Valuation premiums in the AI hardware supply chain remain supported, but consumer electronics are dragged down by rising memory prices
AI summary card
Valuation premiums in the AI hardware supply chain remain supported, but consumer electronics are dragged down by rising memory prices
Morgan Stanley believes that upgrades in NVIDIA Rubin/GB300, AI ASICs, CoWoS/CoWoP-related packaging, PCB/ABF, MLCC, liquid cooling, and power supplies will drive capacity expansion across the Greater China hardware supply chain, while PCs and smartphones will face rising memory costs, demand destruction, and margin pressure in 2026.
- AI server momentum remains strong, with the 2026 forecast for GB200/300 racks at 70-80K, significantly above about 29K in 2025.
- The VR200 platform brings BOM upgrades: estimated cost per rack is about US$7.8M, ODM value-add rises about 38%, PCB content rises about 233%, ABF substrate content about 82%, MLCC about 182%, power supply about 32%, and thermal components about 12%.
- ODMs have underperformed the TAIEX by more than 40 percentage points year to date, but revenue is still growing year over year; the report believes suppliers with delivery records and integration capabilities stand out more.
- ABF substrates are expected to be in short supply starting in 2027, with AI-related applications expected to account for more than 75% by 2030.
- On consumer electronics, the report remains cautious: 3Q26 notebook builds are 28.9M units, down 15% year over year; global smartphone shipments in 2026 are expected to decline 13% year over year, with Android expected to decline 15%.
Report interpretation
Overview
This report is Morgan Stanley's investment summary on the Greater China technology hardware sector, covering servers, AI infrastructure, AI GPU/ASIC platforms, electronic components, ODMs, PCB/ABF substrates, MLCCs, liquid cooling, power supplies, PCs, smartphones, and supply chain migration. The core view is that AI compute hardware remains in a cycle of design upgrades and capacity expansion, and valuation premiums are supported by fundamentals; however, rising costs and tight supply of memory, copper, nickel, and other materials will be headwinds to margins for consumer electronics and some hardware segments.
Core views
The report is most positive on structural opportunities in the AI hardware supply chain, including NVIDIA Vera Rubin/GB300/VR200 platforms, the Kyber architecture, LPX/Vera CPU racks, volume growth in AI ASIC servers such as TPU and Trainium, 800 VDC power supplies, PCB/ABF substrates, data network interconnects, CPO, and liquid cooling. Near-term upside in share prices mainly comes from smooth GB300 deliveries, improved risk-reward for AI server ODMs, and supply-chain capacity expansion. Relatively negative is that consumer electronics demand and margins in 2H26 will be hit by rising memory prices, with the Android smartphone chain under greater pressure than the Apple chain.
Analysis framework
The report forms its views using bottom-up BOM breakdowns, rack shipment forecasts, supplier share assessments, cloud capex trends, CoWoS capacity demand, ABF/MLCC supply-demand analysis, the AlphaWise smartphone survey, PC and smartphone build forecasts, valuation comparisons, and supply-chain import/export data.
Methodology notes
Break down changes in value across ODM, PCB, ABF, MLCC, thermal, power, and other components by GB200/GB300/VR200 rack.
This method is used to identify the incremental component value created by single-rack upgrades and determine the beneficiary segments.
Track demand pull from AI applications for ABF substrates, T glass, and CoWoS capacity.
Based on this, the report judges that ABF will enter a state of shortage after 2027, and evaluates the feasibility and risks of CoWoP replacing CoWoS.
Estimate smartphone demand and cost pressure through consumers' upgrade intentions, brand switching rates, and storage mix.
The survey shows that the global smartphone upgrade rate rose 5 percentage points year over year to 43%, but was mainly driven by iPhone strength.
Compare Greater China hardware companies by price, market cap, EPS, P/E, P/B, P/S, EV/EBITDA, ROE, ROA, and trading volume.
Used to judge whether valuation premiums in the AI hardware supply chain are supported by growth and margin improvement.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- WistronBeneficiary related to AI server ODM and the MI400 series
- Strengths
- Share gains after VR200, and participation in MI400-series Helios rack OAM/UBB production.
- Weaknesses
- Still affected by key component supply, mass-production timing, and ODM valuation volatility.
- Comparison
- The report believes it benefits from both VR200 onward and the MI400-series.
- Risks
- VR200 delays, supply shortages, customer order changes.
- QuantaBeneficiary of GB rack ODM
- Strengths
- The report believes Quanta should gain share in CY26 GB racks.
- Weaknesses
- Affected by rack ramp-up, component supply, and customer capex timing.
- Comparison
- Compared with other ODMs, it is specifically highlighted for its GB rack supply share.
- Risks
- Component delays, cloud capex below expectations.
- WiwynnKey ODM partner for Meta/AMD MI400-series Helios rack
- Strengths
- Should benefit from multi-year, multi-generation cooperation with Meta/AMD, deploying up to 6 gigawatts of AMD Instinct GPUs.
- Weaknesses
- Higher customer concentration and project execution risk.
- Comparison
- Has a prominent role in the MI400-series Helios rack.
- Risks
- AMD platform volume below expectations, changes in Meta deployment timing.
- ABF substrate suppliersBeneficiary segment from AI GPU/ASIC packaging upgrades
- Strengths
- Rising share of AI-related applications, with the report expecting shortages after 2027.
- Weaknesses
- If packaging roadmaps change or supply expands too quickly, the supply-demand balance may change.
- Comparison
- AI demand makes ABF more growth-oriented relative to traditional PC/communications applications.
- Risks
- CoWoP substitution expectations, yield risks, T glass supply tightness.
- PCB suppliersBeneficiary segment from VR200 and AI interconnect upgrades
- Strengths
- VR200 PCB content is expected to increase by about 233%, and AI transceiver PCB TAM is expected to grow strongly in 2025-28E.
- Weaknesses
- Higher requirements for high-density processes, leading to supplier differentiation.
- Comparison
- In discussions of CoWoP feasibility, PCB line width/spacing capability remains a key constraint.
- Risks
- Insufficient yield, inability to meet the <10/10µm process requirement.
- MLCC suppliersBeneficiaries in passive components for AI servers and cloud/edge AI
- Strengths
- AI servers use more high-capacitance MLCCs, and cloud AI is expected to bring about US$1B TAM.
- Weaknesses
- Overall unit share remains small, and is sensitive to pricing and product mix.
- Comparison
- AI server MLCC accounts for about 3% of global units and 5% of value, with a higher value share.
- Risks
- Server demand volatility, inventory and pricing cycles.
- Apple supply chainRelatively preferred direction in consumer electronics
- Strengths
- iPhone upgrade momentum and brand switching rates are better than Android, and supply-chain positioning is relatively more stable.
- Weaknesses
- Still has to bear rising memory costs, and Apple may share costs with non-memory suppliers.
- Comparison
- The report explicitly maintains that Apple/the Apple supply chain is relatively better than Android OEMs/supply chains.
- Risks
- Changes in iPhone launch timing, cost-sharing pressure, end-demand below expectations.
- Android smartphone supply chainDirection with higher consumer electronics risk exposure
- Strengths
- If cost pressure eases or demand recovers, flexibility may improve.
- Weaknesses
- Lower gross margins, higher memory cost as a share of COGS, and greater customer price sensitivity.
- Comparison
- Compared with the Apple chain, it is more vulnerable to price increases, spec downgrades, and volume declines.
- Risks
- 2026 Android shipments down 15% year over year, destruction in mid-range demand, margin compression.
Key data
- Report date2026-07-15The page shows July 15, 2026 03:32 AM GMT.
- 2026e GB200/300 rack forecast70-80K racksAbout 29K racks in 2025, with significant growth forecast for 2026.
- VR200 mass production paceDelayed by 1-2 months versus the original plan, with ramp-up expected to begin in 4Q26Reasons for the delay include issues with components such as the Midplane PCB.
- Single VR200 NVL72 rack costabout US$7.8MEstimate for hyperscalers.
- VR200 ODM value-add changeabout +38%Bottom-up analysis.
- VR200 PCB content changeabout +233%Bottom-up analysis.
- VR200 ABF substrate content changeabout +82%Bottom-up analysis.
- VR200 MLCC content changeabout +182%Bottom-up analysis.
- VR200 power supply content changeabout +32%Bottom-up analysis.
- VR200 thermal component content changeabout +12%Bottom-up analysis.
- AI compute wafer consumptionCY27 could rise to US$46BFrom the report section title.
- 3Q26 notebook builds28.9M units，-5% q/q，-15% y/yThe report keeps this forecast basically unchanged.
- Global smartphone upgrade rate43%，同比+5pptThe AlphaWise Smartphone Survey shows this is the highest level in more than 10 years of surveys.
- 2026 global smartphone shipment forecast-13% y/yThe report believes price increases and demand destruction will weigh on shipments.
- 2026 Android smartphone shipment forecast-15% y/yMid-range Android demand is more elastic and more affected.
- iPhone memory costaverage per unit up 150%+ year over year in 2026Based on supply-chain checks and storage-mix estimates from the survey.
- China's share of U.S. technology imports2018 88% / about US$88bn, 2024 74% / about US$80bnShows that supply-chain redirection has already happened.
Impact & implications
For investment, the AI server hardware chain remains the clearest direction of structural growth, with beneficiary assets concentrated in companies that have capabilities in rack integration, key component supply, liquid cooling, power supplies, PCB/ABF, MLCC, and interconnects. By contrast, earnings improvement in the PC and smartphone chains may be more short-lived, with cost and demand pressures becoming more evident from 2H26. Geographic migration of the supply chain will also affect order allocation, with Vietnam, Thailand, India, and Mexico increasing in importance in U.S. technology product imports.
Risks
- Rising memory prices increase BOM pressure for PCs and smartphones and compress hardware margins.
- Price increases and tight supply of raw materials such as copper and nickel are margin headwinds.
- Component shortages may delay the shipment pace of AI server racks; for example, VR200 was pushed back by 1-2 months due to issues such as the Midplane PCB.
- Geopolitical tensions may affect spending on AI data center infrastructure.
- If cloud capex is below expectations, demand for AI servers, ASICs, network interconnects, and related components will be affected.
- If packaging roadmaps such as CoWoP advance faster than expected, they may reshape the ABF/PCB/packaging supply chain.
- Price increases and spec downgrades in Android smartphones may cause more evident demand destruction.
What to watch
- Whether GB300 server rack deliveries proceed smoothly, and whether the 2026 forecast of 70-80K racks can be achieved.
- The actual ramp-up timing of VR200 in 4Q26 and progress in resolving bottlenecks such as the Midplane PCB.
- Whether TSMC CoWoS capacity expands to 200Kwpm by 2027 in line with strong AI demand.
- Actual volume ramp of more ASIC projects, including TPU, Trainium, and self-developed plans by various CSPs.
- Whether supply of ABF substrates, T glass, PCB, MLCC, liquid cooling, and power supplies remains tight.
- Whether ODM share prices recover after underperforming the TAIEX due to revenue growth and order visibility.
- PC and smartphone demand, price increases, spec downgrades, and gross-margin changes in 2H26.
- Changes in the sources of U.S. technology imports, especially rising supply-chain share of Vietnam, Thailand, India, and Mexico.