The rapid evolution of artificial intelligence from generative models to autonomous 'agentic' systems is placing unprecedented demands on data infrastructure, but industry insiders argue that the sector has been focusing on the wrong bottleneck. While much attention has been lavished on new graphics processing units and specialised silicon, the real challenge at scale is where to put all the data.
Nicolas Frapard, senior manager and regional lead for EMEAI at Western Digital, told UKPulse Media that early AI development naturally emphasised compute power because the priority was simply making models work reliably. 'That naturally put the spotlight on GPUs, accelerators, interconnect bandwidth, and overall compute density,' he said. However, as AI adoption expands to billions of interactions, data growth becomes structural rather than incidental, driving sustained demand for storage.
The economic implications for UK organisations are stark. At exabyte scale, even small inefficiencies in storage architecture become magnified, directly affecting total cost of ownership and return on investment. Frapard noted that while compute tends to become more efficient over successive generations, data volumes continue to expand, meaning storage costs are a growing line item on corporate balance sheets.
A persistent myth in the industry is that storage tiers are obsolete and that all-flash environments can replace traditional hard disk drives across the board. In reality, the cost gap between flash and HDD has widened rather than narrowed, with flash up to 20 times more expensive per terabyte. 'The flash everywhere approach emerged for similar reasons as the compute-centric view – at small scale when performance was the primary concern and data volumes were manageable, it was a good choice,' Frapard explained. For large-scale enterprise deployments, HDDs remain the backbone of storage architecture.
For UK businesses navigating this landscape, the regulatory context adds another layer of complexity. The Information Commissioner's Office expects organisations to demonstrate robust data governance, while the EU AI Act imposes requirements on data quality and traceability for high-risk systems. Firms that fail to plan for storage lifecycle management risk both operational inefficiency and compliance exposure. As AI continues its march towards human-like reasoning, the quiet revolution in data storage may prove as decisive as any breakthrough in chip design.