Semiconductor stocks Asia vs USA — who builds what, who profits?
“Semiconductors” is not a sector — it’s a global supply chain, and the geography is unambiguous: USA = design (Nvidia, AMD, Broadcom), Netherlands = lithography (ASML), Taiwan = manufacturing (TSMC), South Korea = memory (Samsung, SK Hynix), Japan = materials and equipment. This guide shows you who dominates each layer, which stocks are worth holding, and the geopolitical risks you have to price in.
The global value chain in 5 layers
- Layer 1 — Design: US-dominated (Nvidia, AMD, Qualcomm, Broadcom, Marvell). Apple, Google and Meta also have in-house custom designs.
- Layer 2 — EDA tools: US triopoly Synopsys, Cadence, Siemens EDA. Without these tools nobody designs new chips.
- Layer 3 — Lithography: ASML (Netherlands) holds the EUV monopoly — no EUV, no chips below 7nm. Tokyo Electron, Lam Research, KLA in supporting roles.
- Layer 4 — Foundries: TSMC (Taiwan) ~60 % share at leading edge, Samsung (South Korea) ~15 %, Intel trying to enter, GlobalFoundries (USA) at mature nodes.
- Layer 5 — Memory + storage: Samsung + SK Hynix (Korea) + Micron (USA) for DRAM/HBM. NAND adds Kioxia (Japan).
Each layer has a different margin profile: design = 60–80 % gross, EUV = 50 %, foundries = 50 %, memory = 30 % (cyclical). Anyone betting on “semis” should know which layer they’re buying.
90 % of all leading-edge chips worldwide come from Taiwan (TSMC). A conflict over Taiwan would freeze the global semiconductor market for several years. The US tries to counter via the CHIPS Act ($52B) — new TSMC fabs in Arizona, Samsung in Texas, Intel in Ohio. But: time-to-production is 4–6 years.
USA vs Asia — stocks compared
| Region | Player | Layer | Market cap (approx.) | Forward P/E |
|---|---|---|---|---|
| 🇺🇸 USA | NVDA | Design (GPU) | $3.5T | ~32× |
| 🇺🇸 USA | AMD | Design (CPU+GPU) | $280B | ~28× |
| 🇺🇸 USA | AVGO | Design + custom silicon | $1.1T | ~30× |
| 🇺🇸 USA | MU | Memory (DRAM/HBM) | $140B | ~14× |
| 🇺🇸 USA | INTC | Design + foundry | $110B | ~16× |
| 🇳🇱 Netherlands | ASML | EUV lithography | €320B | ~28× |
| 🇹🇼 Taiwan | TSM | Foundry (leading edge) | $900B | ~21× |
| 🇰🇷 Korea | Samsung 005930.KS | Memory + foundry | ~$340B | ~13× |
| 🇰🇷 Korea | SK Hynix 000660.KS | Memory (DRAM/HBM) | ~$110B | ~9× |
| 🇯🇵 Japan | Tokyo Electron 8035.T | Equipment | ~$70B | ~22× |
US designers carry the highest margins, Asian fabs the cheapest valuations. SK Hynix at 9× forward P/E is a value case in any DM context, with HBM growth riding directly on Nvidia GPU demand.
Pro Asia / Pro USA — strategic arguments
- TSMC monopoly at leading edge — anything sub-5nm runs here
- Valuation: SK Hynix, Samsung, TSMC all materially cheaper than US peers
- HBM boom: Samsung + SK Hynix + Micron are the three suppliers, all with pricing power
- Lower China-revenue exposure (especially Korea + Japan)
- Margins + cash flow of designers (NVDA, AVGO) materially higher
- CHIPS Act subsidies flow primarily to US locations
- Smart money concentrated in NVDA/AVGO — institutional demand
- Lower geopolitical tail risk (no Taiwan exposure)
Sample $5,000 semiconductor allocation
| Position | Ticker | Region | Weight | Amount |
|---|---|---|---|---|
| Nvidia | NVDA | 🇺🇸 | 30 % | $1,500 |
| TSMC | TSM | 🇹🇼 | 20 % | $1,000 |
| ASML | ASML | 🇳🇱 | 15 % | $750 |
| Broadcom | AVGO | 🇺🇸 | 10 % | $500 |
| SK Hynix | 000660.KS | 🇰🇷 | 10 % | $500 |
| Micron | MU | 🇺🇸 | 5 % | $250 |
| AMD | AMD | 🇺🇸 | 5 % | $250 |
| Tokyo Electron | 8035.T | 🇯🇵 | 5 % | $250 |
This mix covers all 5 layers with 50 % US / 35 % Asia / 15 % Europe weights. For anyone avoiding single-name risk: SMH (VanEck Semiconductor ETF) holds NVDA, TSM, AVGO, AMD, ASML as top-5 with combined ~50 % weight.
FAQ
Which semiconductor stock is “safest” in 2026?
“Safe” is relative — all semis are cyclical. The lowest volatility tends to be in large-cap designers with steady cash flows (NVDA, AVGO, ASML). Memory (MU, SK Hynix) is the most volatile. For maximum stability, a sector ETF (SMH or SOXX) beats single-name selection.
How serious is the Taiwan risk really?
Hard to quantify, but existential. 90 % of leading-edge chips come from Taiwan, ~70 % of that from TSMC. A military conflict would push the global tech sector into a 3–5 year recession. Markets price the risk at roughly 10–15 % discount vs comparable US names. Is that the right number? Nobody knows — the hedge is a US overweight and a small gold position.
Does Intel belong in a semiconductor bucket?
Speculative. Intel has lost share 2024–2026 (server CPUs to AMD, niche on GPUs). The CHIPS Act funding and foundry build-out are the bull thesis. If the foundry plan executes, Intel is a multi-bagger. If not, Intel remains the cyclical loser of the next decade. Maximum 2–5 % of the bucket as a bet.
What about Chinese semiconductor stocks (SMIC, YMTC)?
Political risk too high for most western investors. SMIC is on the US sanctions list, no access to EUV lithography, and is ~3 generations behind TSMC. YMTC similar. For Chinese tech exposure broadly, KraneShares CSI China Internet ETF (KWEB) is a proxy — but that’s a bet on China tech in general, not semis.
Which semiconductor ETFs are best?
SMH (VanEck Semiconductor) — somewhat concentrated, NVDA/TSM dominant. SOXX (iShares) — same top holdings, slightly broader. SOXQ (Invesco) — cheaper expense ratio. In Europe: HAN-GINS Tech Megatrend Equal Weight (ISIN IE00BDDRF600) for more diversification. All three are up roughly +200–400 % over the past 5 years.
How do US export restrictions affect the sector?
Negative short-term, ambivalent long-term. Nvidia loses ~$5B/year of China revenue from export controls. But: the cuts drive reshoring (CHIPS Act), increase pricing power for non-China chips, and accelerate hyperscaler custom-silicon programs. Net: neutral to slightly positive for US semis, mixed for memory (Samsung, SK Hynix).
Analyze semis, check correlation, compare ETFs
Which semiconductor stocks overlap, which sector ETF covers the full chain? The AI stock analysis delivers quarterly updates on NVDA, TSM, ASML.
- Nvidia, TSMC, ASML — detail pages with smart-money holdings
- Correlation matrix — how tightly correlated are designers and fabs?
- AI portfolio guide — semis in the context of the 4 layers
- What-if calculator — what would $1,000 in TSMC since 2020 have produced?