useful data is held in storage tiers designed for lower access frequency and lower cost.

The survey found near-universal focus on spending for storage infrastructure. Some 98.2% of respondents said total cost of ownership per terabyte was important or very important when making storage decisions.

The results come as companies reassess the balance between compute spending and the less visible costs of storing and retrieving data over time. Much of the recent debate around AI infrastructure has focused on processors and model training, but the report argues that data persistence is becoming a central issue.

WD said the economics of storage are changing as data created by one AI workload can remain useful for later applications. That includes training datasets, model checkpoints, logs, synthetic data and inference outputs that may be retained and reused rather than deleted once an initial task is complete.

Executive view

Irving Tan, Chief Executive Officer, WD, said the findings reflected a broader shift in AI infrastructure planning.

"For the last few years, the AI infrastructure conversation has centred on compute. But AI runs on data," said Irving Tan, Chief Executive Officer, WD.

"Organisations are generating more data, keeping it longer, and finding new ways to create value from the information they already have. Compute requirements will evolve over time, but the need to store, manage and access data at scale is only growing. That foundation will play a critical role in determining how far AI can go."

The white paper was based on a quantitative survey and supported by qualitative interviews with three senior storage industry leaders. It focused on organisations already involved in AI infrastructure and data storage decisions, showing how established AI users are adapting their systems as historical and newly generated data become more closely linked.

One of the clearest messages in the study is that many organisations no longer treat archived data as dormant. Instead, they are recalling historical datasets to support newer AI tasks, changing how they think about the boundary between active and inactive information.

That trend could have practical consequences for spending priorities across corporate IT, particularly as companies weigh the cost of keeping more information accessible against the potential benefits of using it in model training, inference and retrieval-based systems. The IDC figures show that, for many organisations, pressure is already building across the full data lifecycle rather than in compute alone.