AI Demand and Limited Capacity Expansion Reinforce Each Other, Driving DRAM and NAND Prices Higher Together

Overview

The report analyzes prices, supply and demand, product mix and capacity expansion roadmaps for the global memory market in May 2026. The core view is that AI training and inference, cloud service provider stocking and panic buying are amplifying supply-demand gaps in DRAM, HBM and NAND, while newly built fabs require years to generate effective output. Short-term price increases benefit memory suppliers' revenue and profits, but will raise bill-of-materials costs for PC, smartphone, consumer electronics and automotive manufacturers. Long-term supply will be jointly determined by new capacity and yield ramps from Samsung, SK hynix, Micron, Kioxia and Chinese manufacturers.

Core views

First, memory price increases have spread from DRAM to NAND and HBM, and there are still no clear signs of relief in 2Q26. Second, HBM's consumption of wafer capacity, long-term volume locking by cloud service providers and growth in enterprise SSD demand are making traditional end products face more intense capacity competition. Third, short-term supply increases mainly rely on utilization improvements, yield improvements, equipment replacement, cleanroom retrofits and cooperation with existing fabs, while effective supply from large new fabs is mainly concentrated in 2027 to 2029. Fourth, Chinese memory manufacturers are expanding capacity, exports and coverage of the local AI ecosystem, which may reshape overseas market competition over the long term. Fifth, the industry upcycle is not without an endpoint, and future concentrated capacity commissioning, demand destruction, inventory accumulation and price-cycle reversals still warrant vigilance.

Analysis framework

The report combines monthly and quarterly contract price tracking, price matrices by application product, wafer capacity and product mix, process and yield economics, supplier capacity expansion schedules, market share and end-device bill-of-materials costs to assess supply and demand, and tests the market narrative that “this cycle is different from the past” against historical memory cycles.

Methodology notes

• Supply-Demand AnalysisCapacity-Demand Gap Analysis

  Compare bit shipment growth, wafer capacity growth and the ramp-up pace of newly built fabs.

  The report argues that demand growth is faster than effective capacity expansion, and that new fab construction, equipment installation and yield ramping involve multi-year lags, so short-term supply elasticity is limited.

• Price AnalysisContract Price Tracking by Application

  Track quarterly contract prices for DRAM and NAND by mobile devices, PCs, servers and consumer electronics.

  This method is used to identify the sequence in which price increases spread, the degree of pressure on end devices and the bargaining power of different products.

• Industry Structure AnalysisProduct Mix and Capacity Allocation Analysis

  Assess competition for wafer capacity and unit-wafer economics among HBM, general-purpose DRAM and NAND.

  HBM consumes more capacity and has higher strategic priority. Suppliers' tilt toward HBM and cloud service providers will tighten supply of traditional memory products.

• Scenario AnalysisCapacity Expansion Roadmap Analysis

  Map construction, equipment installation and output timing from 2026 to 2029 by supplier and fab project.

  Assess the supply relief window and execution risks through project progress at Samsung, SK hynix, Micron, Kioxia, CXMT and YMTC.

• Cycle AnalysisHistorical Cycle Analogy

  Compare current AI demand, long-term agreements and panic buying with previous memory upcycles.

  The report cautions that technological change has strengthened demand in this cycle, but concentrated capacity expansion, inventory excess and sharp price declines remain traditional cyclical risks for the memory industry.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Samsung Electronics

  A major global DRAM, NAND and HBM supplier whose capacity expansion and HBM4 delivery directly affect industry supply.

  Strengths

  Comprehensive product line with synergies across logic, FinFET and 3D technologies; the report expects it may generate the industry's largest profits in 2026 to 2027.

  Weaknesses

  Needs to catch up with competitors' advanced HBM and capacity pace through P4, P5 and subsequent projects.

  Comparison

  Its scale and technology coverage are more comprehensive, while SK hynix is more aggressive in HBM leadership and DRAM shipment targets.

  Risks

  HBM execution falling short of expectations, new fab delays, demand decline and price reversal caused by concentrated capacity expansion.

• SK hynix

  A leading HBM supplier that plans to pursue the No. 1 position in DRAM shipments through the M15X and Yongin projects.

  Strengths

  Has HBM market leadership, with four fabs planned at Yongin and a clear capacity expansion roadmap.

  Weaknesses

  Logic collaboration capabilities and some technology roadmaps remain challenges, and capacity expansion projects are capital intensive.

  Comparison

  Its HBM competitive position is stronger than most peers, but Samsung's comprehensive technology capabilities and Micron's improved execution create pressure.

  Risks

  HBM product mix adjustments, project delays, Korean labor issues and outflow of technical talent.

• Micron

  A major DRAM, HBM and NAND supplier that shortens capacity expansion timelines through partnerships and new fabs.

  Strengths

  The report recognizes its execution in 1b process HBM4 yield and performance, as well as early adoption of PCIe Gen 6.

  Weaknesses

  Overall scale is smaller than Samsung and SK hynix, and some incremental capacity depends on external fabs and technology transfer.

  Comparison

  Its expansion strategy places more emphasis on cooperation and utilization of existing facilities, which may enable some output faster than purely greenfield projects.

  Risks

  Integration of PSMC cooperation, technology transfer delays, new fab ramping and capital expenditure returns falling short of expectations.

• Kioxia Holdings (285A.T)

  A major NAND supplier whose Kitakami Fab 2 began production in 1Q26 and has shifted toward more active expansion.

  Strengths

  Has a foundation in NAND technology and scale, and emphasizes productivity improvement and capital returns while expanding capacity.

  Weaknesses

  Subsequent fab planning has not yet been fully disclosed, and its concentration in NAND makes the business more cycle-sensitive.

  Comparison

  Compared with multi-product suppliers such as Samsung, Kioxia is more dependent on the NAND and enterprise SSD cycle.

  Risks

  NAND price reversal, declining returns on capacity expansion, and rising costs for high-layer stacking and hybrid bonding.

• CXMT

  A core participant in China's DRAM capacity expansion and the build-out of the local AI memory ecosystem.

  Strengths

  Supported by the local industrial chain and demand, with plans to expand exports and accelerate the transition from R&D to mass production.

  Weaknesses

  Advanced process capacity and yields remain constrained, and multiple patterning increases costs and execution difficulty.

  Comparison

  It still has a technology gap with the world's top three DRAM manufacturers, but its expansion speed and local market support create a competitive variable.

  Risks

  Process yield, equipment restrictions, changes in capital market plans and international trade constraints.

• YMTC

  A major Chinese NAND supplier that is exploring a new fab capable of producing both NAND and DRAM.

  Strengths

  Its existing Wuhan fab has scale, and Fab3 is planned to generate wafer output in early 2027.

  Weaknesses

  DRAM progress remains slow, and there is uncertainty around the product mix and mass production efficiency of the new project.

  Comparison

  Its maturity in China's NAND market is higher than that of its DRAM business, forming both complementarity and potential overlap with CXMT.

  Risks

  Equipment access, technology iteration, yield ramping, international restrictions and overly rapid capacity expansion.

• PC and smartphone manufacturers

  The main cost bearers of DRAM and NAND price increases.

  Strengths

  Leading brands can buffer pressure through scale procurement, price increases, product premiumization and specification adjustments.

  Weaknesses

  Price-sensitive products find it difficult to fully pass on costs, and small and mid-sized manufacturers have weaker supply assurance capabilities.

  Comparison

  PCs face the most pronounced cost pressure, followed by mobile devices, consumer electronics and automobiles.

  Risks

  Gross margin decline, lower sales volumes, order delays and market share concentration toward leading brands.

Key data

• 64GB DDR5 server RDIMM price$450 in 4Q25; $927 in 1Q26; expected $1,350 in 2Q26; expected $1,485 in 3Q26Shows that server DRAM prices are rising rapidly, driven by AI demand and tight supply.
• 2Q26 mobile DRAM price changeLPDDR5 and LPDDR4 are expected to rise 80% to 85% quarter-on-quarterMobile device manufacturers will face significant bill-of-materials cost pressure.
• 2Q26 server NAND price changeNVMe is expected to rise 50% to 60% quarter-on-quarterEnterprise SSD demand continues to lead, with AI inference and KV cache creating incremental demand.
• Total DRAM capacity1.731 million wafers/month in 2024; 1.974 million wafers/month in 2025; expected 2.163 million wafers/month in 2026Capacity growth remains insufficient to quickly eliminate the demand gap.
• HBM capacity as a share of total DRAM capacity17% in 2024; 21% in 2025; expected 26% in 2026Increased HBM allocation will compress available capacity for general-purpose DRAM.
• Long-term growth gapDRAM bit shipments grew at a CAGR of about 17% from 2020 to 2027, while capacity grew about 7.5%The report estimates that maintaining about 20% bit growth in the future will require about 12.3% capacity expansion, which is difficult to achieve in the short term.
• Cloud service provider volume lockingLong-term agreement terms of 3 to 5 years; suppliers may allocate 50% to 70% of output by 2030Leading cloud service providers are prioritizing locked-in supply, which may further squeeze small and mid-sized customers and consumer electronics manufacturers.
• Potential HBM4e price increaseAt least 70%The report believes that after general-purpose memory prices stabilize in 2H26, suppliers may significantly raise HBM4e prices.
• Micron's new Idaho fab in the United StatesPlanned ramp in 2027, with potential capacity of up to about 100,000 wafers/monthThis is a medium-term supply addition, and its actual contribution depends on construction and yield ramping.
• CXMT long-term capacity expansion targetPlans to expand to about 2x by 2030 and about 3x by 2035Increased Chinese DRAM supply may alter the global competitive landscape and price cycle.

Impact & implications

Memory suppliers are expected to gain stronger revenue and profits from price increases, product mix upgrades and AI demand, but technical execution, yield and capacity expansion discipline will determine the degree of benefit. Cloud service providers and leading end-device brands have stronger supply assurance capabilities due to procurement scale and long-term agreements, while small and mid-sized manufacturers may be forced to reduce storage capacity, delay orders or exit some price-sensitive markets. PCs and smartphones will bear the cost shock first, followed by automotive and consumer electronics. Equipment and materials suppliers will benefit in the medium term from the construction of large fabs, but if capacity is released in a concentrated manner from 2027 to 2029, memory prices may re-enter a downcycle.

Risks

• Tensions and insufficient mutual trust between the United States and China may create black swan risks for the supply chain.
• Labor disputes in South Korea and competition for highly paid talent may cause production disruptions or technology outflow.
• Memory prices rising too quickly may trigger specification downgrades, delayed procurement and demand destruction at end markets.
• The concentrated commissioning of large fabs from 2027 to 2029 may lead to inventory excess and sharp price declines.
• New fab construction, equipment installation, technology transfer and initial yields falling short of expectations may delay supply release.
• Increasing complexity in advanced NAND stacking, hybrid bonding and HBM technologies may push up costs and cause product delays.
• Panic buying and long-term volume locking may amplify short-term shortages, and may also create inventory pressure when the cycle weakens.
• Chinese manufacturers' capacity expansion scale exceeding previous expectations may intensify overseas market competition and long-term price pressure.

What to watch

• Whether actual contract prices for DRAM, NAND and HBM in 2Q26 and 3Q26 reach the report's forecasts.
• Whether general-purpose memory prices can stabilize in 2H26, and whether HBM4e achieves a price increase of at least 70%.
• The coverage scale of 3- to 5-year long-term agreements by cloud service providers and the proportion of supplier capacity allocated to them.
• Yields and output at Samsung P4, SK hynix M15X, Micron's partner fabs and Kioxia Kitakami Fab 2.
• Construction and ramp-up progress at Samsung P5, SK hynix Yongin, Micron Idaho, CXMT's new campus and YMTC Fab3.
• Whether smartphone and PC manufacturers continue to reduce storage configurations, raise selling prices or delay orders.
• The growth rate of demand for enterprise SSDs, high-bandwidth flash and KV cache.
• Certification and procurement progress for Chinese memory products entering global original equipment manufacturer supply chains.
• The impact of initial wafer yields, equipment restrictions and advanced process transitions on actual bit output.
• Whether inventory, utilization rates and order cancellations show signs of a cyclical peak.