AI’s memory crunch could squeeze GPUs and PCs through 2030

By Elena Vasquez

By the end of 2026, DRAM prices will have risen 400 percent since 2024, according to JP Morgan.

The forecast captures a supply crisis amplified by AI's appetite for memory. The source report says data centers absorb DRAM as fast as manufacturers can produce it. It calls the consumer effect “chipflation,” with GPU prices rising far beyond economy-wide inflation.

• Tom's Hardware's chips tracker puts Nvidia's GeForce RTX 5090 at $4,920 above its $1,999 suggested launch price from January 2025.
• The source report says some gaming GPUs in Japan cost roughly 41 percent more than their American counterparts.
• The report says GameStop markets a Sony PlayStation 5 Pro roughly $500 above MSRP as opportunistic resellers capitalize on shortages.

The squeeze reaches both sides of the hardware market. Component inflation raises costs for device makers, while scarce finished products give resellers room to mark up stock.

JP Morgan says the Consumer Price Index for software and accessories rose nearly a quarter over the past two years. The Economist's 2026 report found similar increases for video game consoles and laptops. The source report says those increases prompted an uncharacteristic fall in PC shipments.

The report attributes the shortage to AI data centers' demand for memory and manufacturers' reallocation of production capacity. Market stakes are immediate for buyers and broader for AI companies. Consumer electronics makers compete for the same DRAM production capacity used by AI data centers, while developers worldwide need scarce HBM to train and run their models at scale.

What do suppliers expect for 2027?

Many chip producers have already sold much of their 2027 capacity. The report says some now focus sales on 2028 and beyond. It also says some observers expect shortages to extend into the 2030s.

Micron has reportedly sold 75 percent of its 2027 HBM output. CEO Sanjay Mehrotra told Yahoo Finance that those sales were struck at prices “much higher than 2026 prices.”

KB Securities says Samsung has sold almost 80 percent of its 2027 memory output. Roughly 70 percent sits in five-year contracts, according to the Korean investment firm.

The already-sold capacity narrows the near-term market for buyers and gives suppliers pricing power before new production lines can respond fully to rising AI demand. The source report says Micron, Samsung Electronics, and SK Hynix all expect long-term shortages despite major capacity expansions.

Mehrotra said in a recent earnings call to expect supply to be “much tighter in calendar 2027 and 2028” than in 2026. SK Hynix CEO Kwak Noh-jung told Reuters in July: “we forecast that next year will be the worst year in the industry's history from the supply perspective.”

SK Hynix chairman Chey Tae-won said at a Nvidia event that it will take “at least four to five years” to build adequate wafer supply. Reuters reported that production will remain 20 percent behind demand until at least 2030.

The warnings make the consumer problem more than a temporary price spike. They also point to constrained supply through 2028, with production still expected to trail demand in 2030.

Why is this a GPU shortage at all?

At its core, the shortage is a memory problem. The report notes that chip supply cycles are familiar, including earlier peaks and valleys. This shortage differs because AI data centers have intensified pressure on an already strained manufacturing base.

The industry was already shifting from DDR4 to DDR5 memory before the AI boom, and the transition stalled supply as manufacturers converted production lines, according to the report.

The buildout then required vast stocks of high-bandwidth memory, or HBM. This specialized form of DRAM has faster connections and more bandwidth. AI developers need HBM to train and run their models, and demand now far exceeds available supply.

HBM competes for the same wafer capacity as traditional DRAM and other memory products. Suppliers shifted production toward HBM because demand and profit margins favor it. HBM sells for roughly 30 percent more than traditional DRAM, the report says.

Because HBM and conventional memory share wafer capacity, a shift toward one reduces capacity available for the other. The premium price makes that shift attractive to suppliers. It also leaves PC, laptop, and console makers competing for a smaller portion of output, according to the source report.

A move into HBM4 and HBM4E has tightened supply further. Those advanced generations require production changes while suppliers already face limited wafer capacity, according to the report.

A Samsung executive told Reuters in September that HBM could account for nearly 30 percent of total DRAM production in 2027, up from 20 percent the year before. Some forecasts cited by the report put the HBM share near 40 percent by 2029.

The report also projects that AI data centers will consume roughly 70 percent of global DRAM supplies this year. That leaves conventional PC and GPU products competing for a shrinking share of production.

Three companies already control roughly 90 percent of global DRAM output: Micron Technologies, Samsung Electronics, and SK Hynix. Their investment decisions now set the pace of the entire memory market. The source report says all three expect long-term shortages despite major capacity expansions.

All three are also shifting resources toward data centers at the expense of consumer GPUs. Micron stopped direct-to-consumer sales last year to dedicate more resources to AI production. The report links that pivot to continued RAM price increases.

The allocation now runs in one direction. Suppliers are prioritizing AI production, while consumer-device makers absorb higher costs and fewer components.

Could prices ease next year?

Acer CEO Jason Chen told DigiTimes that prices could plateau next year. He cited a potential influx of cheaper Chinese alternatives as a reason.

That outlook remains unproven, according to the report. Other observers question whether AI growth will generate enough sustained demand worldwide to justify the massive production buildout now under way across the memory industry.

The bearish case already has a market precedent. One South Korean publication sent memory stocks tumbling in October 2026 by quoting experts who expected a “flood of new supply” in 2029. The experts predicted the wave would transform the memory market “into a red ocean,” according to CNBC's translation.

Nvidia head Jensen Huang dismissed the prospect of a supply surplus. He told CNBC: “One of the things that's really great about memory is that you can always use more of it.”

The split between Huang and the bearish forecasters turns on demand after new capacity arrives. If AI data centers keep taking roughly 70 percent of global DRAM, a 2029 production surge may not erase the shortage. If demand slows, the same added supply could produce the “red ocean” those experts described.

For consumers, 2027 remains the firm pressure point. Micron has committed most of its HBM output, Samsung has committed most of its memory output, and Chey Tae-won expects production to trail demand until at least 2030. The 2029 flood will matter only if AI demand does not absorb the new wafer capacity first.