A severe and fast-worsening shortage of memory chips is rippling across global tech supply chains, forcing artificial intelligence developers, electronics manufacturers and retailers to compete fiercely for dwindling inventories. Prices for key components—from legacy DRAM to cutting-edge high-bandwidth memory (HBM)—have surged, and analysts warn the bottleneck now threatens broader economic momentum.

Electronics stores in Japan have begun rationing hard drives, while Chinese smartphone makers caution they may soon raise handset prices. At the same time, major tech companies including Microsoft, Google and ByteDance are urgently seeking new supply commitments from memory giants Micron, Samsung Electronics and SK Hynix, according to people familiar with the talks.

The squeeze spans nearly every corner of the memory market. TrendForce data shows prices in some segments have more than doubled since February, drawing in traders betting on further gains. Economists say the shortage could slow AI-driven productivity improvements and delay hundreds of billions of dollars in planned digital infrastructure, even as several economies struggle with inflation and trade tensions.

Sanchit Vir Gogia, CEO of Greyhound Research, said the situation has escalated into “a macroeconomic risk,” noting that the rapid AI build-out “is colliding with a supply chain that cannot meet its physical requirements.”

Interviews with nearly 40 industry participants — including 17 executives at chipmakers and distributors — indicate that the race to produce high-end chips for AI systems has left traditional memory markets dangerously undersupplied. Manufacturers shifted production toward HBM to support Nvidia and hyperscalers such as Google, Microsoft and Alibaba, but the pivot choked output for smartphones, PCs and conventional data centers. Now, many chipmakers are scrambling to rebalance production.

Inventory levels tell the story. Average DRAM stockpiles fell to just two to four weeks in October, down sharply from 13 to 17 weeks late last year. Meanwhile, investor anxiety about an AI-driven bubble is growing as rising memory costs challenge all but the largest firms.

One chip-industry executive said the shortage is likely to postpone future data-center projects. New fabrication capacity takes at least two years to build, yet manufacturers remain hesitant to overexpand for fear of a demand reversal.

Samsung and SK Hynix have pledged new investments but haven’t disclosed the mix between HBM and conventional memory. SK Hynix told analysts the shortfall may last until late 2027. Chey Tae-won, chairman of parent company SK Group, warned that the influx of supply requests is so overwhelming that some customers may struggle to operate at all if allocations fall short.

The supply race is intense. OpenAI recently signed initial agreements with Samsung and SK Hynix to provide chips for its Stargate data-center project, which could require up to 900,000 wafers per month by 2029 — roughly double today’s global monthly HBM output. Samsung said only that it is monitoring the market, while SK Hynix said it is increasing capacity.

Pressure on supplies grew after memory makers publicly retreated from older chip varieties. Samsung told customers last year it would stop making certain DDR4 chips, and although it later reversed course, the signal contributed to tightening supply. Micron similarly informed clients it would phase out shipments of DDR4 and LPDDR4. China’s ChangXin Memory Technologies also reduced DDR4 output.

These cuts came just as the replacement cycle for PCs and traditional data centers accelerated, and smartphone sales outperformed forecasts — a combination that left suppliers unprepared. Samsung raised prices for server memory by as much as 60%. Meanwhile, hyperscalers such as Google, Amazon, Microsoft and Meta made open-ended requests to Micron for as much supply as it could deliver, regardless of cost. Chinese giants Alibaba, ByteDance and Tencent also dispatched delegations to Samsung and SK Hynix to lobby for allocation. “Everyone is begging for supply,” said one person familiar with the discussions.

Despite the strain, chipmaker earnings are soaring. Micron, Samsung and SK Hynix all reported strong performance this year, with Micron beating revenue forecasts and Samsung posting its highest quarterly profit in more than three years. Analysts at Counterpoint Research expect advanced and legacy memory prices to rise roughly 30% through year-end and potentially another 20% in early 2026.

The crunch is hitting consumers as well. Xiaomi and Realme warned they may need to raise smartphone prices, with Realme saying storage costs have risen at an “unprecedented” pace. Some companies may cut back on cameras, processors or batteries to compensate, but storage costs must ultimately be absorbed.

Manufacturers of PCs and components are also feeling the pressure. ASUS has about four months of memory inventory and may adjust pricing, while Taiwanese chipmaker Winbond approved a $1.1 billion capital-expenditure increase to expand production.

Retail markets are responding with rationing and price spikes. In Tokyo’s Akihabara district, stores have limited purchases of hard drives, SSDs and system memory to prevent hoarding. Prices for consumer-grade DDR5 products have tripled since mid-October, pushing many buyers to the secondhand market. Traders in Shenzhen say component prices now fluctuate by the hour.

The shockwaves extend to the U.S., where sellers of recycled low-end server memory report soaring demand from Hong Kong–based brokers supplying Chinese clients. Monthly sales for one California recycler have nearly doubled since September.

With supply not expected to normalize for several years and AI investment still accelerating, the global scramble for memory chips shows little sign of easing — and its economic footprint is growing by the month.