AI Spillover Ignites Mature Node Shortage, UMC Upgraded to Overweight
AI summary card
AI Spillover Ignites Mature Node Shortage, UMC Upgraded to Overweight
Morgan Stanley believes surging AI power IC demand and capacity shifts by major players will trigger a 2027 mature node capacity shortage, favoring second-tier foundries and specialty process platforms.
- Expected mature node capacity shortage in 2H27, marking the start of an upcycle.
- Surging AI power IC demand; 800V architecture significantly increases power semiconductor value.
- Major players like TSMC focus on advanced nodes; exiting mature capacity creates spillover dividends.
- Upgrades UMC to Overweight; doubles target price to NT$138.
- Maintains SMIC Overweight, Hua Hong and Vanguard Neutral; downgrades GWC and Silergy.
Report interpretation
Overview
Morgan Stanley releases in-depth foundry industry report, noting AI infrastructure build-out spills demand from core compute chips to peripheral chips like PMICs, causing structural tightness in mature node capacity. As leaders like TSMC and Samsung shift resources to advanced nodes and gradually consolidate mature capacity, second-tier foundries and specialty platforms will see an upcycle with rising volume and price.
Core views
Demand Side: AI Power IC becomes the core growth engine. The rise in AI server power consumption drives power architectures towards 800V evolution, significantly increasing the value of power semiconductors per rack. Morgan Stanley estimates the AI server power IC market size will reach $15 billion in 2026, with a CAGR of 50% over the next two years. This incremental demand is sufficient to offset weakness in consumer electronics and smartphone chips. Supply Side: Capacity shifts by major players create spillover dividends. TSMC and Samsung are prioritizing capex for advanced nodes and AI advanced packaging, while gradually eliminating or consolidating inefficient mature node capacity (e.g., TSMC plans to phase out some 6-inch and GaN factories). This resource reallocation leads to structural tightening in mature node capacity, creating opportunities for UMC, SMIC, Hua Hong and other second-tier and specialty foundries to win orders and raise prices. Pricing and Trends: Mature node price hike cycle begins. Affected by AI peripheral chips and storage chip spillover effects, 8-inch BCD process and specialty storage (e.g., NOR Flash) capacity are at full load. UMC and Vanguard have initiated wafer price hikes, and it is expected that mature node wafer prices will continue to rise from 2H26 to 2027, keeping utilization rates high.
Analysis framework
The report unfolds along the industrial chain transmission line of 'AI Core Computing Expansion → Peripheral Facility Upgrade → Mature Node Supply/Demand Reversal'. First, disassemble AI server rack (e.g., Nvidia Rubin/Feynman) power architecture to quantify the doubling effect of 800V high-voltage trends on power semiconductor BOM costs; Second, track capex and capacity exit plans of leading foundries (TSMC) to demonstrate rigid contraction in mature node supply; Finally, combine 1Q26 earnings guidance and utilization data from various foundries to validate pricing logic for PMIC, BCD platforms, and specialty storage, using Residual Income Model (RIM) to reprice targets.
Methodology notes
Residual Income Model
The report uses the Residual Income Model to derive target prices for all covered foundry and semiconductor targets. This method evaluates enterprise intrinsic value by forecasting excess returns generated by the company over its equity capital cost (i.e., residual income) and discounting them, which is particularly suitable for capital-intensive foundry industries with high depreciation but stable cash flows.
Spillover Effect of AI Compute Demand onto Peripheral Chips and Mature Nodes
The report does not view AI GPU advanced node demand in isolation, but rather traces the connection, analyzing how surging AI server power consumption drives demand for supporting chips such as Power Management ICs (PMIC), transmitting to mature node (e.g., 8-inch BCD process) foundries responsible for producing these chips. This full-chain perspective transmission analysis helped uncover the maturity node structural shortage logic overlooked by the market.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- United Microelectronics (UMC) (2303.TW)Beneficiary Target: Mature node recovery exceeds expectations, strong AI spillover demand.
- Strengths
- Mature node foundry quality matches TSMC, possesses 12nm FinFET strategic reserve, valuation attractive.
- Weaknesses
- Recovery in non-AI markets (such as traditional consumer electronics) still faces uncertainty.
- Comparison
- Expected P/E ratio in 2027 approximately 20x, lower than Vanguard's 27x.
- Risks
- Global semiconductor demand slowdown, intensified competition in mature nodes may prevent price hikes from materializing.
- SMIC (0981.HK)Beneficiary Target: AI spillover to PMIC/Edge AI, domestic advanced node leader.
- Strengths
- 8-inch BCD process at full load, aggressive expansion of advanced nodes, scarce in domestic AI chip manufacturing.
- Risks
- Depreciation pressure from high capital expenditures may suppress long-term profit margins.
- Hua Hong Semiconductor (1347.HK)Beneficiary Target: Strong PMIC/BCD platform demand, NOR Flash spillover.
- Strengths
- Deep specialty process barriers, 8-inch capacity utilization over 100%.
- Weaknesses
- Consumer electronics demand remains weak.
- Comparison
- Valuation has become reasonable after stock price rise, thus maintained Neutral.
- Risks
- Depreciation pressure brought by Fab 9 ramp-up and impact from minority interests.
- Vanguard (5347.TWO)Beneficiary Target: Interposer production outsourcing by TSMC, 8-inch price hike.
- Strengths
- Singapore 12-inch factory capacity at full load, receiving TSMC advanced packaging spillover orders.
- Comparison
- Risk-reward ratio tends to balance, maintained Neutral.
- Risks
- Expansion of Singapore factory leads to increased capex, long-term profit margin pressure.
- GWC (6488.TWO)Neutral/Harm Target: Silicon wafer price increase expectations already fully priced in.
- Strengths
- Long-term contract (LTA) fulfillment improvement, progress made in compound semiconductor.
- Weaknesses
- Production costs rise, new capacity ramp-up limits profit margin recovery speed.
- Risks
- Global wafer supply unexpectedly increases, peer competition lowers ASP.
- Silergy (6415.TW)Harm Target: Foundry price hikes push up costs, weak consumer demand.
- Weaknesses
- Small exposure to server and optical PMIC, unable to pass foundry price hike costs to downstream.
- Risks
- Gross margin continues to face pressure, fierce price war in China driver IC market.
Key data
- AI Server Power IC TAM (2026E)$15 Billion USDProjected 2-year CAGR reaches 50%
- UMC Target PriceNT$138Significantly raised from NT$68, rating upgraded to Overweight
- SMIC Target PriceHK$85Raised from HK$70, Maintained Overweight
- UMC 2H26 Wafer Price Increase Expectation5%-10%Further price increase expected in 2027
Impact & implications
The bonanza from the AI wave is spreading from a few advanced node giants to a broader mature node ecosystem. Second-tier foundries with specialty processes (such as BCD, embedded storage) and stable capacity will gain stronger pricing power. Conversely, chip design firms with weak bargaining power, heavily reliant on consumer electronics, and facing rising foundry costs will see profit margins under pressure.
Risks
- Global macroeconomic and semiconductor end-demand recovery slower than expected
- Intensified competition in mature nodes prevents price hikes from materializing
- High depreciation brought by new capacity ramp-up suppresses profit margins
What to watch
- Actual exit progress of mature node capacity by leading foundries (such as TSMC)
- Quarterly capacity utilization rates and wafer ASP changes for each foundry
- Actual implementation rhythm of AI server power architecture evolution to 800V