What happened
Solidigm, a trademark of SK Hynix NAND Product Solutions Corp., positions high-capacity solid-state drives as the new "intelligence layer" for agentic AI inference. This shift moves storage from commodity plumbing to a critical component for continuous inference at scale, according to Greg Matson, Solidigm's Senior Vice President. Agentic workloads, like a 15-word prompt generating 40,000 tokens, demand 5-10 gigabytes of context data, pushing storage needs to petabytes. Solidigm responds with 122 terabyte drives and the industry's first cold-plate-cooled enterprise SSDs for fanless Nvidia GPU servers.
Why it matters
GPU underutilisation due to data bottlenecks now drives significant AI infrastructure costs. As agentic inference scales, the gap between GPU memory and data feed creates a critical bottleneck, wasting expensive GPU cycles. Procurement teams and infrastructure architects must now prioritise high-performance storage to ensure continuous GPU utilisation and efficient token generation. This follows earlier reports that memory components dominate AI chip costs, reinforcing storage as a strategic investment rather than a commodity.
