AI investment is rising quickly, but companies are still finding it difficult to secure real returns on investment. TechRadar reported on Aug. 6 that global AI spending this year is forecast to rise 47 percent from a year earlier to $259 billion. Yet only 28 percent of AI projects deliver a measurable return on investment. Many companies are ramping up spending to seize AI opportunities, but are rushing adoption without fully understanding how AI will work inside existing technology environments, especially data storage infrastructure.

AI is often discussed in terms of compute power, GPUs and processing speed, but it is fundamentally a data system. As AI moves from experiments to large-scale deployment, the ability to collect, store, access and manage data becomes more important. The key is to see storage infrastructure not as a simple supporting technology but as a strategic foundation that determines AI outcomes.

Traditional data storage was relatively predictable, and approaches that required little change after initial buildout also worked. In the AI era, the situation is different. IDC expects global annual data creation to more than triple over the next 5 years to reach 718 zettabytes in 2030. AI workloads continually generate new data, including logs, metadata, synthetic outputs, model updates and training datasets. If storage infrastructure is insufficient or not suited to AI workloads, expensive compute resources such as GPUs can be left waiting, ultimately lowering AI investment returns.

One solution is tiered storage. Not all data requires the same level of performance and cost. Data frequently used for model training and inference is kept on high-performance storage, while older data, archived outputs and records for regulatory response are moved to lower-cost storage tiers. Adjusting storage performance and cost to match data value and usage frequency can optimise infrastructure costs while maintaining accessibility needed for AI innovation, governance and future model development.

Regulation and compliance should also be considered from the start of AI planning. Examples include personal data protection rules in each country, rules on data use and access, data governance and transparency for companies with overseas customers, and record retention requirements. Retention periods for training data and model records may be longer than expected, making long-term storage strategy important. If companies address this late, costs and complexity can rise.

AI returns on investment are not determined by algorithms or applications alone. How efficiently large-scale data is stored and managed, and how it is controlled to meet regulations, will determine outcomes. In petabyte and exabyte-scale environments, even cost differences per terabyte can become a major financial burden. Companies need to forecast storage capacity demand alongside GPU investment and include a data infrastructure strategy from the AI business-planning stage.