Industry body SEMI forecasts USD $52B in 2026 gear orders for memory fabs, with the harder AI bottleneck now in manufacturing complexity, not capital.
Equipment spending for the factories that make memory chips is on track to cross USD $52 billion in 2026 for the first time, driven by AI demand, according to industry trade group SEMI. The same forecast projects another 11% rise to $57B in 2027 and a 19% compound annual growth rate through 2029.
The spending surge is AI-driven, but the constraint is no longer capital. It is manufacturing complexity. SEMI's own outlook flags that "effective capacity growth remains constrained" by technology migration (the slow, expensive process of moving a chip factory from one manufacturing generation to the next) and increasingly complex production processes for advanced DRAM, HBM stacking, and higher-layer 3D NAND. Worldwide 300mm memory capacity, where 300mm is the standard silicon wafer size used for leading-edge chips, is projected to reach 4.1 million wafers per month in 2026 and 4.2 million in 2027 (itbrief). Those are nominal installed-capacity figures, not effective output.
The money is flowing toward specialized memory AI systems need to function at all. AI accelerators cannot run at scale without high-bandwidth memory (HBM), the stacked DRAM that sits next to GPUs and other AI chips. Servers need DDR5, the current generation of working memory. AI data centers need high-density 3D NAND flash storage to hold the datasets large models train on.
SEMI projects DRAM equipment spending will rise 29% in 2026 to $37 billion, led by HBM and DDR5 (HPCwire, Semiconductor Digest). Spending on 3D NAND equipment is forecast to climb 28% to $14 billion as storage requirements balloon with model size (Telecomlead).
The reframe matters: memory capex is no longer a side investment. It is a precondition for AI deployment. The dollars are committed. The bottleneck has migrated from capital to process physics. Ajit Manocha, SEMI's president and CEO, attributed the spending pattern to AI workloads reshaping investment priorities across the semiconductor supply chain (AEI Dempa).
A few caveats worth flagging: this is a single-issuer forecast, not independently verified spending data, and the figures cover only 300mm fab equipment investment, not total memory capital expenditure. The report does not break out how the orders will be split among the three firms that dominate advanced memory (Samsung, SK hynix, and Micron), a meaningful concentration question left unanswered (SEMI 300mm Fab Outlook). The wire pickups of this forecast (Manila Times) are restatements of SEMI's own numbers, not independent confirmation.
What to watch: SEMI's next quarterly update to the 300mm Fab Outlook, and the capex guidance in upcoming earnings calls from Samsung, SK hynix, and Micron. That is the first real test of whether the orders convert into usable chips fast enough to keep pace with AI's appetite.