Omdia warns memory shortages to 2027
- Omdia said on April 23 memory shortages will likely last into 2027, as AI demand and the shift to HBM keep DRAM and NAND tight. - The standout number is 62.7%: Omdia now expects 2026 semiconductor revenue growth at that pace, with DRAM nearly doubling and NAND quadrupling. - This matters because the boom is price-led, not volume-led, so buyers face a longer squeeze even as chip revenue surges.
Memory chips are the part of the AI boom that almost nobody outside tech procurement talks about — but they are suddenly the bottleneck. Omdia’s update in late April says the shortage in key memory categories is not a short, ugly patch. It is now expected to last well into 2027. That changes the story from “prices are spiking” to “buyers need a two-year plan.” ### What did Omdia actually say? Omdia raised its 2026 semiconductor revenue growth forecast to 62.7% on April 23, driven mainly by DRAM and NAND. The same update said conventional memory supply will stay constrained through 2026, with meaningful relief unlikely until well into 2027. That is the core news here — not just strong demand, but a longer shortage clock than many buyers were hoping for. (omdia.tech.informa.com) ### Why are DRAM and NAND the problem? DRAM is the fast working memory used while systems are actively computing. NAND is the storage side. AI servers need a lot of both, and they need more of them per machine than older systems did. So even before you get to flashy accelerator chips, memory demand rises hard when hyperscalers and enterprises refresh data-center fleets for AI workloads. (omdia.tech.informa.com) ### Why doesn’t supply just catch up? Because the industry is not simply making “more memory.” A lot of manufacturing focus has shifted toward High Bandwidth Memory, or HBM, which is the premium memory stacked close to AI processors. HBM sells for more, but it also uses capacity differently and tends to yield fewer conventional memory chips. Basically, fabs are chasing the richest part of the market, and that leaves standard DRAM and NAND tighter for everyone else. (omdia.tech.informa.com) ### Why does HBM make the squeeze worse? HBM is not just another memory product on the same shelf. It needs advanced packaging, tight integration, and more complex production flows. Think of it like converting part of a bakery from sandwich bread to custom wedding cakes — revenue per order goes up, but the number of basic loaves coming out can fall. That is why memory revenue can explode even while broad availability stays painful. (omdia.tech.informa.com) ### Is this really a demand story, or just a pricing story? Both, but the catch is that pricing is doing a lot of the visible work. Omdia says DRAM could nearly double in value in 2026, while NAND could quadruple from 2025 levels. That does not mean unit shipments are exploding at the same pace. It means scarce supply and richer product mix are pushing average selling prices sharply higher. (omdia.tech.informa.com) ### Who gets hit first? Data-center buyers can usually pay up. Consumer-device makers have a harder time. Omdia-linked smartphone analysis in April said mobile DRAM and NAND prices rose about 90% quarter over quarter in the first quarter of 2026, with another 30% increase expected in the second quarter. That is the downstream effect — tighter margins, more expensive bills of materials, and eventually pricier devices or weaker profits. (omdia.tech.informa.com) ### So what does this change for companies? It pushes memory from a purchasing detail into a board-level planning issue. If relief is not coming before 2027, buyers cannot assume a near-term normalization. They need longer contracts, more careful inventory strategy, and more realism about server refresh timing. The winners are the companies that locked in supply early. Everyone else may end up paying peak-cycle prices for longer than expected. (cyprus-mail.com) ### Bottom line The important shift is time. A shortage that looked cyclical now looks structural for another year-plus. Omdia is basically saying the memory crunch is becoming part of the AI buildout itself — not a side effect that clears quickly. (omdia.tech.informa.com)