Why the Global RAM and SSD Shortage Could Last Until 2028

The short version: AI moved to the front of the line

AI data centers changed the buyer mix for memory. Everstream explains that AI data centers rely on GPUs that require HBM, a three-dimensional stack of DRAM chips, and that HBM requires more wafer area than standard DRAM . TrendForce says suppliers in 1Q26 continued reallocating advanced nodes and new capacity toward server and HBM products to support rising AI server demand .

The result is a structural allocation problem. Capacity that might otherwise support conventional DDR5, LPDDR, client SSDs, PCs or smartphones is being pulled toward AI and server products where demand is stronger and margins are higher . IDC similarly describes the shortage as being driven partly by a reallocation of manufacturing capacity away from consumer electronics toward higher-margin memory solutions for AI, with knock-on effects for device makers and end users .

Why HBM makes ordinary DRAM tighter

HBM is not separate from the DRAM ecosystem. It is built from stacked DRAM and uses specialized manufacturing and packaging resources . As memory makers allocate more capacity to HBM, less of the same manufacturing base is available for traditional DRAM products used in other systems .

Industry reports also identify SK hynix, Samsung and Micron as shifting wafer production toward HBM for AI accelerators . That makes business sense: AI and server memory can command priority, while commodity memory for consumer devices becomes more exposed to allocation limits and price increases .

Why SSDs are caught in the same squeeze

SSDs are affected because they depend on NAND flash, and NAND is also being pulled toward data-center use. TrendForce says NAND flash demand is increasingly polarized between consumer and AI applications, with enterprise SSDs becoming the largest segment; in the same 1Q26 outlook, it forecast client SSD prices to rise by more than 40% .

Storage suppliers have also warned of allocation constraints driven by data-center, hyperscaler and AI server build demand, with DRAM and NAND prices rising and extended lead times expected . In other words, the SSD issue is not just retail demand for laptop upgrades. Consumer and client SSD buyers are competing with enterprise storage orders tied to cloud and AI infrastructure .

Hyperscalers are locking up supply early

Large cloud service providers have both the budget and the urgency to reserve memory supply ahead of other buyers. TrendForce says U.S.-based cloud service providers locking in capacity are widening the DRAM supply-demand gap and forcing other buyers to accept higher prices . TechwireAsia reported that AI demand and limited capacity are expected to keep memory prices high and shipments weak through at least 2027 .

That helps explain why shortages can persist even if total output rises. New supply may arrive, but the first claim on it often goes to AI servers, HBM, server DRAM and enterprise SSD contracts rather than the consumer PC, phone or client SSD channels .

New factories cannot fix the gap quickly

Memory production capacity cannot be added as quickly as a purchase order. TechwireAsia reported that meeting demand would require memory production to expand by around 12% annually through 2027, while current growth was about 7.5%; the same report described a roughly 40% gap and output that would meet only about 60% of demand by 2027 .

Gigazine’s summary of related reports says most added facilities were not expected to become operational until 2027 or, at the latest, 2028 . That factory timing is a major reason 2028 appears in shortage forecasts: a fab expansion approved today does not instantly create qualified, high-yield DRAM or NAND supply .

The bottleneck spans more than one product

The shortage is harder to route around because it spans multiple memory and storage categories at once. Reports have described severe tightness across HDDs, DRAM, HBM and NAND in 2026 . When several adjacent product classes are constrained at the same time, buyers have fewer easy substitutes.

That matters for both enterprises and consumers. Enterprises may face longer procurement cycles for servers and storage, while PC, smartphone and client SSD markets are more exposed to higher component costs as AI infrastructure absorbs more of the available supply .

Could it end before 2028?

Yes. The 2028 timeline is a forecast, not a fixed date. Tightness could ease earlier if AI infrastructure spending slows, if long-term orders are digested, or if new capacity comes online faster than expected. It could also last longer if cloud and AI customers keep locking up HBM, server DRAM, NAND and enterprise SSD supply faster than producers can expand output .

For buyers, the practical takeaway is to plan around elevated prices and longer lead times rather than assuming a quick return to cheap RAM and SSDs. TrendForce’s 1Q26 outlook forecast sharp quarterly price increases for server DRAM and client SSDs, and TechwireAsia reported sustained memory cost pressure through at least 2027 . The most accurate way to read 2028 is as a risk window: plausible under current AI demand and capacity assumptions, but still dependent on how fast supply and demand move from here.