Global chip shortage deepens as factories cut DRAM production

The global chip shortage is set to deepen in 2027 as AI-driven demand outstrips supply, according to a warning from KB Securities, a South Korean financial firm. The squeeze stems from a 60% surge in AI infrastructure spending, which is projected to hit $1.3 trillion next year. Semiconductor giants like Samsung and SK Hynix are already operating with less than 10 days of inventory, a critical threshold that signals supply constraints are nearing a breaking point.

Hyperscalers—cloud providers and AI companies—are accelerating their investments, with spending expected to grow from 14% of total tech budgets in 2025 to more than 40% in 2026. The shift is pushing memory chip demand far beyond current production capacity. DRAM and NAND supplies are already stretched thin, and the gap between demand and supply could widen by more than 10 percentage points next year, according to Kim Dong-won, head of research at KB Securities.

The problem isn’t just volume. The transition to High Bandwidth Memory 4 (HBM4), driven by Nvidia’s upcoming Vera Rubin release, is further tightening supply. HBM4 production consumes capacity that could otherwise be used for standard DRAM, the workhorse memory used in servers, data centers, and consumer devices. Analysts warn that by 2027, memory availability could drop below critical levels, potentially leading to shortages where supply effectively runs out.

AI demand outpaces semiconductor capacity

Kim Dong-won’s assessment is blunt: “Next year, we could even see available memory supplies effectively run out.” The warning reflects a broader trend where AI’s insatiable appetite for processing power is colliding with semiconductor production bottlenecks. Unlike past shortages tied to pandemic disruptions or geopolitical tensions, this one is structural. The demand isn’t just rebounding, it’s accelerating.

Samsung and SK Hynix are expected to set record-high performances over the next three years, but their ability to ramp up output may not keep pace with hyperscalers’ plans. Both companies are also prioritizing shareholder returns, which could limit expansion of memory production lines. Meanwhile, competitors are eyeing opportunities in the squeeze. Huawei, for instance, is building a 12-inch chip plant in Shenzhen, while AMD recently acquired a firm offering software that mimics DRAM in operating systems. This acquisition represents a temporary solution to extend existing supplies.

Supply chain delays halt even AI giants

The strain on memory chips is visible even in high-profile AI projects. OpenAI’s data center project in Korea, where Samsung and Hynix have joined hands with OpenAI, has slowed due to delays in finalizing key details, including location, power supply, and funding structure. The delay shows how supply chain uncertainties are forcing even well-capitalized firms to pause ambitious plans.

Some chipmakers are attempting to bypass the bottleneck. AMD’s acquisition of a software firm suggests a pivot toward virtual memory solutions, though these won’t replace physical DRAM in the long term. Meanwhile, Huawei’s new plant signals China’s push to reduce reliance on foreign suppliers, a move that could further disrupt global markets if it gains traction. The shift away from DRAM toward specialized memory like HBM4 also risks creating new imbalances, as Nvidia’s demand for high-performance chips pulls capacity away from broader markets.

The immediate question is whether 2027’s shortages will trigger a correction in AI spending or force companies to adopt more aggressive cost-cutting measures. Hyperscalers have shown little sign of slowing down, but if memory supplies vanish entirely, even their deep pockets may not be enough to prevent delays. The semiconductor industry’s next challenge isn’t just meeting demand, it’s deciding how to allocate what little supply remains.