SAN FRANCISCO — A new threat to the financial system began knocking around the internet last summer: faster, better versions of artificial intelligence algorithms that could compete with humans and shake markets. As the cost of critical ingredients spikes and supplies have fallen to just weeks, producers caution that the squeeze could lead to new production delays and inflation, Dec. 3, 2025.
Prices for some kinds of DRAM and NAND memory have risen by more than half since the end of summer, as manufacturers told customers they could not fill pre-existing production orders, according to market research firms, consultants and analysts who track pricing data. The most bearish in February are Daimler AG (3.819%) and Wabco Holdings Inc., which makes truck braking systems. Deliveries for box ship integrations designed under classic designs lie only on the core composite lines. Average DRAM inventory levels at large suppliers have fallen to a few weeks’ supply in the wake of months, according to an extensive Reuters investigation, while SK Hynix has recently predicted that tightness will extend into late 2027.
How the chip shortage rippled from AI labs to everyday gadgets
Fueling the squeeze is a hasty transition by memory makers to high-bandwidth memory, or HBM, the super-fast chips that rest next to processors made by Nvidia and others for use with A.I. applications. Pursuing those lucrative orders, the companies have pumped the brakes on conventional DRAM for PCs and servers, and on mobile memory for phones — a trend that an earlier Reuters analysis reported was already starving less sexy products.
Now that reallocation is running into a data centre replacement cycle and into higher-than-expected demand for new phones and laptops. Distributors say they are seeing a spike in double-ordering as nervous customers panic-buy memory to ensure future supply, and big PC and smartphone makers have started complaining that AI-driven memory inflation will either raise device prices or shrink features. Never before were we in such a desperate situation. “… Immediate Supply not long in coming.” ( credit @Spinellistanley) pic.twitter.com/K6lGBt8COK — VideoCardz.com (@VideoCardz) October 30, 2020 The severity of the memory chip squeeze was best summed up to Reuters by one executive: “Everyone is begging for supply.
Years in the making: earlier warnings about capacity
Today’s crunch did not emerge from thin air. They’ve been telling a new story since 2023, when foundry giant TSMC warned that it had grossly insufficient advanced packaging capacity to serve the AI accelerator market, slowing the pace of new HBM-equipped chips. An industry memo from Sourceability that year projected that its CoWoS bottleneck would persist even as new facilities were built until at least 2024, a prediction outlined in a 2023 industry briefing.
Memory makers slashed output in 2023 and early 2024 as prices fell, then pivoted sharply when demand for AI took off. According to an April 2024 earnings report, Samsung told investors it will more than triple the supply of HBM-developed chips and accelerate the mass production of new HBM3E parts. This shift toward premium AI memory restored profits but squeezed older DDR4 and mainstream mobile memory further, setting the stage for the current memory chip shortage.
Can new investment keep up with the demand from AI?
Chipmakers insist relief is coming. Samsung, SK Hynix, and Micron have all said they plan to invest billions of dollars in new capacity, while recent reports say Micron has been planning a high-bandwidth memory facility at its Hiroshima site in Japan that could ship around 2028. Yet larger AI models consume more memory per server, and cloud providers are scrambling to field thousands of new accelerators; analysts caution that new capacity could be absorbed as fast as it comes online.
And for the time being, the memory chip shortage is remaking global supply chains in real time. Stockpiling traders are hoarding computer components, the secondhand markets for recycled data-centre equipment have exploded, and smaller electronics manufacturers without long-term contracts with major customers are at risk of being priced out. If shortages and price spikes continue until 2027, the AI boom that helped fuel the crunch could paradoxically delay its own productivity gains, leaving consumers to pay more for everything from cloud services to budget smartphones.
