The upcycle in mature process technologies has begun, with AI power ICs driving new demand.

Overview

Morgan Stanley’s report notes that, following more than two years of sustained AI infrastructure investment, mature‑node capacity is entering an upward cycle. The surging demand for power‑management ICs (PMICs) from AI servers is offsetting the weakness in consumer electronics and smartphone semiconductors, potentially triggering a shortage of mature‑node capacity by the second half of 2027. Based on this assessment, the firm has raised its price targets for UMC, SMIC, Hua Hong, and Vanguard, while upgrading UMC’s rating to “Overweight.” However, for upstream players or those under cost pressure—such as Silergy and GlobalWafers—the firm has downgraded their ratings, citing that the positive factors are already fully priced in or that they face margin compression.

Core views

AI-Driven Structural Shortage in Mature Process Nodes: The core argument of the research report is that the AI wave not only benefits advanced process technologies but also generates a strong “spillover effect,” significantly boosting demand for mature nodes. The surge in power consumption of AI servers has triggered explosive growth in demand for power ICs, with the AI server power IC market expected to reach US$15 billion by 2026, posting a compound annual growth rate (CAGR) of 50% over the next two years. Global leaders such as Infineon and MPS have increasingly adopted foundry-based production models, with MPS leveraging HuaHong and VIS, and Infineon turning to UMC. This demand momentum is sufficient to offset the weakness in traditional consumer‑oriented semiconductors, leading to tightening capacity in mature processes starting in the second half of 2026 and potentially evolving into shortages by the second half of 2027. Logic Behind Stock Rating Adjustments: 1. UMC (upgraded to Overweight): Benefiting from an unexpectedly robust recovery in non‑AI demand and spillover from AI-related orders. Wafer prices are expected to rise by 5%–10% in the second half of 2026 and continue to climb in 2027. Its 22nm/28nm and specialty process portfolio is improving, and bridge die technology is opening up new opportunities. Current valuations are attractive, with a 2027 P/E ratio of 20x versus 27x for VIS. 2. SMIC (maintained at Overweight): AI demand is spilling over from compute chips to supporting components such as PMICs and edge AI logic. Its 8-inch BCD platform is fully utilized, and pricing negotiations are helping to mitigate depreciation pressures. Despite its long‑term depreciation burden, near‑term revenue and gross margin guidance remain robust. 3. HuaHong (maintained at Neutral): AI demand is concentrated in PMICs and the BCD platform, while NOR Flash is also benefiting from inventory cycle spillover. Although the price target has been raised, we believe the stock is fairly valued after its recent rebound. 4. VIS (maintained at Neutral): Secured TSMC’s interposer outsourcing orders, alleviating CoWoS bottlenecks and boosting utilization at its Singapore facility. However, given the pressure on long‑term margins from rising capital expenditures, we maintain a neutral rating. 5. Silergy (downgraded to Underweight): First‑quarter gross margin fell short of expectations, primarily due to higher wafer costs passed through by foundries and a declining share of high‑margin automotive revenue. With over 30% of revenue tied to the sluggish consumer market, cost pass‑through remains limited, prompting the downgrade. 6. GWC (downgraded to Neutral): While the industry is recovering, expectations for wafer price hikes in the second half of 2026 appear overly optimistic, with most upside already priced in. Meanwhile, rising production costs and the ramp‑up of new capacity are constraining the pace of margin recovery.

Analysis framework

Supply–Demand Framework and Industry-Chain Transmission Analysis: The research report begins by examining macro‑level industry trends, identifying the sustained nature of AI infrastructure development—expected to continue for another two years—and inferring the resulting shifts in demand for upstream semiconductor components. Rather than focusing solely on GPUs, the firm delved into the bill‑of‑materials (BOM) structure of AI servers, finding that power‑management ICs (PMICs) represent a key incremental market. Revenue Decomposition and Capacity Utilization Tracking: At the micro level, the firm validated signs of demand recovery by analyzing guidance from foundry earnings call transcripts—such as sequential shipment growth, capacity utilization rates, and ASP changes. For example, by comparing the P/E multiples of UMC and VIS and factoring in forecasts for future wafer‑price appreciation (5%–10%), the firm assessed investment value relative to price. Residual Income Model (RIM) Valuation: To derive target prices, the firm commonly employs the Residual Income Model, which emphasizes a company’s ability to generate excess returns above its cost of capital. By adjusting mid‑term growth assumptions—such as raising UMC’s forecast from 5.9% to 12.5%—and recalibrating the cost of equity, the firm arrives at revised target valuations. This approach is particularly well suited for evaluating the intrinsic value of companies in cyclical sectors that possess enduring competitive advantages.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply-and-Demand Framework

  Forecast of Tight Capacity in Mature Process Nodes

  The research report, by analyzing the duration of AI infrastructure deployment on the demand side and the capacity-allocation strategies of major foundries on the supply side—such as TSMC’s reduction of mature-node capacity—concludes that a supply‑demand gap will emerge in the second half of 2027, thereby underpinning price increases and higher utilization rates.

• Industry/Industrial Analysis FrameworkTransmission across the upstream, midstream, and downstream segments of the industrial chain

  AI Spillover Effect

  The research report notes that AI‑driven demand not only benefits GPU manufacturers directly but also, via increased server power consumption, extends to mature‑node chips such as power‑management ICs (PMICs). This spillover effect is the primary driver behind the current upturn in the mature‑node cycle.

• Valuation methodsRIM Residual Income Model

  Target Price Estimation Based on Residual Income

  When calculating target prices for companies such as UMC and VIS, institutions have employed the residual income model. This model assesses intrinsic value by projecting future excess earnings—those exceeding the cost of equity capital—and discounting them to present value, making it particularly well-suited for valuing capital-intensive, highly cyclical semiconductor manufacturers.

• Company Fundamentals and Financial FrameworkVolume-price decomposition

  Revenue Growth Drivers Decomposition

  In analyzing the performance of contract manufacturers, research reports decompose revenue growth into two components: “volume” and “average selling price (ASP/Price).” For instance, they note that SMIC’s 8-inch revenue growth was driven primarily by price adjustments and product mix optimization, rather than new capacity additions, while UMC benefited from a rebound in shipments and anticipated price hikes.

Asset mapping & comparison

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

• UMC (2303.TW)

  Benefits: Recovery in demand for mature process nodes and overflow of AI power‑IC orders.

  Strengths

  Valuation is attractive (2027E PE at 20x), with improvements from the combination of 22/28nm processes and specialty technologies, and bridge chip technology poised to unlock new growth drivers.

  Weaknesses

  Uncertainty remains regarding the recovery of the non-AI market.

  Comparison

  Valuation is below the global benchmark (27x).

  Risks

  Global semiconductor growth has fallen short of expectations, and pricing competition has intensified.

• SMIC (0981.HK)

  Benefit: Strong demand for AI‑related silicon wafers, with domestic leaders in advanced process technology.

  Strengths

  The 8-inch BCD platform is operating at full capacity, and aggressive expansion into advanced process technologies underpins long-term growth.

  Weaknesses

  High capital expenditures have resulted in substantial depreciation pressure.

  Comparison

  It holds a leading position in both mature and advanced process manufacturing within China.

  Risks

  Geopolitical risks and a slowdown in domestic demand for AI chips

• Silergy (6415.TW)

  Adverse Impact: Rising OEM costs that cannot be fully passed on, coupled with sluggish consumer demand.

  Strengths

  It has a certain presence in the automotive and industrial sectors.

  Weaknesses

  Q1 gross margin fell short of expectations, with over 30% of revenue attributable to a sluggish consumer goods market.

  Comparison

  Servers and optical-module PMICs have limited exposure and have not adequately hedged against the downturn in consumer demand.

  Risks

  Intensified market competition has led to price erosion, and the impact of China’s domestic semiconductor support policies has fallen short of expectations.

• World Advanced (5347.TWO)

  Benefits: Securing TSMC’s interlayer outsourcing orders and an increase in capacity utilization.

  Strengths

  Singapore’s 12-inch fab has received technology transfer from TSMC, boosting its visibility.

  Weaknesses

  Increased capital expenditures in 2026 may constrain long-term profit margins.

  Comparison

  Valuation is higher than that of UMC, with a relatively balanced risk-reward profile.

  Risks

  Depreciation pressures in the early stages of new capacity ramp-up, coupled with intensifying competition in the power semiconductor sector.

Key data

• AI Server Power IC Market Size (2026E)US$15 billionThe compound annual growth rate (CAGR) is projected to be 50% over the next two years.
• UMC’s projected wafer price increase for the second half of 20265%-10%Further upside is expected in 2027.
• Semiconductor content of Rubin Ultra rack-mounted power supplies>US$20,000PCS (Power Capacitor System) content per rack
• Feynman Architecture Power Semiconductor BOMUSD 191/kWA significant improvement compared to Rubin Ultra’s $159/kW.
• UMC 2027E PE20 timesBelow the globally advanced multiple of 27 times, indicating attractive valuation.

Impact & implications

For the semiconductor foundry industry, this means that mature‑node processes are no longer “obsolete capacity” but rather an indispensable component of the AI ecosystem, capable of commanding premium pricing. From an investor’s perspective, research reports highlight the importance of focusing on foundries with deep expertise in specialized nodes such as BCD and high‑voltage technologies—namely UMC and HuaHong—which stand to benefit from the outsourcing trend in AI‑powered power ICs. By contrast, analog‑chip design firms that rely heavily on consumer electronics and wield limited pricing power—such as Silergy—may face margin pressure amid rising costs and weakening demand. Furthermore, TSMC’s decision to outsource its interposer manufacturing to World Advanced underscores how advanced packaging bottlenecks are reshaping supply‑chain dynamics, creating opportunities for second‑tier foundries to boost capacity utilization by capturing overflow orders.

Risks

• Global macroeconomic conditions and semiconductor demand growth have fallen short of expectations.
• Intensified pricing competition at mature process nodes has eroded profit margins.
• Geopolitical tensions are impacting supply chain stability.
• The pace of new technology adoption has been slower than expected, impacting the efficiency of capacity conversion.

What to watch

• The actual increase in mature-node wafer prices in the second half of 2026
• Order visibility and changes in the outsourcing ratio for AI server power management ICs (PMICs)
• TSMC’s intermediary-layer manufacturing is outsourced to world-leading facilities, ensuring advanced production ramp-up and high yield rates.
• Quarterly changes in capacity utilization and inventory levels across various contract manufacturers