Goldman Sachs: AI ASIC Remains the Preferred Theme, CPU Demand Underestimated

Overview

This report summarizes Goldman Sachs’ feedback following its marketing roadshow in the U.S.; the team met with over 35 investors in the past week. Overall, investor sentiment toward AI-related opportunities remains highly positive, and AI ASIC continues to be the most favored investment theme. However, discussions are shifting from demand validation to the next phase of the cycle, including CPU demand driven by generative AI, pricing power across supply chain segments, and the sustainability of capital expenditures by hyperscale cloud providers. Goldman Sachs’ core investment recommendations include TSMC, MediaTek, Xinghua, and Wanshi—all maintaining a buy rating.

Core views

AI ASIC and CPU Become Investment Focuses: After the strong performance of AI-related stocks over the past 12 months, semiconductor exposure has become one of the largest positions in many portfolios. As a result, incremental capital allocation has become more selective, with investors focusing on companies capable of continuing to deliver robust earnings growth. The most frequently discussed topics center on the next bottleneck in the AI supply chain, pricing power within the ecosystem, which companies will best sustain excess growth in the coming years, and key downside risks from market overheating. CPU Demand Emerges as an Underrated Beneficiary: One of the most notable shifts in investor discussion is increased attention toward CPUs. Investors increasingly recognize that while AI accelerators remain the primary beneficiaries of training workloads, CPUs may play a more critical role in inference and generative AI applications. Goldman Sachs therefore emphasizes that CPU-related beneficiaries represent an increasingly attractive way to tap into the next stage of AI infrastructure growth—a theme particularly relevant for Xinghua and Wanshi. TSMC: Investors Discuss Pricing Upside Rather Than Demand: Although TSMC remains a core holding for many investors, Goldman Sachs believes expectations are still overly conservative, especially regarding profitability. Investors generally acknowledge TSMC’s strong pricing power, but a recurring discussion point is why the company hasn’t been more aggressive in raising prices. Goldman Sachs’ view is that TSMC may maintain its current pricing strategy, with major price adjustments continuing on an annual basis rather than frequent changes—but investors may still underestimate its ongoing efforts in capacity de-bottlenecking, productivity improvements, favorable product mix, and better operating leverage. MediaTek: Despite Growing ASIC Opportunities, Still Under-Allocated: MediaTek was one of the most discussed stocks during Goldman Sachs’ roadshow. While investors broadly understand the long-term opportunity in its AI ASIC business, many still underestimate the potential pricing and content expansion embedded in its next project in 2028. Investor positioning remains relatively light, with many U.S. investors still under-allocating to the stock. Xinghua: TAM Expansion and Pricing Upside Drive Investor Optimism: Xinghua continues to be one of Goldman Sachs’ favorite stocks, benefiting from high exposure to the server market and new growth drivers from generative AI. Investors are particularly focused on a potential new round of price hikes in the second half of 2026, management’s potential upward revision of BMC TAM assumptions, and strong mixed ASP growth following the ramp-up of the new BMC-AST2700 (about 50% premium over AST2600). Wanshi: CPU Opportunity and TAM Expansion Remain Undervalued: Wanshi remains a relatively less-followed stock among U.S. investors. Goldman Sachs views Wanshi as one of the most undervalued CPU plays in the semiconductor industry, expecting around 40-50% of its revenue exposure to CPUs (including x86 and ARM architectures). The company boasts a stellar customer profile, with its top three customers (NVIDIA, AMD, Google) accounting for over 85% of total revenue.

Analysis framework

Goldman Sachs’ analysis approach is primarily based on investor feedback and insights from the marketing roadshow: First, it collected market sentiment and focus areas through face-to-face meetings with over 35 U.S. investors, identifying three core themes—AI ASIC, CPU demand, and supply chain pricing power. Then, for each covered stock, it analyzed its positioning within these themes, the gap between market expectations and actual performance, and potential upside in earnings. In terms of valuation methodology, Goldman Sachs uses the P/E multiple method for all four stocks: TSMC employs a target P/E of 22x (1.0 standard deviation above its 5-year trading average) applied to 2027 expected EPS; MediaTek uses a target P/E of 25x (1.8 standard deviations above its 5-year average) applied to 2027 second-half to first-half 2028 expected EPS; Xinghua and Wanshi both use a target P/E of 40x applied to 2028 expected EPS, discounted back to 2027 at a cost of capital of 13.6%. The cost of capital assumption includes a beta of 1.5, a risk-free rate of 4.25%, and a market risk premium of 6.25%.

Methodology notes

• Valuation MethodologyPE/PEG valuation

  P/E Multiple Valuation

  Goldman Sachs uses a target P/E multiple multiplied by expected EPS to calculate target prices for all four stocks—a common relative valuation method in the semiconductor industry. Target P/E levels are set based on the standard deviation position within the stock’s historical trading range; for example, TSMC’s 22x P/E is 1.0 standard deviation above its 5-year average.

• Industry/Supply Chain Analysis FrameworkSupply-demand framework

  AI Supply Chain Supply-Demand Analysis

  The report identifies investment themes by analyzing supply-demand relationships across various stages of the AI supply chain, such as continued strong demand for AI ASICs, underestimated CPU demand in inference and generative AI, and the shift of supply chain bottlenecks from accelerators to CPUs.

• Behavioral Finance and Event GamingExpectation Gap/Expectation Management

  Analysis of the Gap Between Market Expectations and Actual Fundamentals

  Goldman Sachs repeatedly emphasizes that market expectations for certain stocks are overly conservative or underestimated—such as TSMC’s profitability expectations, MediaTek’s ASIC opportunities, and Wanshi’s CPU exposure. This expectation gap forms the core logic behind recommending buys.

• Company Fundamentals and Financial FrameworkOperating/Financial Leverage Analysis

  Operating Leverage Driving Profitability Improvement

  The report highlights that TSMC’s capacity de-bottlenecking, productivity improvements, favorable product mix, and better operating leverage will continue driving profitability growth—a key framework for analyzing the company’s earnings growth potential.

• Industry/Supply Chain Analysis FrameworkUpstream-Midstream-Downstream Value Transmission

  Value Transmission Across AI Supply Chain Stages

  The report analyzes various supply chain stages—from AI accelerators to CPUs, from chip design to test sockets—to identify which segments will benefit from the next phase of the AI investment cycle—for example, increased test complexity benefits Wanshi’s test socket business.

• Quantitative/Factor/Portfolio TheoryBeta/alpha analysis

  Beta Coefficient Used in Cost of Capital Calculation

  Goldman Sachs uses a beta coefficient of 1.5 (from Bloomberg) when calculating the cost of capital for Xinghua and Wanshi—a key parameter in the CAPM model used to measure a stock’s systematic risk relative to the market.

Asset mapping & comparison

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

• TSMC (2330.TW/TSM)

  AI Enabler, Leader in Edge Nodes and Advanced Packaging Technologies (CoWoS)

  Strengths

  Technological leadership and execution superior to peers, over 60% global foundry market share, expected to achieve 25% revenue CAGR target

  Weaknesses

  Investor expectations are overly conservative, especially regarding profitability

  Comparison

  A key AI enabler in Taiwan’s semiconductor coverage

  Risks

  Deteriorating recovery in end-demand impacting capacity utilization, slower customer node migration, slowdown in AI investment, poor yield/execution, intensifying competition, unfavorable FX trends or rising costs

• MediaTek (2454.TW)

  Key beneficiary of Google TPU, growing AI ASIC business

  Strengths

  Transforming from traditional smartphone application provider to AI-focused supplier, projected 2025-28 revenue/profit CAGR of 40%/57%

  Weaknesses

  Relatively light investor positioning, many U.S. investors still under-allocate to the stock, doubts about long-term market position

  Comparison

  Lacks clarity with Broadcom on future share distribution

  Risks

  Weaker-than-expected end-demand (especially smartphones), rising foundry costs impacting margins, intensifying competition, slowing ASIC ramp-up

• Xinghua (5274.TWO)

  Pure server/CPU play, BMC market leader

  Strengths

  Largest BMC supplier globally, about 70% market share, 85% of revenue from BMC segment, projected 2025-28 revenue/profit CAGR of 66%/72%

  Weaknesses

  Market lacks consensus on server CPU growth magnitude after 2027

  Comparison

  Continues to be one of Goldman Sachs’ favorite stocks

  Risks

  Weaker-than-expected recovery in server market demand, slower BMC penetration in AI servers than expected, intensifying market competition

• Wanshi (6515.TW)

  Most Undervalued CPU Play, Test Socket Supplier

  Strengths

  Second-largest socket supplier globally, about 8% market share, around 40-50% of revenue exposure to CPUs, top three customers account for over 85% of revenue, projected 2025-28 revenue/profit CAGR of 78%/109%

  Weaknesses

  Still relatively less-followed among U.S. investors

  Comparison

  One of the most undervalued CPU plays in the semiconductor industry

  Risks

  Slower AI/HPC demand, slower new TAM penetration, intensifying competition

Key data

• TSMC Target P/E22x1.0 standard deviation above 5-year average, applied to 2027 expected EPS
• MediaTek Target P/E25x1.8 standard deviations above 5-year average, applied to 2027 second-half to first-half 2028 expected EPS
• Xinghua Target P/E40xConsistent with its 10-year historical trading average, applied to 2028 expected EPS
• Wanshi Target P/E40x1.26 times the 3-year forward average P/E of 31.5x, applied to 2028 expected EPS
• TSMC Long-Term Gross Margin56%+Goldman Sachs expects long-term gross margin to stay above 56%
• MediaTek 2025-28 Revenue/Profit CAGR40%/57%Projected revenue and profit compound annual growth rates of 40% and 57% respectively
• Xinghua 2025-28 Revenue/Profit CAGR66%/72%Projected revenue and profit compound annual growth rates of 66% and 72%, significantly accelerating from 5.9%/4.9% in 2019-23
• Wanshi 2025-28 Revenue/Profit CAGR78%/109%Projected revenue and profit compound annual growth rates of 78% and 109%
• Wanshi CPU Revenue Exposure40-50%Expected around 40-50% of revenue exposure to CPUs (including x86 and ARM architectures)
• Wanshi Top Three Customers’ Share85%+Top three customers (NVIDIA, AMD, Google) account for over 85% of total revenue
• Xinghua BMC Market ShareAbout 70%Largest BMC supplier globally, accounting for about 70% of the global BMC market share
• TSMC Global Foundry Market Share60%+Holds over 60% of revenue share in the global foundry market
• Wanshi Global Socket Market ShareAbout 8%Second-largest socket supplier globally in 2024, accounting for about 8% of the market share
• AST2700 Price PremiumAbout 50%New BMC-AST2700 has a premium of about 50% over AST2600
• AST2800 ASP MultiplierAbout 2xASP will be about twice that of AST2700 when migrating to AST2800 in 2029 and beyond
• Wanshi TAM Annual Growth Rate50-60%Expected at least 50-60% TAM expansion annually in the coming years

Impact & implications

Implications for Investors: With the strong performance of AI-related stocks over the past 12 months, semiconductor exposure has become one of the largest positions in many portfolios, making incremental capital allocation more selective. Investors should focus on companies that can continue delivering robust earnings growth and justify further holding from current levels. Implications for the Industry: Discussions are shifting from AI demand validation to the next phase of the cycle—including CPU demand, supply chain pricing power, and the sustainability of capital expenditures by hyperscale cloud providers. Google’s recent capital-raising activity has drawn investor attention; historically, U.S. investors tended to see external fundraising as a signal of overinvestment or bubble behavior, but sentiment seems to be evolving, with investors increasingly viewing Google’s ability to raise capital as a positive indicator of confidence in long-term AI monetization opportunities.

Risks

• Further deterioration in end-demand recovery impacting capacity utilization
• Slower customer node migration
• Slower AI investment leading to lower long-term semiconductor content growth
• Poor yield/execution resulting in weaker-than-expected profitability
• Intensifying competition eroding ASP/profitability
• Unfavorable FX trends or rising costs impacting margins
• Weaker-than-expected smartphone end-demand
• Rising foundry costs impacting margin outlook
• Slower ASIC ramp-up affecting operating leverage
• Weaker-than-expected recovery in server market demand
• Slower BMC penetration in AI servers than expected
• Slower AI/HPC demand
• Slower new TAM penetration

What to watch

• TSMC’s Q2 2026 earnings report and whether profitability once again exceeds expectations
• MediaTek’s first AI ASIC project’s commercial ramp-up by the end of 2026
• Xinghua’s potential new round of price hikes in the second half of 2026
• Management’s potential upward revision of BMC TAM assumptions at Xinghua
• Xinghua’s migration to AST2800 (starting in 2029 and beyond)
• Wanshi’s SLT market penetration among U.S. AI GPU customers beyond the FT market
• Whether capital expenditure growth by hyperscale cloud providers needs to accelerate further in 2028
• CPU demand growth driven by generative AI workloads