A global surge in memory-chip prices is exposing the resource strains behind the seemingly unstoppable rise of artificial intelligence. From Shenzhen’s Huaqiangbei electronics bazaar to data-center hubs in the U.S. and South Korea, the supply chain that feeds AI’s hunger for data storage is showing signs of stress.
In China’s largest electronics market, 16-gigabyte DDR4 memory sticks that cost 180 yuan earlier this year now fetch as much as 420 yuan. A one-terabyte solid-state drive has doubled to around 620 yuan. Merchants have adopted day-to-day procurement strategies to avoid the risk of being caught with costly inventory. What looks like a local scramble reflects a global squeeze: the price of DRAM chips has jumped 171.8% in a year, while NAND Flash is up nearly 100%, according to researcher TrendForce.
The rally is fueled by the explosion of demand for memory used in artificial-intelligence training and inference, even as production capacity shifts toward new, high-bandwidth products. Micron Technology estimates that a single AI server now consumes eight times the DRAM and three times the NAND of a conventional one. Training large language models can require up to five terabytes of storage for each system, while inference workloads demand vast amounts of high-speed memory. That structural jump has upended the market’s traditional balance.
Suppliers have been forced to make hard choices. Beginning in late 2023, Samsung Electronics, SK Hynix and Micron started cutting conventional DRAM output to free capacity for high-bandwidth memory and DDR5 chips optimized for AI. Global HBM production doubled in 2024 yet still fell short. At the same time, several overseas producers discontinued older lines such as DDR4 and LPDDR4X, further tightening supply.
“The fundamental driver behind this round of price increases is the end of the DDR4 product lifecycle, coupled with early stockpiling linked to geopolitical uncertainty,” said TrendForce senior analyst Xu Jiayuan. Cloud-service giants are hoarding inventory to hedge against potential disruptions, amplifying real-world shortages in a classic feedback loop.
With inventories across the value chain at historic lows, analysts expect volatility to persist at least through the first half of 2026. Smartphones, personal computers and servers are already feeling the pinch. Phone makers are reassessing pricing strategies, while PC assemblers are caught between rising component costs and soft consumer demand. The server sector is suffering the worst delays: lead times for AI machines have stretched from roughly a month to as long as 16 weeks, slowing infrastructure build-outs worldwide.
The reshuffling of the memory industry underscores a deeper transformation in semiconductors. Once treated as simple storage media, memory chips have become central to computing performance. The shift in their strategic role is likely to alter global value-chain dynamics for years to come.
Behind the immediate price squeeze lies a less visible bottleneck: rare-earth metals. Once auxiliary materials, they are emerging as strategic resources for advanced chip production. In August 2025, China’s Ministry of Commerce expanded its export-control list to include rare-earth items used in logic chips below 14 nanometers and memory chips with 256 layers or more. The move formalized the minerals’ importance to the digital economy and sharpened focus on China’s dominance in supply.
Rare-earth elements are also crucial for next-generation storage technology. Researchers at the University of Chicago recently demonstrated a quantum-memory device using single-atom defects in crystals doped with lanthanum and praseodymium.
“Rare-earth elements have specific electronic transitions that allow optical control at precisely tuned wavelengths,” said physics professor David Awschalom, who led the study. “By exciting lanthanides with lasers and capturing electrons in oxide-crystal defects, we can represent binary data within a cubic millimeter.”
Even in conventional manufacturing, rare-earth compounds are indispensable. Cerium-based polishing slurries are widely used in chemical-mechanical planarization, one of the largest rare-earth applications in chipmaking.
As NAND-flash stacking rises beyond 500 layers, wafer-surface flatness requirements grow ever tighter, increasing cerium demand. Small quantities of lanthanum and yttrium are also added to transistor gate dielectrics to fine-tune electrical properties. While the per-chip usage is minimal, the sheer volume of global production translates into a stable and growing consumption base.
China’s position in the rare-earth supply chain is unrivaled. It holds about 36% of known global reserves yet at times has produced more than 80% of total supply, according to the U.S. Geological Survey. More importantly, Chinese refiners account for more than 90% of global separation capacity, giving the country unparalleled leverage. The 2025 export-control policy reinforced that dominance and prompted chipmakers abroad to rethink sourcing security.
The tightening link between memory and materials is reshaping commodity markets. Prices of praseodymium-neodymium oxide, the feedstock for NdFeB permanent magnets, have surged over 60% since early 2024 to about 650,000 yuan per ton, Asian Metal data show.
Demand stems from high-end manufacturing and clean-energy technologies that rely on powerful magnets—electric-vehicle motors, wind-turbine generators and industrial robots among them. With global EV sales projected to surpass 25 million units in 2025, the sector alone could require more than 50,000 tons of NdFeB magnets, each car using two to three kilograms.
Heavy rare-earth elements such as dysprosium and terbium—used to improve magnets’ resistance to heat and demagnetization—have climbed more than 60% to roughly 3.2 million yuan per ton. These metals are critical for ensuring EV motors and wind-power systems operate reliably under extreme conditions. Analysts expect demand to remain robust as performance standards rise.
The boom is creating windfalls for producers. Research by CICC forecasts the global rare-earth permanent-magnet market will reach $32 billion this year, up 14% from 2024, with China accounting for about a third. Zhongke Sanhuan reported revenue of 8.76 billion yuan for the first three quarters, up 28%, and net income of 1.24 billion yuan, a 46% jump.
The Beijing-based firm has long-term supply deals with Tesla, BYD and NIO, and 2026 orders for new-energy-vehicle magnets already exceed 1 billion yuan. JL Magnetic Materials’ revenue rose 32% to 9.23 billion yuan, while profit surged 58% to 1.56 billion yuan. The company has signed contracts worth more than 800 million yuan for 2026 delivery with wind-power groups including Goldwind and Ming Yang Smart Energy.
Technology upgrades are helping blunt the impact of soaring input costs. Zhongke Sanhuan has developed low-dysprosium and dysprosium-free formulations that cut heavy-rare-earth use without sacrificing magnetic strength. JL Mag’s grain-boundary-diffusion technique achieves similar performance gains with reduced material intensity and has won certification from leading motor makers. ZHmag has unveiled a “high-abundance rare-earth” alloy using more common elements such as cerium and lanthanum to maintain efficiency at lower cost.
Such innovations point to a broader shift toward resource efficiency and supply resilience. Yet the overarching picture remains one of strain. Analysts increasingly see the memory-chip shortage as structural rather than cyclical, likely persisting until 2027. As AI workloads grow more complex, the push for higher-performance, higher-density memory will intensify, making materials security a central strategic issue.
For the AI economy, computing power and storage capacity are its twin pillars—and rare-earth resources the lifeblood coursing between them. A shortage in any one link risks destabilizing the entire system. The recent memory-chip rally is more than a price story; it’s a stress test of how the global technology supply chain copes when the building blocks of intelligence itself become scarce.
