The global memory and storage market is undergoing the most severe supply-demand dislocation in its history, and the conventional cycle no longer describes it. Artificial intelligence has broken the industry's oldest rule - that prices fall as new capacity arrives. DRAM revenue grew 144% in 2026 on bit-supply growth of only around 16%, and contract prices for conventional DRAM rose as much as 95% in a single quarter against a historical peak of roughly 35%. Total memory and storage revenue approximately doubles from 2025 to 2026. The mechanism is structural rather than speculative. High Bandwidth Memory consumes roughly three times the wafer area per bit of a conventional DDR5 die once through-silicon-via overhead, known-good-die loss and base-die area are accounted for. As manufacturers concentrate production on HBM for its profitability, HBM rises from around a fifth of DRAM wafer starts toward 40% by 2037 - squeezing conventional DRAM and NAND supply without any producer deciding to reduce it. Producers are currently meeting only half to two-thirds of core customer demand, HBM is sold out through end-2027, and hard disk capacity is sold out for 2026.
Relief is physically impossible in the near term. Samsung and SK hynix have committed a combined ₩800 trillion ($518 billion) to new Korean fabs, Micron has raised planned US spending above $250 billion, and SK hynix is building a $4 billion HBM packaging facility in Indiana - but a new memory fab requires a minimum of three years to first output and five to seven to reach full capacity. Because every producer is expanding against the same signal, the resulting supply lands within a single window, and the market corrects sharply from 2029 to a trough in 2030 before resuming bit-led growth.
Demand itself does not contract in any year. AI inference has installed a permanent floor: test-time scaling, long-context reasoning and agentic workloads have restructured the memory hierarchy downward, pushing data out of GPU memory into CPU RAM and into an entirely new SSD context tier. Meanwhile, embedded flash's inability to scale below 28nm makes the migration to MRAM, ReRAM and ferroelectric memory a matter of timing rather than choice. Memory has ceased to be a commodity input and become strategic infrastructure - priced, contracted and allocated accordingly.
The Global Memory and Storage Technology Market 2027-2037 is a comprehensive analysis of the global memory and storage industry through a full cycle - the AI-driven supercycle of 2026-2028, the capacity-led correction of 2029-2030, and the bit-led recovery that follows. The report combines rebased revenue forecasts anchored to reported 2026 producer results with detailed technology roadmaps, manufacturing capacity analysis, pricing models and 173 company profiles.
Forecasts are provided annually from 2026 to 2037 across every technology, application and region, with revenue decomposed into bit growth and average selling price so that cyclical and structural drivers can be separated. The report includes upside, base and downside scenarios, a ten-point risk register, wafer-capacity and supply-demand balance modelling, node-migration yield and cost curves, and a full technology-readiness assessment for every emerging memory candidate.
Market segmentation - technologies covered
- DRAM: DDR4, DDR5, DDR6; LPDDR4X, LPDDR5X, LPDDR6; GDDR; SOCAMM and CPU-attached memory; graphics and specialty DRAM; planar node progression (1α to 0d) and 3D DRAM
- High Bandwidth Memory: HBM3E, HBM4, HBM4E, HBM5, HBM6; custom HBM (cHBM); TSV, hybrid bonding and thermal management
- NAND Flash: 3D NAND layer scaling; SLC, MLC, TLC, QLC and PLC; CMOS Bonded Array and Xtacking; High-Bandwidth Flash
- Solid-state storage: enterprise SSD, client SSD, SSD POD/context tier, SSD controllers, EDSFF form factors, NVMe and CXL
- Hard disk drives: nearline, mission-critical and energy-assisted recording (HAMR, MAMR)
- Emerging non-volatile memory: MRAM (STT, SOT, VCMA, embedded); ReRAM/RRAM and CBRAM; FeRAM and HfO₂ ferroelectric/FeFET; PCM and ePCM; NRAM; CeRAM; ULTRARAM; selector-only memory and storage-class memory
- Storage systems, tape and optical archive; processing-in-memory and compute-in-memory
Contents include:
- Market forecasts 2026-2037 across a full cycle: supercycle peak, correction, trough and bit-led recovery, with revenue decomposed into bit growth versus average selling price
- Segment forecasts by technology - conventional DRAM, HBM, NAND, HDD, SSD controllers, storage systems and emerging non-volatile memory
- Forecasts by application and region, covering data centre and AI infrastructure, mobile, PC/client, automotive, industrial, consumer and enterprise storage across eight regions
- Scenario analysis and risk register - upside, base and downside cases with a ten-point risk assessment
- DRAM technology roadmaps - node progression from 1α to 0d, the 6F² to 4F² transition, 3D DRAM integration pathways, capacitor-less and gain-cell designs
- HBM technology - HBM3E through HBM6, custom HBM, TSV stacking, hybrid bonding, thermal management and processor integration
- NAND Flash roadmaps - layer scaling beyond 300 and 1,000 layers, CMOS Bonded Array and Xtacking, TLC/QLC/PLC evolution, High-Bandwidth Flash
- AI inference memory architecture - KV cache offloading, the SSD POD context tier, agentic AI and the shift in CPU-to-GPU memory ratios
- Emerging memory technologies - MRAM (STT, SOT, VCMA), ReRAM, FeRAM and HfO₂ ferroelectrics, PCM, NRAM, CeRAM, ULTRARAM and selector-only memory, with a technology-readiness matrix and displacement scenarios
- Advanced packaging and integration - TSV, hybrid bonding, chiplets, fan-out packaging, processing-in-memory and compute-in-memory
- Supply chain and manufacturing - global wafer capacity by technology and region, fab utilisation, supply-demand balance, next-generation fab requirements, node yield and cost curves
- Regional and geopolitical analysis - China's capacity build-out (CXMT, YMTC), export controls, the 2026 tariff landscape and supply chain regionalisation
- Pricing and economic models - DRAM and NAND price cycles, HBM premium pricing, manufacturing cost structure, gross margin cycle and technology cost roadmaps
- Sustainability - carbon and water footprint by technology, the HBM environmental premium, energy efficiency evolution and circular economy
- 173 company profiles plus long-term technology roadmaps to 2037, including quantum, DNA, photonic and neuromorphic memory. Companies profiled include 3D Plus, 4DS Memory, Adata Technology, Advantest Corporation, Ambiq Micro, AMD, Amkor Technology, ANAFLASH, AP Memory, Apacer Technology, Applied Materials, ASE Group, ASM International, ASML Holding, Atomera, Avalanche Technology, Axelera AI, BeSang, Besi, Celestial AI, Cerebras Systems, CXMT, Crocus Nanoelectronics, Crossbar, d-Matrix, Dnotitia, Dosilicon, eMemory, Etron Technology, ESMT, Everspin Technologies, Expedera, Ferroelectric Memory Company, FERROSemi Technology, Floadia Corporation, Fudan Microelectronics, Giantec Semiconductor, GigaDevice Semiconductor, GlobalFoundries, GlobalWafers and more.....