TSM · NYSE
Taiwan Semiconductor Manufacturing Company
The irreplaceable foundry powering the world's most advanced chips — a geopolitical and technological linchpin.
A structured, long-term business thesis on TSMC — analyzed through the six pillars of the Evergreen Framework.
Executive Summary
The world runs on the nodes TSMC manufactures.
TSMC is the pure-play foundry that manufactures the leading-edge silicon behind essentially every modern computing platform — from smartphones to hyperscaler AI accelerators. Its combination of process leadership, yield, and scale is not a lead measured in months. It is measured in generations.
The interesting question today is not whether TSMC is dominant, but how durable that dominance is in a world of geopolitical friction, capex escalation, and customer concentration. The answer, so far, is that every attempt to catch up has made the moat wider, not narrower.
Investment Thesis
Four things that need to be true.
- 01
Process leadership holds
TSMC continues to ship the world's leading-edge node (N3 → N2 → A16) ahead of Intel and Samsung, with better yield and denser design ecosystem.
- 02
AI silicon demand is structural
Hyperscaler accelerators, custom silicon, and edge AI all funnel through TSMC's advanced nodes for a decade, not a cycle.
- 03
Geographic diversification succeeds
Fabs in Arizona, Japan, and Germany reduce single-region risk without meaningfully diluting cost structure over time.
- 04
Pricing power is real
Node price increases stick as customers accept that leading-edge access is more valuable than commodity wafer price.
Business Overview
One business, three technology tiers.
TSMC reports by platform (HPC, Smartphone, IoT, Automotive, DCE) but is best understood as a single manufacturing operation running at three tiers: leading-edge (where the moat is), specialty (where the volume is), and mature (where the cash is).
Leading-Edge (HPC & Smartphone)
N3, N5, upcoming N2 — the nodes that produce Apple SoCs, NVIDIA AI accelerators, AMD server chips, and hyperscaler custom silicon.
Specialty & Advanced Packaging
CoWoS, SoIC, and other advanced packaging capacity — increasingly the true bottleneck for AI accelerators, and a durable second moat.
Mature Nodes
28nm and above serve automotive, industrial, and analog — steady cash generation with far less competitive pressure at the leading edge.
Business Driver Tree
Where value actually gets created.
Six drivers, one owner-operator lens. Each is a distinct compounding engine — and each reinforces the others.
Competitive Advantages
Four moats, reinforcing each other.
Process leadership
A multi-year lead at the leading edge that competitors have consistently failed to close, compounded by superior yield.
Scale & yield
Manufacturing scale allows R&D amortization and yield learning no smaller rival can match; unit economics widen with volume.
Design ecosystem
IP libraries, PDKs, EDA integration, and customer engineering support make TSMC the default target for new tape-outs.
Advanced packaging
CoWoS and SoIC are the physical bottleneck for AI accelerators — a second, less-appreciated moat with its own capex barrier.
Capital Allocation
Where every incremental dollar goes.
TSMC's capital allocation is the most defining feature of its business model: massive, disciplined, multi-year capex in service of a widening technology lead.
Capex (Leading-Edge Fabs)
The dominant use of cash — new fabs, EUV tools, and geographic expansion in Arizona, Japan, and Germany.
R&D
Consistently ~7–8% of revenue — funds the next two nodes, packaging, and process technology.
Dividend
Progressive and durable — TSMC is a rare combination of hyper-cyclical industry and reliable dividend growth.
Buybacks
Historically minimal; management prioritizes reinvestment while returns on capital remain compelling.
M&A
Essentially none — TSMC is a builder, not an acquirer; the business is scaled internally, not bought.
Key Metrics & KPIs
What to actually watch each quarter.
Directional indicators, not price targets. Order-of-magnitude figures for reference — always cross-check against the latest 20-F / quarterly filings.
HPC revenue mix
~50%+
AI-driven; still rising
Gross margin
~55%
Structurally elevated
Operating margin
~45%
Company-wide
Capex (annual)
$35–40B
Order of magnitude
Leading-edge revenue share
N3+N5 majority
Product mix
CoWoS capacity
Doubling annually
AI bottleneck
Risks
What could break the thesis.
Geopolitical risk (Taiwan)
The single largest structural risk — cross-strait tension or supply-chain disruption would affect the entire global tech industry, not just TSMC.
Customer concentration
A handful of customers (Apple, NVIDIA, AMD, hyperscalers) account for a majority of leading-edge revenue; a pullback would ripple immediately.
Overseas fab economics
Arizona and Japan fabs have higher cost structures; if premiums for geographic diversification don't stick, margins compress.
AI demand digestion
A pause in AI capex would show up quickly in HPC utilization, given how much recent growth has been AI-driven.
Long-Term Outlook
Where TSMC plausibly is in ten years.
Assume TSMC continues to ship the world's leading-edge nodes, HPC becomes the majority of revenue, and geographic fabs mature into a global manufacturing footprint. The company looks like today's TSMC with a broader footprint, a larger advanced packaging franchise, and a durable technology lead measured in generations, not quarters.
TSMC does not need to be the only foundry to compound. It needs to remain the default leading-edge choice — the manufacturing partner every advanced silicon roadmap depends on — while the pie itself keeps growing.
Base case: durable low-double-digit revenue growth over the cycle, structurally elevated gross margins, and a progressive dividend supported by robust free cash flow after peak capex.
What Would Change Our Mind?
The signals we're actively watching.
A thesis is only useful if it can be broken. If any of the following hold for more than a couple of quarters, the framework re-underwrites.
A competing foundry (Intel Foundry, Samsung) ships a leading-edge node with comparable yield at scale.
A major customer moves substantial leading-edge volume in-house or to a rival foundry.
Geographic fabs fail to reach acceptable economics, forcing margin compression rather than pricing offsets.
A structural pause in AI capex materially reduces HPC utilization for more than one cycle.
Cross-strait tension escalates into supply-chain disruption that outlasts geographic diversification.