Many companies that set out to build on-premises AI infrastructure have run into an unexpected wall. Storage budgets drawn up just a few months ago no longer add up by the time procurement rolls around—prices have risen that sharply. According to an Omdia survey commissioned by Verge.io (400 North American IT professionals in storage and infrastructure roles, conducted May 6–26, 2026), DRAM prices nearly doubled in Q1 2026, and SSD prices rose more than 50%. Ninety-eight percent of respondents said they are considering or already implementing storage-tier consolidation, and among companies that hadn't yet adopted it, the technology most frequently cited as a top investment candidate was software-defined storage (SDS). With hardware prices now out of control, companies are being forced to rethink the very design philosophy underpinning their storage foundations.
DRAM Doubles, SSDs Up 50%: Omdia Survey Captures North American IT's Distress
According to the survey commentary published by Verge.io, Omdia's characterization of prices as having "nearly doubled" and risen "more than 50%" is based on TrendForce data showing that DRAM contract prices rose 90–95% quarter-over-quarter in Q1 2026, while enterprise SSD contract prices rose 53–58% over the same period. Forecasts for overall NAND flash contract prices were also revised sharply upward—from a projected 33–38% increase as of January 2026 to 85–90% by March—with multiple data sets pointing in the same direction of a sharp spike.
In response to these price swings, 98% of survey respondents said they are either considering or already carrying out consolidation of their storage tiers. When companies that had not yet adopted SDS were asked which of eight technologies they were considering investing in, SDS was the most frequently selected. Omdia Chief Analyst Simon Robinson said, "We were struck by how many organizations are looking at software-defined storage as a potential solution." The fact that SDS came out on top reflects not just a reaction to the price spike itself, but also the perception that it offers a practical way to extend the life of existing storage investments without a wholesale overhaul.
Among the same 400 respondents, roughly three-quarters said rising storage costs are causing them to slow down on-premises AI deployments. Moving to the cloud is one option, but for regulated industries and companies with massive datasets, reversing course from an on-premises strategy is not a decision that can be easily undone. This is the structure at play: a reassessment of storage budgets is directly delaying AI deployment plans themselves.
Why Do SSD Prices Keep Climbing? A Production Shift Among NAND Makers
At the root of the price surge is a shift toward profit prioritization common across memory manufacturers. Samsung and SK hynix, which both produce DRAM and NAND, are preferentially directing their limited manufacturing capacity toward HBM (High Bandwidth Memory—a component made by stacking multiple DRAM chips to boost speed) production in response to surging demand for AI accelerators, while relatively curbing investment in general-purpose DRAM and NAND. According to semiconductor industry reporting, both companies plan to reduce their NAND wafer output in 2026 compared to 2025 (Samsung from 4.9 million wafers per month to 4.68 million; SK hynix from 1.9 million to 1.7 million). The decision to cut production in a rising market has served to tighten supply and demand even further.
Kioxia and SanDisk, which produce only NAND and cannot manufacture HBM, are stepping in to fill the supply gap left by Samsung and SK hynix. At the same time, NAND makers—including these two companies—are increasingly prioritizing production of enterprise SSDs for data centers, which carry fatter margins than general-purpose products. According to semiconductor industry reporting, the five NAND makers—Samsung, SK hynix, Kioxia, SanDisk, and Micron—have nearly sold out their entire 2026 NAND production capacity. While some new facilities, such as Kioxia and SanDisk's Kitakami plant, began operations in late 2025, new investments typically take several quarters of gradual ramp-up before they translate into meaningful supply, so there is no prospect of a rapid supply expansion in the near term.
The business logic—that directing production capacity toward high-value-added AI products yields higher margins than selling general-purpose products at thin margins in bulk—is rewriting the industry's supply structure as a whole. As a result, the supply of general-purpose enterprise SSDs that companies have relied on for on-premises storage has grown relatively scarce, and prices have risen ahead of any corresponding supply response.
The results are visible in the financial figures. According to a TrendForce report published on June 11, 2026, the enterprise SSD industry's Q1 2026 revenue reached $18.5 billion (approximately ¥2,941.5 billion, at ¥159/$1), up 86.1% quarter-over-quarter. The breakdown was $7.05 billion for Samsung (up 92.8% quarter-over-quarter), $4.64 billion for the SK hynix group, $3.09 billion for Micron, $2.22 billion for Kioxia, and $1.47 billion for SanDisk. The 53–58% rise in contract prices mentioned above translated directly into higher revenue for these manufacturers. A tailwind for sellers has become a headwind for buyers.
What Is Software-Defined Storage (SDS)? A Look at How It Works
SDS refers to a mechanism that decouples storage control functions—such as data placement, replication, and tiering—from dedicated hardware and runs them at the software layer. Traditionally, storage came as vendor-specific appliances that bundled hardware and control software together. SDS instead runs these control functions as software on general-purpose servers, allowing disks of different generations and from different vendors to be treated as a single pool.
This approach works because of a simple fact: data naturally splits into "frequently used data" and "rarely used data." Major SDS products such as VMware vSAN and Weka include functionality that automatically tiers data according to access frequency, making it possible to reserve fast, expensive new SSDs strictly for genuinely hot data while shifting rarely accessed cold data onto existing, cheaper HDDs. Rather than buying the latest, most expensive SSDs every time capacity runs short, this creates room to reuse older hardware already on hand—which is precisely why SDS is drawing attention amid the current price surge.
Verge.io CMO George Crump describes the situation this way: "Every byte sits on media purchased at 2026 prices. What remains unsolved is the gap between the capacity companies buy and the capacity their workloads actually need." He adds, "Reduce the number of architectures. Move cold data down to lower tiers. Stop buying a second set of hardware purely to protect the first set. None of this is exotic advice, but most companies have never even measured it." With prices out of their control, simply auditing and eliminating wasteful, redundant purchases has itself become a cost-cutting strategy.
At the same time, the currently published Omdia survey commentary (commissioned by Verge.io) does not touch on technical questions surrounding a shift to SDS—compatibility with existing hardware, the cost of migration itself, or the impact on latency. There remains a considerable gap between the fact that 98% of North American IT professionals say they are considering this move and the reality of actually completing such a migration.
Winners and Losers: NAND Makers' Revenue Gains vs. IT Departments' Cost Squeeze
At this juncture, NAND manufacturers are the clearest winners. Kioxia, a Japanese company, is one of them. Its full-year financial results for the fiscal year ended March 2026, announced in May 2026, showed revenue of ¥2,337.6 billion (up 37% year-over-year) and operating profit of ¥870.3 billion (up 92.7% year-over-year), an eight-year high for operating profit. This was the first time the company's revenue exceeded ¥2 trillion, with its joint production arrangement with SanDisk at the Yokkaichi and Kitakami plants contributing to the growth.
Kioxia and SanDisk announced in January 2026 that they had extended their joint venture agreement at the Yokkaichi plant by five years, from the end of 2029 to the end of 2034 (the Kitakami plant's agreement already ran through the end of 2034 at that point). With both plants now committed to continuing operations through the end of 2034, the companies have signaled confidence in continuing their business through this period of elevated prices.
By contrast, the ones on the losing end are enterprise IT departments—the buyers of storage. Roughly three-quarters have been forced to slow down on-premises AI deployments, and they now have no choice but to rework their budgets amid DRAM prices that have nearly doubled and enterprise SSD contract prices up more than 50%. In the Omdia survey, 27% of respondents reported holding 1–10PB of data on-premises, a roughly equal share reported 11–25PB, 18% reported 25–50PB, and 6% reported over 100PB. These four categories together account for roughly the high 70% range, with the remainder spread across other bands such as 50–100PB. The larger a company's data footprint, the more the price increases translate into real dollar impact.
On top of this, more than a third of respondents said 50–75% of their data was "dark data" whose actual usage is unknown, and 13% reported that figure at 76–90%. This makes visible, in numbers, just how inefficient it is to keep buying SSDs at current prices for data that may or may not even be in use.
The handling of dark data further complicates this picture. A situation in which more than half of a company's held data is of unknown utility means, conversely, that the company itself cannot determine whether that data is a future asset for training AI models or simply a liability being stored at inflated prices. Now that real-world SSD prices have jumped several-fold, the cost of postponing that determination has grown far larger than before.
This shift in demand is also a business opportunity for SDS vendors. Companies such as VMware vSAN, Ceph, MinIO, Pure Storage, Weka, VAST Data, and VergeIO now find themselves fielding inquiries from companies eager to extend the life of their existing hardware. While NAND manufacturers hold the upper hand on hardware pricing, a new competition is emerging over control of the software layer that sits on top of it.
How Outdated Is the 22.6x Figure?
The most eye-catching number in this survey is the comparison Crump presented: a 30TB QLC SSD costs 22.6 times as much per unit of capacity as an HDD of the same capacity. This ratio, up sharply from 4.9x as of Q2 2025, is based on QLC SSD data as of Q1 2026 published by VDURA in April 2026. But citing this peak figure as if it represents "the current gap" is not accurate. Tracking what happened afterward reveals a more complicated picture than simple continued widening.
Looking at the real-world prices for 30TB TLC SSDs and HDDs tracked over time by StorageReview, the SSD-to-HDD price ratio jumped from 7.0x in Q3 2025 to 23.2x in Q1 2026, before narrowing to 16.3x in Q2 2026 and standing at 18.6x as of the most recent Q3 2026 data. SSD prices themselves have continued rising consistently, from $3,460 in Q3 2025 to $22,600 (approximately ¥3.59 million) in Q3 2026—but when measured as a ratio, Q1 turned out to be the peak.
The reason the ratio narrowed isn't that SSD price increases have leveled off. Rather, HDD prices themselves rose sharply over the same period, from $755 to $1,216—a 61% increase. Even HDDs, long treated as the "cheap receptacle" for data, have surged in price, compressing the relative gap with SSDs.
This fact bears directly on the very premise underlying SDS's promised savings. The basic idea behind SDS is to combine expensive SSDs with cheap HDDs, splitting data between hot and cold tiers to control costs. But if the unit price of the HDD side itself has more than doubled in a single year, the savings SDS promises may end up smaller than originally assumed. Using the 22.6x figure as if it were still a valid benchmark for procurement planning risks focusing exclusively on the SSD-side price surge while overlooking the simultaneous price increases occurring on the HDD side.
Ripple Effects in Japan and Kioxia's Presence
Japanese companies are caught up in the same rising procurement costs as their North American IT counterparts. The enterprise SSD industry's $18.5 billion in Q1 2026 revenue alone translates to roughly ¥2,941.5 billion (at ¥159/$1, as of August 13, 2026)—and Kioxia, which holds a share of that market, is a Japanese company with main plants in Yokkaichi, Mie Prefecture, and Kitakami, Iwate Prefecture. Because prices for data center-grade SSDs and DRAM are set by global supply and demand, companies in Japan planning to build on-premises AI infrastructure cannot escape the same rising procurement costs.
The decision by Kioxia and SanDisk to extend their joint venture through 2034 signals that they do not regard the current price surge as a temporary supply-demand disruption. It reflects the NAND makers' own judgment that this is a structural shift lasting at least several years—and that rational business decision is, in turn, coming back as a binding constraint on enterprise IT departments.This asymmetry will not resolve itself unless the semiconductor price cycle itself changes, and the growing interest in SDS is likely to become entrenched not as a passing trend but as part of a procurement strategy reworked around the assumption of persistently elevated prices.
The burden facing Japanese companies looking to refresh their server storage is also visible in concrete figures. With a 30TB enterprise SSD now priced at roughly ¥3.59 million per unit, procuring even a few dozen units moves budgets into the hundreds of millions of yen range. Even setting aside exchange rate fluctuations, the scale of such procurement plans is now an order of magnitude different from just a few years ago. Given that ramping up new facilities inevitably involves a gradual increase in output, it's hard to imagine supply-side capacity suddenly expanding in the short term.
With few levers available to bring prices themselves down, the option left to companies is to change how they use storage. Tiered management and architectural streamlining through SDS serve as a stopgap measure to defer buying pricier SSDs, while also offering an opportunity to uncover wasteful, redundant investments that had never even been measured before. What lies behind the fact that 98% of North American survey respondents are taking action is not so much frustration over near-term price hikes as a structural sense of powerlessness—an inability to move hardware prices on their own. Until the new production capacity being planned by Samsung and SK hynix gradually ramps up and is actually reflected in the NAND makers' available supply, there is no sign that this structure will loosen.