What Foundry Concentration Means for Supply Risk
Leading-edge chip manufacturing in very few hands is not a tail risk to file away — it is a hidden common factor acting on portfolios daily through pricing, allocation, and correlation.
Most sophisticated investors have already processed foundry concentration — as a geopolitical scenario. It sits in the mental file marked tail risk: low probability, catastrophic severity, nothing to do until the headlines change. That filing is the mistake. Concentration in leading-edge chip manufacturing is not a dormant scenario; it is an active economic structure that works on a portfolio every day, through channels that have nothing to do with conflict — and the portfolios most exposed to it are frequently the ones that hold no semiconductor names at all.
Here is the framing we use: foundry concentration converted a component cost into a systemic input, and a systemic input behaves like a hidden common factor. When the manufacture of advanced logic sits with very few firms — and at the true frontier, effectively one — every business whose products, cost structure, or roadmap depends on advanced silicon shares a supplier several layers down, whether it knows it or not. Software platforms, cloud infrastructure, autos, industrial equipment, medical devices, consumer hardware: the exposure does not appear in any of their filings as a named risk with a number attached, but it is common to all of them, which is precisely what makes it a factor rather than a stock-level fact. The work of this framework is to move foundry concentration out of the tail-risk file and into the daily one, by naming the three channels it acts through.
Three layers of why the concentration exists — and persists
The three channels: pricing, allocation, correlation
The pricing channel is the everyday one. A firm that is the only practical source of frontier manufacturing prices like it: wafer pricing at the leading edge reflects scarcity value rather than cost-plus competition, and those economics flow downstream into the cost of every advanced product built on it. For the portfolio, this cuts two ways. Holdings that consume advanced silicon face a structural input cost floor that competition will not erode on any near timescale; the concentrated manufacturer and the layers that supply it — the equipment barbell treated in the companion piece — sit on the collecting side of the same rent. Concentration, in other words, does not just create risk; it dictates where a durable share of the industry's economics settles, and a portfolio is on one side of that flow or the other whether it has considered the question or not.
The allocation channel operates whenever capacity is tight. In a competitive market, shortage resolves through price alone; in a concentrated one, it resolves through rationing — someone at the manufacturer decides who gets wafers, guided by contract terms, prepayments, strategic importance, and relationship history. In those periods, queue position becomes a competitive asset as real as intellectual property: firms holding long-term capacity agreements ship product and take share, while firms buying capacity as they need it discover that money alone cannot manufacture their product. The equity-relevant point is that allocation outcomes redistribute market share within industries during shortages, so concentration risk is not only a macro exposure — it decides relative winners inside sectors that seem far from silicon.
The correlation channel is the one portfolio math cares about most and prices least. Diversification works by combining exposures that fail independently; a common supplier several layers deep quietly deletes that independence. A portfolio spread across cloud platforms, device makers, autos, industrials, and medical technology looks diversified at the sector level while sharing a single manufacturing dependency at the input level — so a serious disruption to frontier capacity would arrive not as a stock event but as a factor event, marking down seemingly unrelated holdings together. Sterling's Embedded Intelligence treats this as a mapping exercise rather than a forecasting one: the useful output is not a probability estimate for a disruption but an honest count of how much of the portfolio shares the dependence, because that count — not the scenario odds — is what the owner actually controls.
The illustration: when allocation decided who could build cars
The automotive chip shortage that began in 2020 is the cleanest available demonstration of the allocation channel, and it is worth reading as history with the mechanism in view. When the pandemic first hit, carmakers cut their component orders sharply, expecting a long demand slump. Consumer electronics demand surged instead, and the capacity the auto industry had walked away from was reallocated to customers who wanted it. When vehicle demand rebounded much faster than the carmakers had assumed, they returned to the queue — and found their place gone. The result was one of the strangest industrial episodes in memory: assembly lines for vehicles idled for want of inexpensive, mostly trailing-edge parts, with the missing component often worth a vanishing fraction of the finished product it was blocking.
Three durable lessons sit in that episode. First, the allocation channel is real and decisive: in a rationed system, the order in which you stand matters more than the price you are willing to pay, and firms that had maintained their commitments shipped while firms that had cancelled waited. Second, the exposure extends far beyond the leading edge — trailing-edge capacity is more diversified by firm, yet still concentrated enough by process and geography that a demand whipsaw rationed a global industry. Third, the damage appeared in equities that no semiconductor screen would have flagged: the shortage expressed itself as lost production, lost revenue, and pricing chaos in the auto complex — the correlation channel surfacing in an industry filed nowhere near technology. A single episode, no conflict required, and every channel of the framework visible at once.
The strongest case against this framework
The serious objection is that the structure is already correcting. Governments are funding fabs across multiple regions precisely to dilute the concentration; major customers are qualifying second sources and spreading designs across manufacturers where they can; and the concentration was always overstated anyway, because it describes the leading edge while most of the world's chips are built on mature nodes with a genuinely broader supplier base. On this reading, concentration risk peaked and is now being competed and subsidised away — and pricing a structural premium on it means paying for a risk in decline.
The rebuttal begins with a distinction the objection blurs, and it is the second reusable idea of this piece: concentration has two axes — who and where. Firm concentration determines pricing power and allocation authority; location concentration determines physical and political disruption risk. Much of the visible correction addresses the second axis without touching the first: when the same dominant manufacturer builds capacity in additional countries, location risk genuinely falls while firm concentration — the rents, the queue, the single point of pricing — remains exactly as it was. Subsidised capacity operated by others, meanwhile, runs into the learning-curve physics from the reasoning chain: replicated capacity is higher-cost capacity, and building fabs does not transfer the tacit knowledge that makes frontier manufacturing work. Such capacity is best understood as insurance rather than competition — worth having, unlikely to discipline pricing on any near horizon. Second-sourcing at the frontier is constrained by design economics: porting an advanced design between processes is expensive enough that second sources tend to trail by a generation, which is diversification of a kind, but not at the edge where the value sits. The trailing-edge point we simply grant — with the auto episode as the caveat that even the diversified end rationed a global industry when the cycle whipsawed. The honest synthesis: location risk is genuinely improving, firm concentration is not, and a portfolio owner should track the two separately because almost every reassuring headline concerns the first while almost every economic consequence flows from the second.
How to apply this framework
- Map the embedded exposure before estimating the scenario. For each major holding, ask whether its products, cost structure, or roadmap depend on advanced silicon, and from whom. The output is a count of how much of the portfolio shares one supplier — the number diversification math silently assumes is small.
- Separate the who from the where. Capacity announcements in new regions reduce location risk; only genuine competitive capability at the frontier reduces firm concentration. Read every resilience headline against that distinction — most address the axis that does not set prices.
- Treat wafer pricing at the leading edge as the rent gauge. Sustained pricing power there is the concentration expressing itself in the ordinary course of business — and a reminder that the exposure has a collecting side a portfolio can hold as well as a paying side it already holds.
- In tight markets, ask about queue position, not just demand. When capacity rations, long-term agreements and prepaid capacity decide who ships and who waits. Within any hardware-dependent industry, allocation terms are a competitive asset that separates relative winners before the income statement shows it.