How can organizations help ensure they're optimizing ROI from their AI investments?

Investment in AI continues to accelerate at pace, with global spending projected to reach $2.59 trillion in 2026, a 47% year-on-year increase.

Yet only 28% of AI projects are currently generating a measurable ROI, with many failing to deliver expected business outcomes.

As organizations race to capitalize on AI opportunities, many continue to increase investments without fully understanding how AI will operate within their existing technology environments, particularly their data storage infrastructure.

Without careful planning in this area, businesses risk higher costs, growing technical debt and increased compliance complexity, all of which can undermine the value AI is set to deliver.

No data, no AI

The IT industry often talks about AI in terms of performance, computing power and processing speed. But at its heart, AI is a data system.

In recent years, high-performance processing GPUs and NPUs have been getting much of the attention and investment. While compute remains essential, AI data is constantly evolving, expanding and requiring ongoing management throughout its lifecycle.

As organizations move from AI experimentation to deployment at scale, the importance of a robust data infrastructure becomes increasingly clear.

Businesses need the ability to capture, store, access and manage growing volumes of data efficiently at every stage of the AI journey.

To maximize AI ROI, organizations should view storage and data infrastructure as strategic enablers rather than supporting technologies. Long-term AI success depends on ensuring storage infrastructure is aligned with wider technology and business objectives from the start.

AI requires a different approach to storage

Historically, data storage needs were relatively predictable. A ‘set-and-forget’ approach was often sufficient. The AI era demands a fundamentally different mindset for several reasons.

For one, the scale of data associated with AI exceeds anything many organizations have previously encountered. According to IDC, annual global data creation is expected to more than triple over the next five years, reaching 718 zettabytes by 2030, representing a CAGR of 26.9%.

AI workloads also continuously generate additional data through logs, metadata, synthetic outputs, model updates and training datasets. Just as importantly, AI performance is heavily influenced by the quality of the underlying data infrastructure . The ability to access large, well-managed datasets efficiently has a direct impact on business outcomes.

When discussing return on AI investments with the C-suite, technical leaders should connect infrastructure decisions to financial performance. GPUs are typically among the most significant items within AI budgets, and any time spent waiting on storage input/output represents underutilized investment.

Storage architectures that are not designed for AI workloads can constrain performance and reduce overall returns from compute investments.

One example for an AI-optimized architecture approach is tiered storage. Not all data delivers the same value at every stage of the AI lifecycle, nor does it require the same level of performance. Frequently accessed datasets used for active model training and inference may benefit from high-performance storage, while historical data, archived outputs and compliance-related records can be moved to lower-cost capacity tiers.

By aligning storage performance and cost with the value and usage profile of different datasets, organizations can optimize infrastructure spending while maintaining access to the data needed to support AI innovation, governance and future model development.

Organizations that invest in cost-effective scalable, future-ready, AI workload optimized storage and data infrastructure are better positioned to unlock value from AI initiatives, while building a foundation that can support future growth.

AI data, compliance and regulatory risk

Legal and regulatory considerations should also form part of AI planning from the start.

As AI laws are established, organizations must continue to comply with existing data-related obligations, including regulations like UK GDPR and the Data (Use and Access) Act 2025.

Any organization with EU customers should also consider EU requirements around data governance, transparency and record-keeping. Retention periods for training datasets and model records can often be longer than anticipated, making long-term storage planning a decisive factor.

Addressing these needs early helps organizations avoid costly remediation efforts later and supports more effective governance as AI deployments scale.

Talking a proactive approach to protecting AI ROI

As AI adoption grows and data volumes continue to expand, demand for storage capacity is increasing rapidly. This is creating new supply chain and procurement challenges across the industry.

As a result, organizations can no longer assume that capacity will be readily available whenever it is needed, particularly for large-scale AI projects.

In practice, this means considering forecasting storage requirements alongside GPU and infrastructure investments, exploring longer-term capacity planning arrangements, and incorporating storage needs into AI business cases from the beginning.

A proactive storage strategy is becoming essential for businesses looking to support future AI workloads with confidence, minimize operational risks, and maximize long-term returns from their AI investments.

The last word

AI ROI depends not just on the quality of its algorithms and applications, but also on how effectively organizations manage, store and govern their large-scale data estates.

Even a small difference in cost per terabyte can become significant when applied across petabyte- and exabyte-scale environments.

Organizations that integrate forward storage planning into their AI strategy, treat data infrastructure as a strategic asset and prepare early for future capacity requirements...

Indexed and credited by AIPROPX. Originating outlet: TechRadar. Open at source →