By Christopher Mims

If you had put all your savings into a few pallets of computer memory chips a year ago, you'd have at least doubled your money by now. And prices are projected to continue their meteoric rise.

Behind the value of one of the world's fastest-appreciating assets is the voracious appetite of AI companies. These same chips -- mainly what's known as RAM, but also the storage chips often called flash or solid-state memory -- are required for almost every digital device on the planet. And just three companies make more than 90% of them: SK Hynix, Samsung and Micron.

Prices for memory shot up 50% in the last quarter of 2025 and are projected to increase another 40% to 50% by the end of the first quarter of 2026, according to Counterpoint Research, fueled mainly by builders of data centers, who are willing to pay huge premiums.

Because AI firms are crowding out other buyers of memory, unexpected consequences are likely to reverberate across countless industries. Effects could include delayed data centers, higher prices for laptops, TVs and other consumer electronics, and possible chip shortages for automakers that would delay vehicle production, in a potential repeat of the pandemic car crisis.

"I have tracked the memory sector for almost 20 years, and this time really is different," says Avril Wu, senior research vice president at Taipei, Taiwan-based TrendForce, which tracks the global semiconductor industry. "It really is the craziest time ever."

No relief in sight

Analysts last year said limited electrical power was the primary speed bump for new AI supercomputer construction in 2026 and beyond. A memory crunch was barely on their radar.

Samsung slowed construction of a new factory to produce memory just two years ago because of a global lull in demand. The company fast-tracked its completion in late 2025, and it's now expanding capacity in existing fabrication plants.

Market leader SK Hynix said in late October that it had already sold out its entire inventory for all of 2026, but it announced massive investments in new manufacturing capacity, expected to be funded by the record revenues.

Several months ago, Boise, Idaho-based Micron saw "a pretty significant surge in demand" from its data-center customers for 2026 and 2027, says Sumit Sadana, the company's chief business officer. On top of that, there has been a less sharp -- but steady -- increase in demand for all other applications, as device makers cram in ever more memory. Micron just announced it will stop producing its popular brand of PC memory to focus on supplying high-end memory for AI.

On Friday, Micron broke ground on what the company says will eventually be a $100 billion "megafab" in Onondaga County, N.Y.

Consisting of several factories to be built over the next 20 years, the entire development will be devoted to producing memory chips.

Unfortunately, almost none of that new capacity will come online until 2027, and it won't make a meaningful difference in supply until 2028, says Wu. For now, manufacturers are running at full tilt with factories that predate the AI boom. "Whatever wafer production is in place right now is coming from investment from three or four years ago," she adds.

"We foresee a challenging situation in terms of being able to meet customer demand for the foreseeable future," says Micron's Sadana.

As AI startups and tech companies alike continue to try to outdo each other with AI supercomputers -- Elon Musk's xAI recently announced it would invest more than $20 billion in a gargantuan data center complex in Mississippi -- analysts are predicting memory prices won't stabilize for a year or two.

This doesn't mean that memory will run out for Apple, Alphabet's Google, Nvidia, Amazon or other tech titans. Many prepurchase memory far in advance, allowing them to lock in prices and guarantee supply. But ambitious new AI data-center construction projects might have to be built with less memory at first and then upgraded later, says Wu.

In consumer electronics, margins are already razor-thin. Smaller manufacturers will likely have no choice but to raise prices, potentially denting demand. The situation has intensified so rapidly in the past two months that tech research firm IDC issued an update to its year-ahead forecast for smartphones and PCs because the anticipated price hikes are expected to reduce consumer demand. In the new worst-case scenario, smartphone sales in 2026 could dip 5%, and PC sales by nearly 9%, as a result of significantly higher prices for the gadgets.

Makers of automobile electronics, telecom equipment and other components face a separate but related issue: They often require older types of memory that manufacturers are moving away from making. If you run one of these component makers, "you gotta buy a plane ticket and get that allocation from manufacturers right now," says MS Hwang, a research director at Counterpoint Research who has been in the memory industry for more than 30 years. "Those guys are now selling their capacity not only for 2026, but also 2027 and 2028," he adds.

The situation is so dire that some firms are considering buying memory from Chinese manufacturer CXMT, even though U.S. lawmakers have signaled their unease with such deals. Other gadget makers are seeking used memory chips. Caramon, a company that reclaims old memory from decommissioned servers for PCs, is benefiting. The value of its sales went from about $500,000 a month to nearly $900,000 a month in just a few months, says company director Paul Coronado.

AI's hunger continues

What we're seeing is a "permanent reallocation" of supplier capacity toward AI companies -- and away from other devices, write the IDC analysts.

Data centers, both conventional and for AI, will consume more than 70% of the high-end memory chips all manufacturers will produce in 2026, and would take even more if they could, according to TrendForce.

Memory chips tend to keep a low profile in the semiconductor industry, compared with the CPUs and GPUs they serve. Even so, they require much of the same cutting-edge technology now enriching Nvidia, which is helmed by Chief Executive Jensen Huang, and its manufacturing partner TSMC, to the detriment of the once-formidable Intel.

Every square centimeter HBM takes is one less that can be used to make memory for other devices. "Every time we manufacture one extra bit of HBM, we lose three bits of supply of conventional DRAM," says Sadana.

AI is also more demanding: Nvidia's latest systems support up to 288 gigabytes of HBM for every one logic chip, compared with the typical 8GB for smartphones and 16GB for laptops. Since HBM is a more lucrative business, consumer-electronics makers have to fight one another for the leftovers.

While rapid price appreciation will continue for now, it's hard to gauge memory-chip pricing beyond mid-2027, says Hwang. He predicts they will soon be considered one of the pricier components in a device, rising from under 10% to as much as 30% of the total cost of phones and other gadgets.

And as AI companies lay claim to ever more manufacturing capacity, the question is, How much will other manufacturers have to pay for memory? "There is no limit," Hwang says.

Write to Christopher Mims at christopher.mims@wsj.com

(END) Dow Jones Newswires

January 16, 2026 22:00 ET (03:00 GMT)

Copyright (c) 2026 Dow Jones & Company, Inc.