Not that critics of data center expansion needed another reason to oppose those facilities in their backyards, but they have one: the memory chip shortage.
As fabricators reallocate their resources to meet the demands of the AI industry, the big losers will be consumers, as the prices of the gadgets they love soar.
“AI servers need enormous amounts of memory, and the three makers that supply the world, Samsung, SK Hynix, and Micron, are reallocating wafer capacity to high-bandwidth memory for AI chips because it is far more profitable and effectively sold out,” explained Jeff Barrington, managing director of Windsor Drake, an investment banking and M&A advisory firm in Toronto.
“That starves consumer-grade DRAM and NAND, and the prices show it,” he told TechNewsWorld. “Contract DRAM jumped about 90% in the first quarter and another 60% in the second, after rising 172% across 2025.”
Data centers sit at the center of the shortage because AI servers are memory monsters, explained Francisco Jeronimo, vice president for EMEA and devices at IDC, a market research company in Framingham, Mass.
“A single AI server consumes 10 to 20 times more memory than a conventional workload server, so as the hyperscalers build out, they pull a hugely disproportionate slice of global supply,” he told TechNewsWorld.
That scenario is complicated by the same handful of suppliers serving both the data center and the consumer side. “The high-bandwidth memory feeding AI data centers comes off the same DRAM wafers as the RAM in a laptop or phone,” Jeronimo said. “Capacity is finite, so a wafer committed to an HBM stack for an AI data center is one that never becomes memory for a mid-range handset.”
“It is close to a zero-sum game,” he added. “We forecast that data centers will absorb about 70% of all memory produced worldwide in 2026, up from 20% to 30% as recently as 2022.”
Why AI Memory Pays More
Sandip Patel, of Frisco, Texas, a senior cloud solution architect with Microsoft, noted that it’s common to think of memory chips as a commodity that scales with consumer demand. “AI flipped that,” he told TechNewsWorld. “Now a handful of cloud providers can outbid the entire consumer electronics industry for the same wafer capacity, and that’s exactly what’s happening.”
“Why chipmakers are chasing that demand instead of sticking with consumer electronics comes down to simple economics,” he said. “AI-grade memory sells at a much higher margin than the commodity DRAM that goes into your phone or laptop.”
“Fab capacity is finite, so when a manufacturer has to choose, they put their wafers toward the product that makes them more money,” he continued. “It’s not personal. It’s just where the returns are right now, but it leaves less room on the line for consumer chips.”
Tzvika Shahaf, senior vice president of product strategy at the Blancco Technology Group, a global company specializing in data erasure and mobile device diagnostics, noted that not only can chip makers get more money for AI chips, but they can also secure longer deals.
“Hyperscalers are driving significant demand with multi-year supply contracts that lock guaranteed capacity at premium prices,” he told TechNewsWorld. “The result is that large chip manufacturers have shifted production toward the hyperscalers and slowed down on consumer-grade chips.”
When Supply May Catch Up
The semiconductor industry is highly cyclical because it depends on multibillion-dollar manufacturing facilities, explained Arie Brish, a business professor at St. Edward’s University in Austin, Texas.
“The industry is almost either in an over-demand situation or an over-supply situation,” he told TechNewsWorld. “The over-demand is usually driven by explosive growth in some segment, such as Covid driving semiconductor demand for videoconferencing, which caused an extreme shortage of semiconductors for other industries. This time the demand is driven by AI.”
“During such supply shortages, semiconductor suppliers invest heavily in production capacity to meet the explosive demand,” he said. “These capacity expansions take time and money. Once they catch up, we will start to see some relief.”
“This relief is expected to emerge gradually in the next few months,” he added. “Once the semiconductor suppliers catch up with demand, it is likely to drive a significant drop in semiconductor stock prices.”
Jonathan Schaeffer, CEO and founder of Synsira, a Canadian software company that develops AI-powered tools, agreed that relief will be gradual. “Manufacturers are expanding capacity, but semiconductor supply chains do not turn on a dime,” he told TechNewsWorld. “High-bandwidth memory is complex to make, and new supply can be absorbed quickly if AI demand keeps rising.”
“The better long-term answer is not just more supply,” he contended. “It is also better efficiency.”
“AI systems need to do more with less through better algorithms, smaller specialized models, data compression, edge computing, and hybrid approaches,” he continued. “If the answer to every AI problem is ‘build another giant data center,’ then we are not being imaginative. We are just being expensive.”
“The history of technology shows repeatedly that if demand for a product is large, economies of scale and innovation take over,” he said. “With AI, computing infrastructure demand will decrease as the AI algorithms become better, faster, and less resource-intensive.”
Recovering Existing Memory
Blancco’s Shahaf pointed to another source of relief for a memory-hungry world. “There is a more immediate lever the industry has largely overlooked — the memory and components already sitting in retired IT assets,” he said.
“For years, the default approach to end-of-life IT assets was to destroy-first — erase it, shred it, recycle the raw materials,” he explained. “That’s effectively throwing working memory away at the same exact time the world can’t make enough of it.”
“A paradigm shift toward recover-first could put a meaningful amount of much-needed memory, storage, and other components into circulation instead of into a shredder,” he predicted.
“With new fab capacity years away, recovering and recirculating what already exists seems to be the most realistic source of relief,” he added.
Not in My Back Yard
Michelle Lopes Maldonado, associate director of AI policy at the Information Technology & Innovation Foundation, a research and public policy organization in Washington, D.C., predicted that the memory market could take up to two years to recover from the AI-driven surge in demand.
“Since new HBM fabs can take years to reach volume, constraints may not meaningfully ease for consumers for another year or two, particularly if faster fab and grid permitting efforts don’t move forward,” she told TechNewsWorld.
“Opposition to data center siting over power and water use could have the effect of slowing capacity buildout needed to ease bottlenecks, including new fabs, which is why streamlined permitting and process transparency also play a critical role in addressing this issue,” she said.
Resistance to data centers is rising fast at the local level, noted Mark McNees, director of social and sustainable enterprises at the Jim Moran College of Entrepreneurship at Florida State University in Tallahassee, Fla. “In Florida, there are moratoriums in at least 20 counties and communities,” he said.
“It is driven mostly by power and water costs, not chip supply,” he explained.
“That opposition slows where data centers get built, which over time shapes where demand for chips and everything else lands.” However, he added, “It is a siting and cost story before it is a supply story.”
What’s the Next Constraint?
Community opposition to data centers will have implications on the expansion rate and the risks associated with how hyperscalers view their scaling possibilities, but in the immediate term, it will not be a relief for the memory shortage, observed Abhijit Sunil, a senior analyst with Forrester Research, a national market research company headquartered in Cambridge, Mass.
“Even if some projects are delayed, cloud providers can redirect investment to more favorable regions since an unprecedented amount of investment has already been committed to AI infrastructure,” he told TechNewsWorld.
“One thing enterprise leaders should care about is how AI is resource-intensive and what could be the next bottleneck, both inside and outside the data center,” he advised. “Before AI investments stabilize, there could be other bottlenecks and opportunities in the supply chain that are yet to arise.”
Read the full article here
