ASML's roadshow, the memory cycle and Chinese AI models jointly outline the semiconductor demand thesis

Overview

This report is global technology webcast material from Morgan Stanley's European semiconductors team. Its core content includes notes from the ASML CEO/CFO roadshow, discussion of the memory industry cycle, the potential catalyst of Apple AI entering China, progress on the open-source Kimi K3 model, CXMT's STAR Market IPO and next-generation memory innovation. The report covers intersecting themes across semiconductor equipment, memory, AI infrastructure and China's domestic DRAM supply chain.

Core views

The core view is that ASML still has a foundation for long-term structural growth; although the memory cycle has entered a phase of slowing momentum, it appears more likely to be an extended cycle than a rapid collapse; and AI capital expenditure is the key variable for assessing the resilience of memory demand. The report maintains a selectively constructive stance toward the memory industry, while noting that DRAM prices may peak in 4Q26, inventories are beginning to rise and the earnings upgrade rate is declining, implying that the market needs to focus more closely on marginal changes. Chinese AI and domestic DRAM supply have also become important variables affecting the global semiconductor landscape.

Analysis framework

The report adopts a thematic framework, dividing semiconductor investment themes into the ASML management roadshow, AI capital expenditure, the memory price cycle, long-term supply agreements, earnings revisions, valuation multiples, Chinese AI models and CXMT's domestic DRAM supply. The analysis focuses not on a single-company valuation model, but on determining the industry's cyclical position and structural demand through management feedback, industry prices, inventory, earnings revision rates, P/B valuations, IPO financing and product technology roadmaps.

Methodology notes

• Industry Cycle AnalysisThree-Factor Memory Cycle Framework

  AI spending, long-term agreements and cycle position

  The report frames the memory industry around three debates: whether capital is genuinely flowing toward monetizable AI infrastructure, whether the market should re-rate the industry because of long-term agreements, and whether inventory and price signals indicate a cyclical inflection point.

• Management InterviewASML NDR Roadshow Notes

  Long-term growth drivers

  Long-term drivers such as demand expansion, value-based pricing, flexibility and medium-term margin leverage are extracted from feedback from the ASML CEO/CFO roadshow.

• Relative ValuationNTM P/B and Comparable Company Table

  Whether valuation will re-rate

  The report compares changes in the P/B ratios of memory companies such as Samsung and SK Hynix, noting that multiples have declined from their highs but remain above long-term ranges, to assess whether the market has fully priced in the cycle.

• Technology Roadmap AnalysisAI Model and Memory Innovation Framework

  Compute, context windows and memory bandwidth

  Kimi K3, HBM, DDR5/LPDDR5X and memory bandwidth constraints are used to assess the pull from AI workloads on semiconductor demand.

Asset mapping & comparison

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

• ASML HOLDING NV

  Core covered company; the report includes management roadshow notes and an EUV shipment outlook.

  Strengths

  Long-term growth drivers are clear, with value-based pricing, end-market expansion and medium-term margin leverage emphasized; the report mentions expected EUV shipments of 92 units and 104 units in FY27 and FY28, respectively.

  Weaknesses

  Near-term equipment demand remains affected by the semiconductor capital expenditure cycle and the pace of customer capacity expansion.

  Comparison

  Compared with memory manufacturers, ASML is more exposed to upstream equipment and advanced-process bottleneck assets, combining cyclical sensitivity with structural barriers to entry.

  Risks

  Slower customer capital expenditure, changes in the pace of EUV orders, geopolitics and export restrictions to China.

• Samsung Electronics

  Global memory comparable company, used to assess the memory cycle and valuation.

  Strengths

  Memory leader, benefiting from DRAM/NAND and AI-related memory demand.

  Weaknesses

  The P/B ratio has declined to approximately 1.7x, indicating that the market is already concerned about the cyclical peak and slowing earnings momentum.

  Comparison

  Compared with SK Hynix, Samsung has a lower valuation multiple, but remains exposed to the common memory price and inventory cycle.

  Risks

  Peak DRAM prices, rising inventories and declining earnings upgrade momentum.

• SK Hynix

  Representative asset for the HBM and memory cycle.

  Strengths

  AI and HBM demand provide structural support.

  Weaknesses

  One-year forward EPS has recently declined, while the P/B ratio is approximately 2.5x and has retreated from its recent high.

  Comparison

  Compared with Samsung, SK Hynix has more direct exposure to the AI memory theme, but its valuation and earnings expectations are also more sensitive.

  Risks

  HBM demand below expectations, a DRAM price peak and valuation contraction.

• CXMT

  Variable for China's domestic DRAM development and supply expansion.

  Strengths

  China's only large-scale DRAM manufacturer, planning to raise funds through a STAR Market listing, with strategic shareholders including Alibaba Cloud, Meituan and Xiaomi.

  Weaknesses

  Revenue is primarily derived from DDR/LPDDR; the structural HBM gap has not been resolved, and the node is D1z without EUV.

  Comparison

  The report believes CXMT will fill the commodity DDR5 gap left as the Big Three shift toward HBM, while HBM remains a structural constraint.

  Risks

  Technology-node catch-up, earnings sustainability, valuation pressure and insufficient advanced memory capabilities.

• Apple Inc.

  The report title identifies Apple AI entering China as a key catalyst for AAPL.

  Strengths

  If Apple AI is launched in China, it could stimulate end-device AI demand and related hardware upgrades.

  Weaknesses

  The body provides limited available evidence and does not offer detailed implementation timing or quantified contribution.

  Comparison

  Unlike upstream semiconductor equipment and memory, Apple is more exposed to end-market demand catalysts.

  Risks

  Regulatory approval, user adoption, commercialization of AI functions and supply-chain execution.

Key data

• Report Date2026-07-21The cover shows July 21, 2026 11:26 AM GMT, and the body states that the information is current as of 21 July 2026.
• Kimi K3 Parameter Scale2.8T total parametersMoE architecture, with 16 of 896 experts activated.
• Kimi K3 Context Window1M token context windowThe report also mentions native visual capabilities and plans for open-source weights.
• Kimi K3 Pricing$3 / $15 per M tokensThese correspond to input and output token prices, respectively.
• Kimi K3 Risk IndicatorApproximately 51% independent hallucination rateThe report indicates that a verification process should be retained.
• DRAM Price OutlookMay peak in 4Q26The year-over-year growth rate of DRAM contract prices has declined from the cyclical high, consistent with a peak around 4Q26.
• Earnings Revision RateDeclined from a peak of 92% to 77%Net upgrades remain positive, but momentum in the upgrade cycle has weakened.
• DRAM and NAND InventoryRose in 2Q26Driven primarily by module makers.
• Samsung P/B1.7xThe report states that the P/B multiple has declined from its recent high.
• SK Hynix P/B2.5xStill above the long-term range, but down from the high.
• CXMT Fundraising SizeApproximately ¥57.9bn, up to approximately $9.8bnShanghai STAR Market IPO, with an offering price of ¥8.66 per share; trading begins July 27.
• CXMT Implied ValuationApproximately ¥579bn, approximately $85bnThe report describes it as the largest STAR Market IPO ever and Asia's largest IPO in 2026.
• CXMT Business MixApproximately 99% of 2025 revenue came from DDR/LPDDRThe node is D1z DDR5/LPDDR5X, without EUV.
• CXMT Valuation MetricsFY26e P/E approximately 5.8x, P/B approximately 3xTrailing P/E of approximately 309x was also disclosed.
• Morgan Stanley Rating Time Horizon12-18 monthsThe disclosure page states that target prices and relative ratings generally correspond to a period of 12 to 18 months.

Impact & implications

For investment implications, ASML's long-term thesis remains supported by EUV shipments, value-based pricing and demand for advanced processes; upside in the memory chain requires caution regarding peak prices, rising inventories and slowing earnings revisions; AI capital expenditure remains a key confirmation indicator for semiconductor demand; and CXMT's listing and Kimi K3's release show accelerating Chinese self-reliance in DRAM supply and AI models, although technology, bandwidth and model reliability remain constraints.

Risks

• The year-over-year growth rate of DRAM contract prices has declined from its high, and prices may peak around 4Q26.
• DRAM and NAND inventories rose in 2Q26, driven primarily by module makers.
• The earnings upgrade rate declined from a peak of 92% to 77%, indicating slowing momentum in the upgrade cycle.
• Although memory valuations have declined, they remain in a relatively high range and could continue to contract if prices or EPS are revised downward.
• Kimi K3 has an independent hallucination rate of approximately 51%, and progress in model capabilities still requires reliability validation.
• Although CXMT can supplement commodity DDR5 supply, the HBM gap remains a structural constraint.
• Morgan Stanley discloses investment-banking or potential business relationships with several covered companies; investors should be aware of conflicts of interest.

What to watch

• 2Q26 and subsequent capital expenditure by hyperscalers, to verify whether AI infrastructure demand remains sustainable.
• Year-over-year changes in DRAM contract prices and whether a price peak emerges around 4Q26.
• Whether DRAM and NAND inventories continue to accumulate at module makers.
• Whether the net earnings revision rate for memory companies continues to decline from 77%.
• Whether ASML's FY27/FY28 EUV shipment expectations remain around 92 units and 104 units.
• CXMT's trading performance after its July 27 listing, fundraising execution and pace of capacity expansion.
• Actual benchmarks, pricing competition and improvements in the hallucination rate following the release of Kimi K3's open-source weights.
• Progress in launching Apple AI in China and its impact on AAPL and demand for edge AI hardware.