Enterprises are facing storage shortages, long lead times, and dramatic price increases, which may cause some to defer on-prem AI projects.
Enterprises are seeing jumps in storage prices that started last year and will extend into 2026, researchers predict, due to unprecedented demand related to AI deployments and the growth in data in general. As a result, some enterprises may defer on-prem AI projects for a few months until availability increases and prices stabilize.
“Storage needs are always expanding,” says Falko Kuester, engineering professor at UC San Diego and the director of Open Heritage 3D, a non-profit project that collects images, videos, LIDAR, point clouds, and other scans of historic sites and shares them with the public and with the research community.
In order to create a shared “ground truth” of this data, the organization makes all the raw data and scans available in an easily accessible online format, which creates an ever-growing amount of data. The scans are processed, analyzed, and annotated in various ways, and the amount of data is multiplied, sometimes exponentially, Kuester says. Not to mention the growing resolution of imaging and scanning technologies. That adds up to hundreds of terabytes of data, and it’s expected to hit a petabyte within the next 18 months.
“That’s the nature of the beast,” he adds. “We’ll keep consuming as much storage as we have. That’s a self-fulfilling prophecy in data science.”
Heritage 3D isn’t alone.
Some 40% of companies with 1,000 or more employees are currently storing more than 10 petabytes of data, according to a December report from Komprise. And 85% of organizations expect to increase their storage spending in 2026, up from 59% in 2024’s survey. The top priorities? Cost optimization, cited by 64% of respondents, followed by data preparation for AI at 61%, and cloud migration at 54%.
Higher prices ahead
All indicators are showing a steep price increase for memory and storage in 2026.
Brad Gastwirth, for example, says he met with many of the most important players in the market at CES earlier this month, and his analysis suggests there will be a 50% or more price increase in DRAM and NAND.
“Memory and storage are now system-level performance constraints, not secondary components,” says Gastwirth, global head of research and market intelligence at Circular Technology, a supply chain consultancy. With multi-quarter AI infrastructure demand growth by hyperscalers and enterprises, and limited supply, we’re seeing a fundamental reset in demand and supply, he says. “This is not a temporary spike.”
“We’ve been predicting that we were going to start seeing rising memory prices since last summer, so we’ve been well prepared for it,” says Scott Tease, vice president and general manager of AI and high-performance computing at Lenovo’s infrastructure solutions group. He expects prices to go up as much as four-fold compared to the start of 2025. “So very significant.”
“A 64-Gig DIMM, which is a mainstay, a sweet spot memory, right now we’re procuring that in the low two-hundreds,” Tease adds. “That’s likely going to approach $800 for that same exact DIMM in the next few months. Every device we make — this phone, that laptop, our servers — they all have that same memory building block in there. So, it is going to affect the entire industry.”
A January report by TrendForce predicts that DRAM prices will be 55% to 60% higher this quarter compared to the same time last year, while NAND Flash prices will be up 33% to 38%.
A global shortage
And DRAM suppliers will continue to reallocate capacity to support AI server demand, TrendForce predicts, limiting supply in other markets. While large players like hyperscalers and OEMs have their supplies locked in, smaller buyers might be out of luck. Suppliers’ inventories are approaching depletion, the research firm said in a January report.
Other types of memory are also experiencing supply constraints, but for different reasons. Take for example, MCL NAND Flash, which has steady demand in enterprise markets, driven by industrial control, medical devices, and networking equipment. According to TrendForce and other industry sources, Samsung will be ending MLC NAND production in June of 2026, and other suppliers have stepped up to fill the gap. TrendForce predicts that total MLC NAND Flash capacity will fall by 42% this year.
And it’s not just these two components.
“There’s not enough supply to meet demand right now. and that’s true of all the memory and all the storage technology,” says Tom Coughlin, IEEE fellow and president of Coughlin Associates.
Manufacturers are being cautious about increasing supply because they’ve been burned before, he says, when, by the time they increased the capacity, the market had moved on. “To build a semiconductor plant, that’s another 15 months and costs $50 billion,” he says. “They had to sell below cost, and slash manufacturing, and for semiconductor plants that’s a mess.”
To protect themselves from swings in market demand, hard drive and SSD manufacturers are signing long-term contracts with large-scale customers like data centers, he says. “Some of those contracts go out to 2027.”
That makes it harder for other players to buy what they need, since the supply is all tied up.
One of those hard drive and SSD manufacturers, Western Digital, confirmed this. “Supply will remain tight next year across the board, and it’s not just us,” says Ahmed Shihab, the company’s chief product officer. “Every earnings call the last few weeks has had the same thing. There’s something structurally different.”
He agrees that AI is driving most of the new demand. “Whether you’re driving training or driving inference, the scale of change is off the charts,” Shihab says. “We have capacity contracts with our customers going out to 2026 and beyond, and the average capacity of the drives we’re shipping is going up.”
According to TrendForce, some SSD models are now facing delivery delays of more than a year.
Growing demand for QLC
The hard drive shortage is accelerating adoption of QLC SSDs — quad-level cell solid state drives — as an alternative, and they will account for 30% of the enterprise SSD market, TrendForce predicts.
“There’s a push to go to QLC,” agrees Coughlin. “It will enable higher capacities in a smaller footprint.”
So, what should enterprises be doing if they’re getting squeezed by higher prices on one hand and shortages and delays on the other? One thing they can try is to try to get more life out of the memory they’ve already got, says Coughlin.
“Some of the products have gotten more reliable than they had been, but there are also some tricks,” he says. For example, minimizing the number of writes to QLC NAND Flash will help extend the life of those components. “So, they’re trying to consolidate data so you only write what you need to write, and write in a more efficient way,” he says.
“Now is not the time to buy a whole bunch of extra storage,” says Forrester analyst Brent Ellis. “If you’re a mid-sized enterprise and you’re thinking of purchasing a small AI cluster and you need storage to support that new cluster, you might choose to delay that a few months.”
