An AI-driven flash shortage is emerging as one of the defining infrastructure challenges of the machine learning boom. More than a problem solved by waiting out the typical hardware cycle, the shortage is forcing a larger rethink on how enterprises can get more from existing capacity.

Unlike past supply crunches, this shortage is hitting even the biggest buyers. The current pattern suggests the shortage is unlikely to resolve on a normal product-cycle timeline, according to Phil Manez (pictured, right), vice president of go-to-market execution at Vast Data Inc.

“We’re seeing the biggest customers on the planet not be able to get the allocation that they need, both for kind of business-as-usual growth and, at the same time, trying to bring all these new applications and AI into their environment,” Manez said. “It’s serious — and it’s something we expect to really play out through this year, potentially really into maybe the middle of next year.”

Manez and Ace Stryker (left), director of AI and ecosystem marketing at Solidigm, a trademark of SK Hynix NAND Products Solutions Corp., spoke with theCUBE’s Dave Vellante and Rebecca Knight at Vast Forward 2026, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed the growing AI-driven flash shortage and the need for more efficient storage architectures as a response. (* Disclosure below.)

AI-driven flash shortage demands efficiency

AI systems are being used more often not just by people, but by agents and automated systems making constant requests through application programming interfaces. That surge in machine-to-machine traffic is reshaping how compute and memory resources are consumed, according to Stryker.

“These models with these context windows that are just growing and growing, and these longer loops, more iterations, in a given interaction with a model,” he said. “All of that has incredible storage implications. And it does not appear that this is a cyclical thing, that this is likely to wane any time soon. That’s where we find ourselves in 2026.”

With flash supply constrained, the emphasis has shifted toward architectural efficiency — particularly in how inference workloads consume memory and storage resources. A major focus lately has been inference context and the key-value cache, as expanding context windows and more iterative model interactions keep larger volumes of active data readily accessible, according to Stryker. That efficiency imperative is also shaping how Solidigm works with partners and customers, including Vast, an enterprise data platform provider focused on AI-optimized storage infrastructure.

“[The answer is] space and do more with less — it’s been a general sort of name of the game in AI outcomes for the last few years, which has been efficiency across the board,” Stryker explained. “The demands that AI is placing on capacity are unprecedented. You can’t procure your way out of this.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of Vast Forward:

(* Disclosure: TheCUBE is a media partner for Vast Forward. Sponsors of theCUBE’s coverage, including presenting sponsor Solidigm, do not have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE