A Chip Foundry Is a Capital Machine That Must Never Sit Idle
Building a leading edge fabrication plant costs tens of billions and the equipment starts losing value immediately. Utilisation is not an operational detail, it is the entire business.
The Cost Structure
A leading edge semiconductor fabrication plant requires capital expenditure in the tens of billions of dollars. The lithography equipment inside it is among the most complex machinery manufactured anywhere, produced by a very small number of suppliers.
Once built, the incremental cost of producing an additional wafer is comparatively small. The economics are dominated by depreciation on assets that must be recovered whether or not the plant is busy.
This is the defining feature. A fab running at full capacity generates excellent margins. The same fab at 60 percent utilisation can lose money, because the depreciation does not pause.
The question is never what a chip costs to make. It is whether the plant is full, because everything else follows from that.
The Foundry Split
The industry separated into two structures. Integrated device manufacturers design and fabricate their own chips. Fabless companies design chips and outsource production to a foundry that manufactures for many customers.
The fabless and foundry split won for a structural reason: the foundry aggregates demand from many designers, which keeps its plants full in a way no single company's own demand could. Utilisation is the constraint, so the model that maximises utilisation wins.
The consequence is severe concentration. Leading edge manufacturing capability sits with very few firms, because only companies with enough aggregated demand can justify each generation of investment.
Why Each Generation Costs More
Every process node, meaning each step toward smaller transistors, requires more expensive equipment and more complex processes than the last. The cost of a new fab has risen substantially with each generation.
That rising cost eliminated competitors one by one. Firms that could not fund the next node stopped trying to compete at the leading edge and shifted to mature processes, where the equipment is depreciated and the products are less demanding.
| Segment | Economics |
|---|---|
| Leading edge | Enormous capex, few players, highest prices |
| Mature nodes | Depreciated equipment, steady demand |
| Specialty processes | Analog, power, sensors, different competition |
Mature node capacity is a genuinely good business precisely because the capital is already sunk. The 2021 shortage was concentrated in exactly this category, where the chips are cheap, essential to cars and appliances, and nobody had built new capacity because the returns did not justify it.
Yield
Manufacturing defects mean not every chip on a wafer works. Yield, the proportion that function correctly, determines the effective cost per working chip.
On a new process, early yields can be poor and improve over months as the process is refined. A company that reaches acceptable yield faster than a competitor has a genuine cost advantage on identical equipment, which is why process engineering expertise is guarded so carefully.
Yield also interacts with chip size. A larger die has more chance of containing a defect, so very large chips are disproportionately expensive to produce. That relationship is behind the industry move toward assembling smaller pieces into a package rather than manufacturing one large die.
The Cycle
Semiconductors are famously cyclical, and the mechanism is capacity lead time. A fab takes years to build, so capacity is committed based on a demand forecast that is years old by the time it arrives.
Shortage encourages investment. The investment arrives after demand has normalised. Oversupply follows, prices fall, investment stops, and the shortage returns. Long lead times and long lived assets produce a cycle that no participant can escape by forecasting better.
The Political Layer
Concentration of leading edge capacity in a small number of locations became a geopolitical concern rather than an industrial one. Government subsidy programmes in several regions now fund domestic capacity for reasons of supply security rather than economic return.
This distorts the ordinary economics. Capacity built with subsidy does not need to clear the same return hurdle, which affects everyone competing against it.
The Bottom Line
Foundry economics are fixed cost economics. Utilisation determines profitability, aggregating demand from many customers is what keeps plants full, and rising capital costs per generation have concentrated the leading edge into very few hands. Yield execution separates competitors using identical equipment, and multi year build times guarantee a cycle nobody can forecast their way out of.