Semiconductor Cycle 2026: TSMC, ASML, AMD, Micron Investing Thesis
Semiconductor cycle 2026 thesis: TSMC foundry dominance, ASML EUV monopoly, AMD MI300 progress, Micron HBM up cycle. Multi-year capex thesis and risks.
Semiconductor Cycle 2026: TSMC, ASML, AMD, Micron Investing Thesis
TL;DR
The semiconductor industry entered a new up cycle in late 2024, driven by AI-related capex and lagging consumer recovery. TSMC dominates leading-edge foundry with ~60% market share and is building 2nm in Taiwan plus 3nm in Arizona. ASML holds a monopoly on EUV lithography with a $300M+ price tag per machine and 18-month lead times. AMD is gaining data center GPU share via MI300X (now ~$8–10B run-rate) but remains a distant Nvidia challenger. Micron is benefiting from the HBM (high-bandwidth memory) up cycle as DRAM pricing recovers; Samsung and SK Hynix lead HBM3E supply. Many investors consider 2025–2026 the up phase of a typical 18–24 month semiconductor cycle, with potential correction in 2027. The AI thesis carries elevated valuation risk and customer concentration is meaningful across the sector.
Why the Semiconductor Cycle Matters in 2026
Semiconductors are the most cyclical large-cap sector in equities. The industry historically experiences 4-year boom-bust cycles driven by capex lead times — when demand spikes, foundries cannot expand capacity for 18–24 months, driving shortages and pricing power. When demand softens, oversupplied capacity drives margin compression for 12–18 months.
The current cycle is unusual because AI capex has decoupled from consumer cycles. While smartphone, PC, and traditional server demand remained weak through 2024, AI data center demand drove TSMC's leading-edge utilization to over 100% (overtime production). By 2025, consumer recovery has begun to layer on top of sustained AI demand, creating the strongest combined demand picture since 2021.
Key 2025–2026 dynamics:
- HBM supply remains tight through 2026 across SK Hynix, Samsung, Micron
- TSMC 3nm is fully booked through 2026; 2nm production starts late 2025
- ASML EUV order backlog at €40B+, lead times 18+ months
- AMD MI300/MI325 ramp taking measurable data center GPU share
- China indigenous capacity expanding via SMIC 7nm via DUV multi-patterning
This article analyzes the four highest-conviction names in the cycle for an EU investor: TSMC, ASML, AMD, and Micron. We also frame the broader semiconductor cycle thesis with concrete risks.
Investment Thesis
The semiconductor up cycle thesis rests on four pillars:
Pillar 1: AI capex underwrites multi-year demand. Hyperscaler 2026 capex of approximately $330 billion drives demand for advanced packaging (CoWoS at TSMC), HBM (SK Hynix, Samsung, Micron), and EUV lithography (ASML). This is structural, not cyclical, until the AI capex pause arrives.
Pillar 2: Capacity additions are slow. Building a leading-edge fab takes 3–5 years and $20B+ capex. ASML cannot ship more than ~50 EUV systems per year due to manufacturing constraints. The supply side is structurally lagged versus demand, supporting pricing power for 2025–2026.
Pillar 3: Memory cycle has turned. DRAM and NAND pricing bottomed in 2023 and have recovered through 2024–2025. HBM specifically commands 5x premium over commodity DRAM and is supply-constrained. Memory makers are increasing wafer allocation to HBM at the expense of commodity DRAM, supporting pricing on both.
Pillar 4: Geopolitical reshoring drives capex. US CHIPS Act ($52B), EU Chips Act (€43B), and Japanese subsidies (~$25B) collectively fund roughly $120B in new fab construction outside Taiwan. This benefits ASML (equipment for new fabs), AMAT, LRCX, and KLA in the medium term.
The bear case: cyclicality always wins eventually. Every prior semiconductor up cycle (1995, 2000, 2007, 2017, 2021) ended in 18–24 months with 30–50% earnings declines for the most cyclical names. Whether 2025–2026 is structurally different (AI changes the cycle dynamics) or simply the next iteration of the historical pattern is the central debate. Many investors consider position sizing critical regardless of view.
Top Picks Breakdown
TSMC (Taiwan Semiconductor Manufacturing — TSM ADR)
| Metric | Value |
|---|---|
| Market cap | ~$1.0 trillion |
| Forward P/E | ~22x |
| Revenue (TTM) | ~$95 billion |
| Revenue growth | ~30% YoY |
| Gross margin | ~55% |
| Operating margin | ~45% |
| Capex (2025) | ~$42 billion |
| Leading-edge market share | ~60% |
| AI revenue share | ~25–30% |
TSMC is the indispensable foundry. Every Nvidia GPU, AMD GPU, Apple SoC, and most Qualcomm/Broadcom chips are fabricated at TSMC. Leading-edge nodes (5nm, 4nm, 3nm) command 50%+ of revenue. The 2nm node ramps in late 2025 and is already over-allocated.
Bull case: Leading-edge monopoly, AI demand drives 30%+ revenue growth, Arizona fabs reduce geopolitical concentration over time, US CHIPS Act subsidies offset capex burden.
Bear case: Taiwan geopolitical risk is binary and uninsurable, capex intensity (~45% of revenue) limits FCF, customer concentration on Apple (~25%) and Nvidia (~12%).
ASML
| Metric | Value |
|---|---|
| Market cap | ~$320 billion |
| Forward P/E | ~38x |
| Revenue (TTM) | ~€30 billion |
| Revenue growth | ~25% YoY |
| Gross margin | ~52% |
| Operating margin | ~32% |
| Order backlog | ~€40 billion |
| EUV market share | ~100% (monopoly) |
| High-NA EUV first delivery | 2024 |
ASML manufactures the lithography systems required to print sub-7nm chips. Extreme ultraviolet (EUV) lithography is an effective monopoly — no other company has shipped a production EUV system. High-NA EUV (next gen, $380M+ per machine) began shipping in 2024 to Intel and TSMC.
Bull case: Monopoly on critical equipment for advanced nodes, 18-month lead times protect pricing, fab buildout cycle in US/EU/Japan extends visibility through 2027, recurring service revenue grows with installed base.
Bear case: China export restrictions limit DUV revenue (~10% of revenue at risk), customer capex pauses translate quickly to order softness, valuation premium to broader semicap (AMAT, LRCX, KLA all trade at 22–28x).
AMD
| Metric | Value |
|---|---|
| Market cap | ~$220 billion |
| Forward P/E | ~32x |
| Revenue (TTM) | ~$28 billion |
| Revenue growth | ~22% YoY |
| Gross margin | ~52% |
| Data center GPU revenue (run-rate) | ~$8–10B |
| Data center CPU share (vs Intel) | ~35% (gaining) |
| AI revenue share | ~30% |
AMD is the credible Nvidia challenger in data center GPUs and the dominant share gainer against Intel in CPUs. The MI300X launched late 2023 has gained traction at Microsoft, Meta, and Oracle for inference workloads. MI325X (2025) and MI350 (2026) extend the roadmap.
Bull case: Nvidia priced like a monopoly while AMD priced like a distant #2; even modest share gains (15–20% data center GPU share) drive massive earnings upside, EPYC server CPU continues taking Intel share, total addressable market doubling on AI.
Bear case: CUDA moat means AMD competes on price not feature, gross margins structurally below NVDA (~50% vs 75%), execution risk on MI350/MI400 roadmap, customer concentration similar to NVDA.
Micron Technology (MU)
| Metric | Value |
|---|---|
| Market cap | ~$130 billion |
| Forward P/E | ~14x (cyclical, low at peak earnings) |
| Revenue (TTM) | ~$30 billion |
| Revenue growth | ~60% YoY (cycle recovery) |
| Gross margin | ~38% (improving from cycle low) |
| HBM revenue share | ~20% (growing rapidly) |
| Customer mix (HBM) | Nvidia, AMD, AWS, Google |
Micron is the only US-based major DRAM maker (after Samsung and SK Hynix). The HBM3E ramp at Nvidia is the primary 2025 growth driver. Cycle dynamics: DRAM and NAND pricing has recovered from 2023 lows, and HBM supply remains tight through 2026.
Bull case: HBM cycle is structurally tighter than commodity DRAM, US-based supply increasingly important for geopolitics, low forward P/E reflects skepticism about cycle peak earnings.
Bear case: Memory is the most cyclical semiconductor segment, commodity DRAM oversupply could emerge in 2027, Samsung HBM3E qualification at Nvidia (delayed but possible) would compress Micron market share.
Valuation Analysis
| Stock | Forward P/E | Cycle position | EPS growth (3y) | PEG | Margin trajectory |
|---|---|---|---|---|---|
| TSMC | 22x | Mid up cycle | 22% | 1.0 | Stable high 50s |
| ASML | 38x | Mid up cycle | 25% | 1.5 | Expanding |
| AMD | 32x | Early up cycle | 35% | 0.9 | Expanding |
| Micron | 14x | Mid up cycle | 30%* | 0.5 | Cycle peak risk |
| Nvidia | 47x | Mid up cycle | 35% | 1.3 | At peak |
*Memory EPS growth highly cycle-dependent
Cyclical valuation paradox: Memory companies (Micron, SK Hynix) trade at the lowest P/E ratios at the peak of the cycle and the highest at the trough — because earnings are most volatile. A 14x forward P/E for Micron sounds cheap but reflects the market's anticipation that fiscal 2026 may be near peak earnings.
Compared to historical norms: ASML at 38x is at the upper end of its 5-year average (28–35x range). TSMC at 22x is roughly in line with historical average. AMD at 32x is below its 2024 peak of 45x. Micron at 14x is mid-range.
Sector vs S&P 500: The Philadelphia Semiconductor Index (SOX) trades at approximately 28x forward earnings vs the S&P 500's 22x — a 27% premium. Historical premium has averaged 10–20%, so the sector is moderately expensive relative to the market.
EU Investor Access
All four stocks plus broad semiconductor exposure are accessible via UCITS structures:
Direct ADR/share access:
- TSM (TSMC ADR): Listed on NYSE, available on every major EU broker
- ASML: Listed on Amsterdam (ASML.AS) — native EUR, no FX cost
- AMD: Listed on Nasdaq, broad EU broker availability
- MU (Micron): Listed on Nasdaq, broad EU broker availability
Recommended UCITS semiconductor ETFs:
- iShares MSCI Semiconductors UCITS ETF (SEMI / IE00BFNM3D14): 0.35% TER, ~$2.5B AUM, top holdings NVDA (15%), TSMC (12%), AVGO (11%), ASML (8%), AMD (5%)
- VanEck Semiconductor UCITS ETF (SMH / IE00BMC38736): 0.35% TER, ~$2.0B AUM, similar holdings, slightly higher concentration
- HANetf Sprott Uranium Miners UCITS — adjacent (semiconductor + power thematic)
Tax efficiency: All UCITS ETFs are Ireland-domiciled accumulating, internalize US dividend withholding at 15% via treaty. Direct US stock holdings (AMD, MU, TSM) require W-8BEN form for 15% withholding rate. ASML on Amsterdam pays Dutch dividends — Polish investors face 15% Dutch withholding plus 4% top-up to reach 19% Belka rate.
For Polish investors specifically: IKE/IKZE accounts shelter capital gains from 19% Belka tax. Holding semiconductor positions in IKE for >5 years is meaningfully tax-advantaged given the cycle volatility.
Real-World Example Portfolio
A €100,000 portfolio with semiconductor cycle tilt for a moderate-aggressive EU investor:
| Position | Allocation | Amount | Rationale |
|---|---|---|---|
| iShares Core MSCI World (IWDA) | 35% | €35,000 | Global core |
| iShares MSCI Semiconductors (SEMI) | 15% | €15,000 | Diversified semi exposure |
| TSMC (TSM) | 8% | €8,000 | Foundry monopoly, AI exposure |
| ASML | 7% | €7,000 | EUV monopoly, equipment cycle |
| AMD | 5% | €5,000 | Distant #2 GPU upside |
| Micron (MU) | 3% | €3,000 | HBM cycle, smaller speculative |
| Xtrackers AI & Big Data (XAIX) | 10% | €10,000 | AI thematic overlap |
| Vanguard FTSE All-World (VWCE) | 12% | €12,000 | Diversification |
| Cash / short bonds | 5% | €5,000 | Dry powder |
Combined direct semi (23%) + via SEMI ETF (15%) equals roughly 38% effective semiconductor exposure — very high conviction in the cycle. For a more conservative posture, scale direct positions down by half and rely more on the diversified SEMI ETF.
Note that XAIX (AI thematic) overlaps significantly with SEMI (NVDA, AVGO, AMD all appear in both). True diversification requires careful counting — many investors consider this when sizing.
Risk Factors
Cycle reversal. Every prior semiconductor up cycle ended in 18–24 months with significant earnings declines. The 2025–2026 cycle could roll over in 2027 if hyperscaler capex pauses, consumer demand stalls, or oversupply emerges. Memory is most exposed; ASML and TSMC are more resilient.
China geopolitical risk. Taiwan-focused risk affects TSMC most directly but cascades to NVDA, AMD, AAPL, and ASML supply chains. A Taiwan event would freeze global semiconductor production for 12–24 months. This is a binary tail risk that diversification cannot eliminate.
China indigenous capacity. SMIC has produced 7nm chips via DUV multi-patterning despite EUV restrictions. CXMT (Chinese DRAM) and YMTC (Chinese NAND) are scaling rapidly. Long-term, this may erode pricing in commodity segments.
ASML China exposure. Approximately 30% of ASML's 2024 revenue came from China DUV systems before further restrictions. Ongoing tightening could reduce China revenue share by 10–15 percentage points over 2025–2027.
Capex absorption. Hyperscaler capex of $330B in 2026 assumes AI demand continues to scale. If enterprise AI adoption disappoints or model improvements plateau, capex could pause sharply. NVDA, AMD, Micron most exposed; ASML less so given multi-year backlog.
Inventory cycles. Customer double-ordering during shortages (HBM, leading-edge wafers) can lead to inventory corrections in the following year. Watch lead times — when they shorten from 30+ weeks to under 16 weeks, the cycle is rolling over.
Multiple compression. Even with earnings intact, sector multiples can compress if rates rise or risk appetite falls. SOX index has historically de-rated 30–40% in bear cycles independently of earnings.
The AI thesis carries elevated valuation risk across the entire semiconductor complex.
Time Horizon Considerations
Short-term (0–12 months): Most volatile period. Quarterly earnings, hyperscaler capex commentary, and macro rates dominate. Memory most volatile (Micron, SK Hynix). Not a sector for short-term tactical positioning unless professionally trading.
Medium-term (1–3 years): This is the cycle window. Leading-edge foundry (TSMC), equipment (ASML), AI accelerators (AMD, NVDA) all ride the AI capex thesis through fiscal 2026 and into 2027. Most cycle returns are typically realized in this window. Position-sizing decisions made in 2025 matter most for this horizon.
Long-term (3–10 years): Beyond 2028, semiconductor returns depend on whether Moore's Law continues delivering economic improvements (3D stacking, advanced packaging, new architectures) or whether we hit a productivity wall. Historical returns: SOX has delivered ~13% CAGR over 20 years versus S&P 500 ~9%. The cyclical drawdowns (2008: -50%, 2018: -25%, 2022: -35%) make long-term holding psychologically difficult.
The most defensible semiconductor holding period is 3–5 years with willingness to trim exposure when cycle indicators (lead times, inventory, capex commentary) flag a turn.
FAQ
Q: Is now a good time to buy semiconductor stocks? A: Many investors consider 2025–2026 the mid up cycle phase, with strong earnings momentum but elevated valuations. Position sizing matters more than directional view. The AI thesis carries elevated valuation risk if hyperscaler capex pauses unexpectedly.
Q: TSMC vs ASML — which is the better semiconductor cycle play? A: TSMC offers more direct AI exposure (~25–30% revenue) and lower multiple (22x vs 38x), but Taiwan geopolitical risk is uninsurable. ASML offers monopoly economics and diversified customer base across foundries, but trades at higher multiple. Many investors hold both for complementary exposure.
Q: Should I avoid Micron because memory is too cyclical? A: Memory is more volatile but offers the lowest entry P/E (14x) of the group. The HBM cycle is structurally tighter than commodity DRAM. Position-size accordingly — memory is a satellite, not a core.
Q: How does AMD compare to Nvidia for AI exposure? A: AMD trades at 32x forward earnings versus NVDA at 47x with similar growth rates, suggesting more upside if AMD takes meaningful data center GPU share. But CUDA software moat keeps NVDA dominant for training. Pair trade (long both) is common.
Q: How do I track semiconductor positions across cycles and brokers? A: For consolidated multi-broker tracking with cost basis in EUR or PLN, Freenance supports semiconductor sector allocation reports and unrealized P&L tracking — useful for cyclical positions held across multiple years and through cycle turns.
Related Articles
Want full control over your finances?
Try Freenance for free