Morgan Stanley: AI Triggers Chipflation, Memory Becomes Structural Bottleneck

Overview

This report explores the 'Chipflation' (chip inflation) phenomenon, where the AI wave is transforming memory chips from cyclical commodities into structural bottlenecks. Morgan Stanley points out that as AI servers become memory-intensive systems, HBM and enterprise SSD demand is exploding, severely squeezing traditional consumer electronics capacity. Although policy support and China capacity expansion may alleviate pressure, constrained by equipment delivery cycles and export controls, the mid-term memory market will face significant structural divergence and supply tightness.

Core views

Demand Structure Dramatic Change: AI is reshaping the memory demand landscape. Report data shows AI cluster HBM usage surged 1800x, AI chip HBM usage grew 7x. Server share of total DRAM demand is expected to climb from 37% in 2023 to 59% in 2028; enterprise SSD share of NAND demand will also skyrocket from 20% in 2021 to 65% in 2028. This shift in demand focus means the memory market is no longer dominated by consumer electronics, but turning towards data center infrastructure. Supply Squeeze and Shortage: Due to HBM consumption of advanced process wafers, conventional memory supply is severely squeezed. Report calculations show that if AI and server demand are prioritized, PCs and smartphones combined may face about 13% DRAM supply gap (about 19bn Gb) by 2027. Supply allocation shows clear hierarchy: Tier 1 hyperscale cloud vendors can lock in supply via strategic contracts, while low-end OEMs and non-AI buyers face spot price volatility, spec cuts, and even new product delay risks. Price Transmission and Macro Impact: Memory price hikes have begun transmitting downstream. Electronic components PPI rose about 30% YoY, although memory cost's direct push on CPI is limited (only about 0.1 percentage points), it constitutes pressure on terminal electronics pricing. Low-end PCs and smartphones, due to weak cost pass-through ability, are more susceptible to demand destruction. Supply Side Variables and China Factor: Mainland China is a potential supply adjustment variable, accounting for about 30% of global net added wafer capacity 2023-2028. In an optimistic scenario (assuming export controls relax), combining YMTC expansion and Samsung Xi'an, Solidigm Dalian fab node migration, China NAND incremental output could account for 17%-33% of global by 2028. But the report emphasizes this is not the base case, core constraint lies in access to advanced lithography equipment. Equipment Side Outlook: Regarding market concerns about ASML capacity, the report believes it is overhyped. Based on company AGM information and new campus production progress, analysts are confident ASML can meet demand for about 90 EUV tools next year, and raised FY28 EUV shipment forecast to 104 units, showing upstream equipment bottlenecks are expected to gradually ease.

Analysis framework

The report adopts a 'top-down demand breakdown + bottom-up supply constraint' analysis framework. First, by quantifying memory usage differences between AI systems and traditional equipment, it establishes the trend of structural demand shift; then constructs a 'Memory Sufficiency Framework', deducting HBM and server-dedicated capacity layer by layer from total supply to calculate the remaining available supply gap for the consumer side. Meanwhile, introduces the 'Allocation Stack' concept to analyze resource acquisition priority for buyers with different bargaining power in a shortage environment, thereby judging the industry chain profit distribution pattern. For the supply side, it distinguishes between 'base capacity' and 'geopolitically adjusted potential capacity', evaluating their actual impact on the market separately.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply-demand framework

  Memory Sufficiency Model (Memory Sufficiency Framework)

  The report does not just look at total supply-demand, but deducts total supply by technical layer (HBM, Server DRAM, Consumer DRAM) to calculate 'remaining available supply' for specific segments. This method can precisely identify structural shortages, avoiding local crises being masked by total balance.

• Competition and Strategy FrameworkValue chain analysis

  Customer Tiered Allocation Stack (Buyer Hierarchy / Allocation Stack)

  When supply is tight, suppliers allocate resources according to customer strategic value. The report divides buyers into Tier 1 to Tier 5, revealing vastly different cost and supply risk exposure between cloud giants with long-term contracts and small factories relying on spot markets in the 'Chipflation' environment.

• Cycle and Prosperity FrameworkCapacity/Equipment Cycle (Juglar)

  Time Lag Effect from Equipment Delivery to Qualified Output

  There is a 1-2 year physical time lag (including installation, validation, yield ramp-up) from ordering lithography machines to finally producing qualified memory chips. This common sense explains why even increasing capital expenditure or policy support now cannot solve short-term structural shortage problems.

Asset mapping & comparison

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

• ASML (ASML.AS)

  Key equipment supplier, capacity release determines industry supply ceiling

  Strengths

  EUV monopoly status, new campus production ensures delivery capability, FY28 shipment expectation raised

  Weaknesses

  Expansion constrained by supply chain components and geopolitical export controls

  Comparison

  Compared to other equipment vendors, ASML is the most core bottleneck to solving memory bottlenecks

  Risks

  Export controls tighten, downstream capital expenditure below expectations

• SK Hynix / Samsung / Micron

  Core manufacturers of HBM and DRAM, directly benefit from AI demand explosion

  Strengths

  HBM technology leading, secured Tier 1 cloud vendor long-term contract lock-in

  Weaknesses

  Advanced process capacity squeezed by HBM, conventional DRAM output limited

  Comparison

  SK Hynix leads in HBM field, Samsung has comprehensive layout in NAND and foundry

  Risks

  Yield ramp-up below expectations, consumer electronics demand collapse drags overall revenue

• YMTC / China Memory Vendors

  Potential supply increment source, highly affected by geopolitics

  Strengths

  Mature process expansion rapid, domestic market demand support

  Weaknesses

  Lack EUV equipment, advanced node mass production hindered

  Comparison

  Compared to Korean vendors, significant gap in high-end HBM field, but competitive in mature NAND

  Risks

  US export controls further upgrade, equipment maintenance and components supply cut off

Key data

• 2027 Consumer Side Memory Gap13%DRAM supply shortage ratio faced by PCs and smartphones due to AI priority allocation
• Server DRAM Demand Share37% → 59%2023 to 2028E, servers replace consumer electronics as the largest DRAM demand source
• AI Cluster HBM Usage Increase1800xCompared to traditional systems, AI cluster high-bandwidth memory usage grows exponentially
• China NAND Potential Supply Share17-33%2028E optimistic scenario China factory incremental output share of global, not base forecast
• ASML FY28e EUV Shipment Forecast104 unitsAnalyst raised forecast value, believes market capacity concerns are excessive

Impact & implications

The report believes the memory market is undergoing a transition from cyclical fluctuations to structural bottlenecks. For AI infrastructure investors, this means accepting higher memory costs and longer supply lock-in cycles; for consumer electronics manufacturers, especially low-end brands, profit margins will be further compressed, even facing risks of products not being released on time. At the policy level, although countries attempt to alleviate pressure through subsidies, constrained by physical construction cycles and equipment bans, it is difficult to change the supply-demand imbalance pattern in the short term. Capacity release from upstream equipment vendors like ASML will be the key variable to alleviate contradictions in the medium to long term.

Risks

• AI capital expenditure slowdown leads to HBM demand below expectations
• Consumer electronics demand shows unexpected collapse due to price hikes
• Geopolitics leads to further tightening of equipment export controls
• ASML and other equipment vendors expansion progress lags behind plan
• China memory capacity breakthrough speed faster than base assumption, triggers price war

What to watch

• Major cloud vendors AI capital expenditure guidance and HBM long-term contract signing status
• ASML quarterly EUV orders and delivery data
• DRAM/NAND spot price trends and inventory level changes
• US semiconductor export control policy update dynamics towards China
• PC and smartphone shipment volumes and per-unit memory load trends