It’s Not 1999: Why the Market Is Wrong About Semiconductors (Again)

Summary

Semiconductors have quietly become one of the most consistently profitable sectors in the global economy. And yet, institutional portfolios continue to underweight the space, trapped by memories of dot-com volatility and uncomfortable with the technical complexity of the supply chain. But the numbers—and the industry structure—tell a different story. The post-2015 rise in average ROE above 20% is not a fluke. It’s a function of secular demand, tightening supply constraints, and enormous technological moats that resemble a distributed, layered version of Nvidia across the Taiwanese ecosystem. The capex cycles are real, but the investment logic isn’t speculative—it’s hyperscaler-driven and anchored in enterprise adoption. Investors who still view semiconductors as “cyclical” are missing a potential decade long compounder in disguise.

Thesis

The Market’s Blind Spot: Every once in a while, markets collectively misprice something not because they haven’t noticed the data—but because they remember it too well. In the case of semiconductors, that memory goes back to 1999. Back then, semis were hot. Dot-com demand and a frothy investor mindset sent chip stocks soaring. But post-2000, the story turned: oversupply, brutal competition, margin compression, and the rise of software over hardware. From 2001 to 2015, semiconductors lost their shine. They were still crucial, yes—but increasingly commoditized, low-ROE, and deeply cyclical. That impression stuck. Now fast forward to 2025. The average return on equity in the semiconductor sector has exceeded 20% for nearly a decade. Yet the investor base hasn’t returned. According to YWR’s QARV rankings, semis screen as high quality on profitability—but continue to trade at valuations more typical of heavy industry than elite tech. The ROE Re-Rating That Already Happened The re-rating in semiconductor fundamentals didn’t happen overnight—it began quietly around 2015. That was when the sector’s return on equity started a structural climb. Instead of bouncing between 5–10% like in the prior decade, ROEs began to hover in the 15–25% range, even in non-boom years. What changed? • Scale and complexity became a moat. The cost of building a fab rose from $1B to $20B to now ~$50B. Only a handful of companies on Earth can build or operate one at scale. • Demand shifted from cyclical gadgets to systemic infrastructure. AI, cloud, edge computing, automotive semis—these are now mission-critical parts of the economy. • Capital discipline emerged. Foundries and design firms began prioritizing long-term ROIC instead of short-term volume growth. Semiconductors are no longer a commodity cycle—they’re a tollbooth industry. And yet the market hasn’t updated its framework. Misunderstood Cyclicality: It’s Not 1999, It’s 2010 Cloud All Over Again One of the biggest misconceptions is that we are reliving 1999: massive tech hype, too much spending, and an inevitable bust. But the reality is very different. What’s happening today more closely resembles the 2010–2020 cloud infrastructure buildout than the dot-com bubble. Back then, hyperscalers like Amazon, Microsoft, and Google steadily ramped capital spending to build out real, cash-generating cloud services. Capex wasn’t speculative—it was tied to genuine demand. The exact same thing is happening now, only the layer has shifted down the stack to hardware. AI datacenters aren’t being built by Reddit forums or SPACs—they’re being built by: • Fortune 500 enterprise buyers • Cloud providers with $100B balance sheets • Governments building sovereign compute capacity This isn’t speculative—it’s strategic. The Moat: Complexity and Capital Semiconductors have one of the deepest, most layered moats in existence. But because they’re invisible to most consumers, their depth is underappreciated. Here’s what makes the modern semi ecosystem so defensible: 1. Capex as a Barrier $50 billion. That’s what a cutting-edge fab costs. And that’s not including: • 3–5 years to build and ramp • PhDs in material science • Geopolitical risk of supply chains • Specialized lithography tools only a few firms can supply (ASML) This kind of scale creates a natural duopoly or monopoly in nearly every sub-segment. 2. Specialization is Stickiness The semi stack is fragmented—but not in a bad way. Each layer (design, IP, manufacturing, packaging, testing) is specialized and sticky. Examples: • TSMC dominates advanced manufacturing • Synopsys and Cadence own chip design tools • ASML monopolizes EUV lithography • MediaTek and Realtek are critical Taiwanese enablers 3. Talent and Know-how Hard tech is hard. The knowledge base in Taiwan’s semi-supply chain is generational. This isn’t a skillset you hire for on LinkedIn. Moore’s Law is Slowing. That’s Bullish. Paradoxically, the slowdown in Moore’s Law is good for semi margins. When transistor scaling was fast and cheap, the end users were constantly getting increased capacity without much cost. That fueled supply gluts. Now? It’s the opposite: • Each performance gain requires exponentially more investment • Capacity takes longer to build • Customers are locked in longer, and willing to pay more Slower Moore’s Law = More units required for more compute. Valuation Disconnect: Why Is the Market Still Skeptical? Despite the structural changes above, semiconductors continue to trade at: • Lower P/E multiples than software peers • Modest EV/EBITDA levels • Limited institutional overweighting in portfolios Why? Because they’re misunderstood: • Investors fear inventory cycles and headline volatility • Tech analysts focus on software, not supply chains • Many fund managers lack the technical comfort to model chip-level fundamentals • Geopolitical risk (Taiwan) is seen as a cloud, even though many fabs are diversifying The end result is a mispriced moat. A structural compounder priced like a cyclical laggard. The Taiwanese Mini-Nvidias Perhaps the most compelling part of the story is the under-the-radar ecosystem in Taiwan. Investors focus on Nvidia—but miss the mini-Nvidias all around it. Companies like: • Alchip: custom AI chip design • Global Unichip: design service partner to hyperscalers • Realtek: essential IP blocks for SoCs • MediaTek: edge compute champion • Novatek: display chips for everything from TVs to EVs These firms operate at margins, IP control, and technical depth reminiscent of early-stage Nvidia. But they trade at half the valuation and are often miscategorized as commodity hardware. If you believe AI is real infrastructure, then the whole Taiwanese semi stack is the picks-and-shovels trade.

Risks

1. Geopolitical Risk — Taiwan as a Single Point of Failure The semiconductor supply chain’s beating heart still lies in Taiwan, particularly TSMC and dozens of high-value component makers. While many companies are attempting to diversify manufacturing to the U.S., Japan, and Europe, Taiwan remains irreplaceable in the short term. Mitigating factor: many semi firms are building redundancy outside Taiwan, though this will take years. 2. Inventory and Capex Cycles Still Matter While we argue the cyclicality is overstated, it hasn’t disappeared entirely. Hyperscaler capex surges can reverse or slow if: • Enterprise AI adoption stalls • Regulators step in on AI energy or privacy grounds • Capital markets tighten, leading to datacenter cost scrutiny This could lead to: • Temporary overcapacity in chips • Order cancellations • Inventory adjustments impacting quarterly results Mitigating factor: Structural demand from edge computing, autos, and global digitalization provides a longer tail. 3. Valuation Compression from Macro Events Even if fundamentals remain strong, the multiple investors are willing to pay can shift with: • Rising interest rates (which hurt long-duration cash flow stocks) • Tech rotations into software or AI services • Risk-off flows away from emerging markets (especially impacting Taiwanese firms) Mitigating factor: Many semi stocks are already trading at low relative valuations vs. quality peers. 4. Technology Disruption or Standardization Ironically, success breeds imitation. If new chip architectures (e.g. RISC-V) commoditize certain elements of the stack, or if hyperscalers bring more design in-house: • Some mid-tier IP firms could see margin compression • Fabless designers could get squeezed • High-end differentiation might erode over time Mitigating factor: The complexity of physical chip production and ecosystem interdependence makes total commoditization unlikely.

📈 Price Targets

No price targets submitted with this view.

Tags

• Semiconductors
• Technology