VAST Data Reclaims Flash to Address SSD Crunch

Companies initiating AI projects are facing a hard truth: Prices for enterprise storage are climbing by as much as 40%, as demand for the media central to all leading enterprise storage systems outpaces supply.

The cause: demand from hyperscalers and datacenter providers for flash storage to fuel AI workflows. Suppliers such as Micron, SK Hynix, Solidigm, Samsung, and others can’t meet the needs of the datacenter storage vendors that fuel these customers. The situation has drastically limited SSD supply and sent prices sharply upward.

It’s a situation that’s prompted at least one AI storage market leader into action. “We’ve identified an issue that we think is severe enough where it deserves maybe broader industry attention,” said Jeff Denworth, cofounder at VAST Data, in a briefing for industry analysts this week. VAST has launched a “Flash Reclamation Program,” in which it will repurpose enterprises’ existing flash drives to support VAST’s architecture, which the vendor says reduces data bulk by up to 200%.

VAST’s campaign calls for sales engineers to take “squandered, poorly utilized and valuable masses” of SSDs otherwise controlled by competitors and turn that into storage for VAST Data’s “AI operating system” of data management software. There will be no extra licensing fees for taking over the drives, Denworth said.

The news is significant on several fronts. It illustrates how vendors such as VAST, which competes against DDN and WEKA, among others, have shifted from serving storage in HPC environments to managing data for use in accelerated computing workflows, in which keeping GPUs humming busily is central to ROI.

VAST Data’s reclamation project is also a clever move that could encourage enterprises formerly hesitant to deploy VAST’s “disaggregated, shared-everything” (DASE) architecture to make the move to VAST, whose design depends on flash to meet the high-performance demands of AI workflows. Let’s take a closer look.

SSD Scarcity, VAST’s Opportunity

Storage and data management are key to the growth of enterprise AI. And NVMe SSDs are essential to enterprise storage systems from all the market-leading suppliers. While hyperscalers gobble up precious components faster than manufacturers can supply them, enterprise storage providers are in the unenviable position of raising their prices too.

VAST thinks it can capitalize the situation by proving that, if enterprises can’t buy enough SSDs to support their current storage solutions, VAST can bring greater efficiency to what they already have in place. This should help users make the move from their “legacy” systems to VAST, which represents a relatively radical option that deploys an architecture in which all data is in one flash tier, accessible by any processor any time. The result is performance highly suited to AI workflow demands, the vendor says.

VAST says its approach to erasure coding, which splits data into fragments that can be spread across RAID storage, is highly efficient, consuming 2.7% RAID overhead compared to 20% for “other all-flash” competitors. VAST also uses a tool called similarity data reduction, which removes not only duplicates but highly similar segments of data for representation in storage. (Data can be retrieved in full when read.)

VAST’s SSD reclamation project puts the company into the spotlight as a key player. Sales are booming, Denworth said, and growth is expected to be in the "hundreds of percentages" this year. About 1,000 customers contributed to the vendor’s claim of reducing data in enterprise storage by a factor of 3.40 to 1. VAST Data has been cited as a potential IPO candidate, given its rumored $30 billion valuation.

The Storage Industry’s Tectonic Shift

Some features claimed by VAST are supported in various iterations by other vendors, which also are focused on fueling AI and are adopting their own strategies during the SSD shortage. Cloudian, for instance, claims flash storage isn’t required for 80% of stored data. Instead, it uses a tiered storage setup that deploys flash for AI workflows and HDD for "cold storage" in a unified architecture with a single operator interface.

Cloudian isn’t alone. Storage vendors are quickly adding a range of solutions that will enable them to function more efficiently within the NVIDIA-driven AI pipeline, especially in terms of memory usage. At the recent CES conference in Las Vegas, NVIDIA revealed a new capability associated with its new BlueField-4 DPUs titled the Inference Context Memory Storage Platform. In this approach, KV caching, which handles key values associated with individual tokens in an AI model, is shared with the NVIDIA DPU, and only the most important tokens are processed by high bandwidth memory (HBM) in the GPU.

Among the storage vendors developing products with BlueField-4 are AIC, Cloudian, DDN, Dell Technologies, HPE, Hitachi Vantara, IBM, Nutanix, Pure Storage, Supermicro, VAST Data, and WEKA, according to NVIDIA. (Of these, most, including VAST, already support KV caching.) This technique will be increasingly needed as NVIDIA shifts into gear with its Vera Rubin GPU, expected to be released in the second half of 2026. According to VAST Data’s Jeff Denworth, the Rubin shift will call for 100 exabytes of additional capacity in the market.

Futuriom Take: VAST Data has taken a leading role in addressing the SSD shortage among storage and data management vendors. Expect others to follow suit with similar efforts. And watch VAST Data for signs of growth and a potential IPO.