The thesis
Capital-light is not capital-free.
Fabless companies still avoid owning most manufacturing assets. Scarcity increasingly forces leaders to finance or guarantee capacity anyway, often without the residual asset, process learning, subsidy, or scarcity upside that ownership can provide.
In a supply-constrained, geopolitically fragmented semiconductor industry, control of manufacturing capacity is strategy. Ownership is the strongest form of control, and contractual control is increasingly capital-intensive.
The old bargain
Specialization let designers focus capital and talent on products while specialist foundries carried factory ownership. That bargain remains powerful. What changed is the assumption that access to every critical layer would remain abundant, flexible, and someone else’s financing problem.
When capacity tightens, the boundary between an asset-light company and the manufacturing system it depends on becomes financial. Reservations, firm purchases, prepayments, inventory, guarantees, strategic investments, and long-duration agreements move onto the customer’s risk map.
The old CapEx-only proof of fabless capital efficiency is incomplete because material capacity exposure can appear as commitments, prepayments, inventory, guarantees, and investments.
Caveat. Do not state that every fabless company is more capital-intensive than every IDM.
NVIDIA disclosed $119 billion of manufacturing, supply, and capacity commitments at April 26, 2026.
Caveat. Keep separate from cloud, investment, and other vendor commitments.
Open source
AMD disclosed approximately $12.166 billion of unconditional commitments at December 27, 2025.
Caveat. Commitments include wafers/substrates and non-manufacturing items.
Open source
Qualcomm disclosed $15.1 billion of purchase obligations at September 28, 2025, primarily with integrated-circuit suppliers and including multi-year capacity commitments.
Caveat. Follow filing definition.
Open source
A more complete scorecard
Reported CapEx and PP&E remain essential measures. They are simply insufficient on their own. A complete manufacturing-capital review keeps each exposure in its own accounting and economic category, preserves the reported definition and period, and asks who owns the residual asset.
Purchase commitments are not automatically prepaid cash and should not be treated as economically identical to fab construction.
Caveat. Core caveat for all charts.
Ownership is conditional
The slogan earns attention; the counterexample earns trust. Competitive, utilized manufacturing can create allocation priority, process learning, residual option value, public-policy relevance, and scarcity rent. Uncompetitive or underused manufacturing can consume capital and destroy value.
The defensible conclusion is narrower: control the bottleneck that governs the product. Own it when ownership creates durable advantage. Otherwise negotiate allocation, portability, remedies, and downside sharing strong enough to survive scarcity.
The strongest semiconductor businesses own the bottleneck, finance it, or control it through contracts that survive scarcity.
Caveat. Editorial synthesis, not a quoted fact.
Intel is a caution, while TSMC is evidence that competitive, highly utilized manufacturing ownership can produce superior economics.
Caveat. Avoid claiming that all owned capacity creates value.
Open sources
Intel’s experience shows that fab ownership without process competitiveness and utilization can destroy value.
Caveat. Editorial inference supported by losses, restructuring, and project changes.