The AI frenzy is driving a memory chip supply crisis

FE Team | Published: December 03, 2025 23:58:46

An acute global shortage of memory chips is forcing artificial intelligence and consumer-electronics companies to fight for dwindling supplies, as prices soar for the unglamorous but essential components that allow devices to store data, reports Reuters. Japanese electronics stores have begun limiting how many hard-disk drives shoppers can buy. Chinese smartphone makers are warning of price increases. Tech giants including Microsoft, Google and ByteDance are scrambling to secure supplies from memory-chip makers such as Micron, Samsung Electronics and SK Hynix, according to three people familiar with the discussions.
The squeeze spans almost every type of memory, from flash chips used in USB drives and smartphones to advanced high-bandwidth memory (HBM) that feeds AI chips in data centers. Prices in some segments have more than doubled since February, according to market-research firm TrendForce, drawing in traders betting that the rally has further to run.
The fallout could reach beyond tech. Many economists and executives warn the protracted shortage risks slowing AI-based productivity gains and delaying hundreds of billions of dollars in digital infrastructure. It could also add inflationary pressure just as many economies are trying to tame price rises and navigate US tariffs.
"The memory shortage has now graduated from a component-level concern to a macroeconomic risk," said Sanchit Vir Gogia, CEO of Greyhound Research, a technology advisory firm. The AI build-out "is colliding with a supply chain that cannot meet its physical requirements."
This Reuters examination of the spiraling supply crisis is based on interviews with almost 40 people, including 17 executives at chipmakers and distributors. It shows industry efforts to meet voracious appetite for advanced chips - driven by Nvidia and tech giants like Google, Microsoft and Alibaba - created a dual bind: Chipmakers still can't produce enough high-end semiconductors for the AI race, yet their tilt away from traditional memory products is choking supply to smartphones, PCs and consumer electronics. Some are now hurrying to course-correct.
Details of the global scramble by tech firms and price increases described by electronics retailers and component suppliers in China and Japan are reported here for the first time.
Average inventory levels at suppliers of dynamic random-access memory (DRAM) - the main type used in computers and phones - fell to two to four weeks in October from three to eight weeks in July and 13 to 17 weeks in late 2024, according to TrendForce.
The crunch is unfolding as investors question whether the billions of dollars poured into AI infrastructure have inflated a bubble. Some analysts predict a shakeout, with only the biggest and financially strongest companies able to stomach the price increases.
One memory-chip executive told Reuters the shortage would delay future data-center projects. New capacity takes at least two years to build but memory-chip makers are wary of overbuilding for fear it could end up idle should the demand surge pass, the person said.
Samsung and SK Hynix have announced investments in new capacity but haven't detailed the production split between HBM and conventional memory.
SK Hynix has told analysts that the memory shortfall would last through late 2027, Citi said in November.
"These days, we're receiving requests for memory supplies from so many companies that we're worried about how we'll be able to handle all of them. If we fail to supply them, they could face a situation where they can't do business at all," Chey Tae-won, chairman of SK Hynix parent SK Group, said at an industry forum in Seoul last month.
OpenAI in October signed initial deals with Samsung and SK Hynix to supply chips for its Stargate project, which would require up to 900,000 wafers per month by 2029. That's about double current global monthly HBM production, Chey said.
Samsung told Reuters it is monitoring the market but wouldn't comment on pricing or customer relationships. SK Hynix said it is boosting production capacity to meet increased memory demand.
Microsoft declined to comment and ByteDance didn't address questions about the chip strain. Micron and Google didn't respond to comment requests.
After ChatGPT's release in November 2022 ignited the generative AI boom, a global rush to build AI data centers led memory makers to allocate more production to HBM, used in Nvidia's powerful AI processors.
Competition from Chinese rivals making lower-end DRAM, such as ChangXin Memory Technologies, also pushed Samsung and SK Hynix to accelerate their shift to higher-margin products. The South Korean firms account for two-thirds of the DRAM market.
Samsung told customers in May 2024 that it planned to end production of one type of DDR4 chips - an older variety used in PCs and servers - this year, according to a letter seen by Reuters. (The company has since changed course and will extend production, two sources said.)