Memory shortage could stretch toward 2030 as AI absorbs supply

The global memory shortage is no longer just a short-term pricing problem for PC builders. It is becoming one of the clearest hardware bottlenecks behind the AI boom. Demand from AI data centers has pushed Samsung, SK hynix and Micron to prioritize high-bandwidth memory, while conventional DRAM for PCs, smartphones and other devices faces tighter allocation. That shift is why analysts and industry executives now talk about a shortage that could last years, not months.

The most severe warning came from SK Group chairman Chey Tae-won. Reuters reported in March 2026 that Chey said the global chip wafer shortage could persist until 2030 because AI demand continues to outpace supply and new wafer capacity takes at least four to five years to build. The warning is not limited to HBM. Chey also said the industry faces a wafer shortage of more than 20%, which means adding capacity is slower than the demand curve coming from hyperscale AI infrastructure.

AI memory is crowding out ordinary DRAM

The pressure is strongest in high-bandwidth memory, the type of memory used alongside advanced AI accelerators. HBM commands higher prices and is tied to large cloud and GPU contracts, so memory makers have strong reasons to direct more production toward it. The tradeoff is that older, more ordinary DRAM categories can become less available or more expensive. That is where consumers and device makers start to feel the AI buildout indirectly.

Reports citing Nikkei Asia say memory makers may be able to meet only about 60% of demand by the end of 2027. That figure should be treated as a market estimate, not a guarantee, but it lines up with the broader pattern: new fabs and packaging capacity do not arrive instantly, and much of the investment is aimed at AI-grade memory first. Even when a facility is completed, mass production takes time to ramp.

Price pressure may reach more devices

The shortage matters because memory sits inside nearly every computing product. Servers need HBM and DRAM. Laptops and phones need LPDDR and standard DRAM. Gaming handhelds, VR headsets, workstations and networking hardware all depend on memory supply. If AI buyers lock in more output through long-term contracts, smaller customers may face higher prices or less predictable availability.

For the AI industry, memory has become as important as compute. Faster models and larger inference workloads need more bandwidth, not just more processing cores. That is why the market is rewarding memory makers and pushing suppliers to expand. But the same demand also makes the supply chain less flexible.

The practical conclusion is that RAM prices and availability may remain volatile through the second half of the decade. The shortage may ease in individual product categories, but the pressure from AI infrastructure is likely to remain. Consumers may notice it in laptop configurations, upgrade prices or slower price drops. Cloud companies will notice it in capacity planning. Memory has moved from a background component to a hard limit on how quickly AI systems can be built.