The Customers Who Stopped Being Chip CompaniesNarrow moat

TSMC (TSM) — moat facet

TSMC's customers increasingly aren't chip companies at all, which uncaps their orders and re-couples TSMC's revenue to hyperscaler capital budgets.

For most of TSMC's history its customers were fabless semiconductor companies: firms whose entire business was designing chips and selling them. That is no longer the whole picture, and TSMC says so — its filing observes that over the years its customer profile and the nature of its customers' business have changed dramatically, driven by the structural shift to high-performance computing, and that only a limited number of customers are successfully exploiting the new model1.

Capital spending plans for 2026 (US$B)Alphabet$195-205BMicrosoftabout $190BMeta$125-145BTSMC$60-64BCompany guidance: Alphabet 2Q26, Microsoft 3Q FY2026 and Meta 1Q26 calls; TSMC 2Q26 call
Alphabet alone plans to spend about three times TSMC's capital budget, and TSMC's orders now depend on budgets like these.

What changed is that the companies buying leading-edge wafers increasingly are not chip companies at all. Cloud providers designing their own accelerators come to TSMC directly, as do the AI labs commissioning custom silicon through design partners. Their reason for existing is not to sell chips but to reduce what they pay for someone else's.

This is better for TSMC than it first appears. A fabless customer's orders are capped by its own ability to sell chips; a hyperscaler's are capped by its capital budget, which is currently enormous: Alphabet alone has guided 2026 capital spending to $195-205 billion2, about three times TSMC's own $60-64 billion3. And because these buyers are competing with each other, TSMC ends up supplying every side of the same contest — the classic position of the arms dealer.

The catch is that a capital budget can be cut in a way a product roadmap cannot. Watch the share of revenue from high-performance computing against the concentration figures. The more of TSMC's business that traces to hyperscaler capex rather than to end-product demand, the more its revenue behaves like a capital-goods company's — which is not how it is currently valued.

Moat trajectory: Widening

The arrival of hyperscalers and AI labs as direct customers is a genuine improvement to the customer base. Their orders are limited by capital budgets rather than by their own chip sales, they are competing with one another and TSMC supplies all sides, and they have deeper pockets than most fabless companies ever had. Widening, with the understanding that a capital budget can be cut faster than a product roadmap can be cancelled.

The number that tests this moat
Reported
HPC share of revenue, full year
58% in 2025, from 43% in 2023

TSMC says its customer profile changed dramatically as system companies began designing chips. HPC's rising share shows that shift in the numbers, and ties TSMC's revenue more closely to hyperscaler capital budgets.

Source: TSMC Form 20-F, FY2025 ↗
References
  1. ReportedTSMC says its customer profile and the nature of its customers' business have changed dramatically with the shift to HPC.
    TSMC Form 20-F, FY2025 (net revenue, noncurrent assets by country, customer concentration, capacity above 17 million wafers, risk factors; earthquake losses of about NT$3B in 2Q24 and NT$5.3B in 1Q25) — FY2025 (and prior years) · publ. Filed early 2026 · source ↗
  2. ReportedAlphabet raised its 2026 capital spending guidance to $195-205 billion.
    CNBC live coverage of Alphabet's Q2 2026 results and call (July 22, 2026) — Cloud revenue up 82% to $24.8 billion with a cloud backlog of $514 billion; YouTube ads up 13%; Q2 capex $44.9 billion, up 100%; 2026 capex guidance raised to $195-205 billion from $180-190 billion — Q2 2026 (quarter ended June 30, 2026) · publ. July 22, 2026 · source ↗
  3. ReportedTSMC's 2026 capital budget is $60-64 billion.
    TSMC 2Q26 earnings call transcript (full-year 2026 revenue growth slightly above 40% in US dollars; 3Q26 revenue guided to US$44.6-45.8B at a 65-67% gross margin; 2026 capital budget raised to US$60-64B; 2nm ramp to dilute gross margin by about 3-4 points in 2H26; overseas-fab dilution 2-3% early, 3-4% later; an additional US$100B for Arizona; 13 leading-edge and advanced-packaging fabs under construction in Taiwan) — 2Q26 call · publ. July 16, 2026 · source ↗
Sources
Generated September 23, 2026