AI storage vendor Vast Data is launching what it calls a Flash Reclamation Program to help customers in the current AI-driven memory drought.

In an exclusive briefing attended by SDxCentral, Vast cofounder Jeff Denworth said the move would help clients navigate the 2026 SSD "Dry Year," many of whom, he said, have only had the last month to solve the problem as memory prices began to spike toward year-end to sate the scaling surge of AI use cases.

Vast's Flash Reclamation Program, as it's currently called, aims to replace existing storage or database software with its own platform, with the promise of delivering two to three times the capacity customers had originally planned for.

To do this, Denworth explained, Vast engineers perform a full survey of a customer’s infrastructure, while R&D rapidly qualifies whatever SSD hardware is found. They can then reclaim NVMe drives from legacy arrays, data lakes, and existing servers, potentially extending those NVMe investments by a factor of two to three.

Denworth claimed the same applies to environments generating large amounts of AI data. Promising around 1.7:1 data reduction plus error-correction efficiency, Vast aims to double the usable capacity of systems not originally running its software.

These numbers are likely to not only please engineers, but bean counters too, according to the Vast exec.

"Not only does it reduce capacity, it also pins the price point to 2025, which is roughly 50% less than what customers today have to pay," Denworth said, while adding the program doesn't impact on current contracts or SLAs.

"If customers want to give us a bunch of SSDs, that would look like other SSDs that customers already build with us. So there's just standard capacity licensing that we provide. In that case, there's nothing exceptional that we need to. So it comes free as part of the package."

After Flash, there's a cache catch...

In an interesting detail, Denworth also gave the customer insight on another 'drought trigger' as arising from the use of key-value (KV) caching. This is a token cache that registers prior interactions with a model to avoid activating the GPU when those interactions are repeated during the process of AI inferencing – how AI models generate their output.

Denworth believes KV caching creates a huge new data set that the industry hasn’t planned for, compounding the already severe flash shortage. The exec also pointed to another driver in Nvidia's recently released approach to key value caching in the form of Inference Context Memory Storage (ICMS) platform.

"There's going to be a need for another 100 exabytes of additional capacity in the market, which, from our perspective, doesn't necessarily exist."

Alongside Nvidia's ICMS launch last week, Vast natively integrated its operating system with Nvidia's BlueField-4 DPUs in a bid to service inference sessions in the agentic era, as powered by Vast's Disaggregated Shared-Everything (DASE) approach to architecture.

"We're basically bringing an end to the era where a bunch of clients have to share server resources to get access data. So now every client can have their own server, and because of this new parallel architecture that we built where no Vast servers need to talk to each other, and we strip out a bunch of the intermediary infrastructure in terms of servers and in terms of switching, but you know, at the end of the day, data reduction is still a really important thing."

Denworth explained that by eliminating separate infrastructure components such as file servers, object storage servers, and database servers, and instead integrating these functions into existing data processing units (DPUs) within the host system, significant power savings can be achieved while consolidating data services.

"It's an advantage as compared to, ultimately, what ambition Nvidia had originally envisioned the product to do, as we can bring all of these optional background data services into the picture," Denworth said. "Because with our DASE architecture, we could just throw in a couple other servers in the background that can do things like audit controls and protect machines from disk failures, to data reduction and implementing lifecycle policies.

"So there's now these optional data services features that customers get for Key-Value cache with Vast and and we're going to do our damnedest to kind of compress the net capacity impact that this has on the market."

The vendor promised to announce further details on its Flash Reclamation Program.