The insatiable appetite of AI leaders is causing significant pressure on the demand for computer memory, sending prices soaring and raising the risk of more expensive PCs and smartphones for end consumers.
A new specter of shortage looms over global IT. This time, it’s not about semiconductors, but about computer memory. RAM memory sticks, coveted by OpenAI, Google, Meta and all the AI giants to equip their brand new data centers, are in very high demand. According to data from TrendForce, a consultant, the main computer memory suppliers saw their average inventories fall from 13 to 17 weeks at the end of 2024 to just two to four weeks in October. Consequence: RAM sticks have seen their prices explode by 500% in recent months.
The crisis is further aggravated by the fact that the main suppliers of computer memory, namely Samsung, Micron and SK Hynix, are now snubbing the BtoC markets to focus on the much more lucrative AI market. Micron thus closed its Crucial line, while Samsung’s branch dedicated to memory today generates a turnover close to that of its smartphone division: 18.2 billion compared to 22.8.
Blame it on AI
Computer memory has become an essential component of computers and smartphones, but also game consoles, connected televisions and vehicles. All of these consumer industries need the same memory chips that the AI giants are accumulating to power their data centers.
However, for Samsung, Micron and SK Hynix, the equation is quickly resolved: OpenAI, Google, Meta and others pay higher prices and place massive orders with high visibility over time. Consumer electronics, on the other hand, is characterized by lower margins and fragmented demand among hundreds of products and manufacturers. This realignment means that the increase in costs could well take place over time, being linked to a medium/long term trend and not to a temporary disruption such as a war or an accident.
However, even for AI giants willing to stretch the dollars, getting their hands on enough memory can prove difficult, says Greyhound Research’s Sanchit vir Gogia. “The industry is facing a structural supply constraint, not just a temporary delay. Memory modules that support AI and high-performance computing clusters are no longer arriving on time or in sufficient quantity, even for hyperscalers paying high premiums.”
Consumers will pay the bill
After semiconductors and energy, would computer memory pose a new potential obstacle to the progress of AI? At least that’s what the expert thinks. “In Europe, AI clusters are being postponed. Across Asia Pacific, deployments are being redesigned along the way. In China, corporate orders are being declined due to running out of inventory. This is no longer a simple component issue. It’s the ceiling of the AI economy. Memory has become the limiting factor in global AI expansion.”
Others are more optimistic, like Antoine Chkaiban, consultant at New Street Research, a market intelligence firm. “I don’t think this will be a brake on the deployment of AI, as most of the giants like Nvidia, AMD and Broadcom have anticipated things and already reserved capacity.”
But for consumers, the bill could well prove steep. “All companies that make PCs, smartphones and servers will be affected by the shortage. The result is that consumers will pay more,” said William Keating, director of research at Ingenuity, another research firm. PC makers like CyberPowerPC and Maingear have already announced price hikes, citing the exploding costs of memory and external hard drives.
Increasing production: why it’s not so simple
One way to resolve the crisis would of course be for suppliers to increase their production capacities. Easier said than done. First, building computer component production plants is expensive and time-consuming. It is therefore a long-term investment, and manufacturers, fearful of finding themselves in overproduction in the event of an economic downturn, are reluctant to get started.
“There is currently no reason to believe that we are no longer in a cyclical market. For manufacturers, the question therefore arises of the duration of the current cycle. The answer will determine the extent of the additional production capacity that they will decide to build in the coming years. At this stage, the ratio of investments in relation to current production remains lower than that observed during the 2018 cycle (a year when strong demand from data centers had already been pressure on memory prices, editor’s note)”, estimates Antoine Chkaiban.
The turmoil on the geopolitical scene adds yet another layer of complexity. The Sino-American trade war is generating tensions for Western companies over supplies of rare earths and critical minerals, massively refined and exported by China. Hard drives, for example, rely on magnets made from neodymium, one of the most sought-after types of rare earth. However, the country recently restricted its exports of magnets as a retaliatory measure against the United States. The customs duties put in place by Donald Trump on steel are making the infrastructure around memory (servers, boxes, racks, data centers) more expensive.
Thus, as Sanchit vir Gogia summarizes, “CIOs must now view memory strategy as strategic capital planning rather than a simple operational purchase. Those who anticipate, diversify their sources and rearchitect upstream will stay ahead of the curve. Those who assume availability will face delays, budget shocks and stunted ambition.”
