Tech / AI value chain — who owns what, and where the cracks are

First entry in the daily-thinking log. Picking Wednesday's theme (tech/AI value chain) rather than kicking off with the broader "reform momentum" pillar — that gets its own scoping doc (docs/REFORM_MOMENTUM_PILLAR.md, next wake). Today is about reading the current state of the tech supply chain from the equity-ETF standpoint.

What I'm watching

• Taiwan (EWT): TSMC is the single most important company

in the AI stack. Its 2-3nm capacity fills and pricing power set the floor for Nvidia, AMD, and Apple silicon margins.

• Korea (EWY): SK Hynix and Samsung for HBM memory, which

has become the binding constraint in large-model training clusters. Margins on HBM3e/HBM4 are much richer than commodity DRAM.

• Japan (EWJ): semiconductor equipment (Tokyo Electron,

Disco, Advantest, Lasertec) — picks-and-shovels exposure to the whole node transition, less concentrated than a single fab risk.

• Netherlands (EWN): ASML, the single-supplier on EUV and

high-NA EUV. Their order book is the leading indicator of capacity intentions 2-3 years out.

• US (SPY): this is where most of the hyperscaler capex

sits (Microsoft, Google, Meta, Amazon). The demand side of the chain, not the supply side.

• China (MCHI): increasingly a parallel stack — SMIC,

Huawei's Ascend, domestic HBM efforts. Geopolitically awkward; sanctions regime matters more than fundamentals here.

What changed / what matters

The AI capex story dominated equity markets through 2023-2024 and into 2025. The question for 2026 is no longer "is the capex happening" (it is) but "at what rate is ROI materialising, and what's the multiplier from capex-dollars to downstream equity returns by country."

Two things I'd watch in the next 90 days:

1. Hyperscaler ROI disclosure. Microsoft, Meta, Google Q1-Q2 earnings will reveal whether the $200B+ annual capex is translating to incremental revenue. If ROI compresses (e.g., cloud AI revenue grows slower than capex), the whole chain re-rates down — including TSMC, ASML, Hynix. If ROI holds, the chain stays bid but we're already priced for perfection.

2. HBM supply elasticity. Current HBM market is ~30% supply constrained. Samsung has struggled to qualify HBM3e at Nvidia, leaving Hynix with pricing power. If Samsung qualifies meaningfully in Q2-Q3, Hynix margins compress (bad for EWY); if not, the supply bottleneck persists and HBM pricing stays elevated (good for EWY, neutral to slightly bad for US hyperscalers buying the HBM).

The country-ETF angle: EWT and EWY are levered long the AI stack. Netherlands (EWN) is almost a pure proxy for EUV demand. Japan (EWJ) is the lowest-beta way to stay exposed (equipment diversification). US (SPY) is the demand side but also 65%+ of ACWI already — the user's "€10 in 10 countries" structure naturally underweights it by design.

Candidate picks within this theme

• EWT (Taiwan) — TSMC's node leadership and pricing power is

the durable moat. Tail risk = Taiwan Strait geopolitics, which is real but unknowable.

• EWY (Korea) — HBM oligopoly value chain. More binary than

EWT because Samsung qualification risk is live.

• EWN (Netherlands) — purest-play EUV proxy. Single-stock

risk (ASML is most of the ETF's equity beta).

• EWJ (Japan) — equipment diversification, lower volatility,

cheaper than Korea/Taiwan on earnings. Underrated defensive AI proxy.

• MCHI (China) — omit for user's 10-pick context.

Sanctions regime dominates fundamentals; the risk/reward asymmetry is unfavorable for a DCA'er.

Provisional lean going into Saturday's synthesis: EWT, EWY, EWJ, EWN likely occupy 3-4 of the 10 slots, with EWT and EWN being highest-conviction. That's a lot of "semiconductor beta" — Saturday's synthesis should flag if this is too much concentration risk for a €200/month DCA.

What I'd revise if I saw

• HBM3e qualification announcement from Samsung → cut EWY

conviction significantly

• TSMC cut to 2026 capex guidance → cut EWT + EWN together
• Major Taiwan Strait escalation → cut EWT + EWY, raise EWJ

Cross-references

• Current macro data: .cache/backtest/etf-metrics.json
• ACWI sector exposure: ~30% tech. Your 10-country DCA picking

3-4 semi-exposed countries = higher tech beta than ACWI.

• Next Wednesday entry: will track hyperscaler Q1 capex prints +

HBM price tape.