How Semiconductor Cycles Actually Break

The misconception: end demand is the least volatile link

Semiconductor content travels through a long chain: wafer fabs and packaging, then chip suppliers, then distributors, then contract manufacturers, then brand OEMs, then retail or enterprise buyers. Every link holds inventory, and every link sets its order rate not on what it sold last month but on what it expects to sell and what it fears about availability. That second term is the whole story.

Because each tier orders to replenish a target stock level as well as to satisfy demand, a modest change in final consumption is amplified as it propagates upstream — the bullwhip effect, familiar from industrial economics but unusually violent in semiconductors because lead times are long, the products are hard to substitute, and a missing three-dollar controller can idle a finished good worth thousands. The upstream supplier therefore experiences volatility that has very little to do with how many phones or cars were actually sold. Sterling's Embedded Intelligence treats this as the founding distinction of the sector: end demand sets the trend, the channel sets the cycle. Sterling's read is that most disappointing chip investments are misdiagnosed demand calls that were really inventory calls.

The mechanism: lead time masquerading as demand

The central mechanism is a signalling failure. In a shortage, the price a customer pays is not the only cost of a chip — the wait is a cost too. When lead times stretch, the rational response for a purchasing manager is to order earlier, order more, and order from more than one supplier. None of that is irrational hoarding; it is prudent supply assurance. But it is indistinguishable, in the supplier's order system, from genuine demand growth.

This is why the most useful early evidence is a divergence rather than a level. Distributor sell-in running above sell-through; inventory days rising at the chip supplier while revenue is still growing; a book-to-bill below one against a backlog management still describes as robust; the first customer request to push out a delivery date rather than cancel it. Each of these says the same thing: the marginal buyer has stopped fearing scarcity. Capex guidance cuts and formal demand warnings come later, and they are confirmations, not signals — by the time a company concedes the cycle, the equity has usually already been repriced.

The illustration: shortage to glut, without a demand collapse

Why the break is asymmetric, and why margins move more than revenue

Two structural features make the downside sharper than the upside. The first is fixed-cost intensity. A leading-edge fab is an enormous depreciating asset with a largely fixed cost base; utilization, not volume, is the profit variable. When wafer starts fall, the depreciation and fab overhead do not, so gross margin compresses faster than revenue declines. The same operating leverage that produces spectacular incremental margins on the way up runs in reverse with equal force.

The second is pricing behaviour in commodity segments. Where the product is fungible — most obviously in commodity memory — sellers with sunk capacity and a spot market will rationally price toward marginal cost to keep fabs loaded, because idling capacity destroys yield learning and still incurs depreciation. Differentiated analog, mature-node microcontrollers and custom logic have more price discipline and longer contract structures, which is why their corrections tend to show up as volume and utilization declines first and price concessions much later. The practical consequence for a portfolio owner is that segment identity determines the shape of the break, not just its timing: commodity exposure breaks through price, differentiated exposure breaks through volume and eventually through the inventory the customer is sitting on.

The chain does not break all at once

Sequencing is where most of the analytical edge lives. Consumer-exposed and commodity segments typically inflect first, because their channels are shortest and their pricing is most transparent. Industrial and automotive analog inflects later, because design cycles are long, qualification makes substitution hard, and the inventory sits further downstream where visibility is worst. Equipment — the tool makers and their suppliers — has the longest apparent immunity, because backlog and deferred revenue cushion reported results for several quarters after the chip makers' order books have turned. That cushion is precisely why equipment estimates can look resilient into a downturn and then air-pocket when the backlog is worked off and new orders have not replaced it.

Design-heavy fabless businesses sit in a different position again. They own no fabs, so their fixed-cost leverage is lower and their downside margin protection is better — but they carry inventory risk and they lose the pricing umbrella that scarcity provided. Integrated device manufacturers keep the upside of utilization and absorb the full brunt of the empty fab. Reading the cycle correctly means holding both facts at once: the same downturn produces different injuries at different points in the chain, and the equity market usually prices them in roughly the order described above rather than simultaneously.

The strongest competing view — and what survives it

The most serious objection is that this framework is dated: that accelerated computing, electrification, industrial automation and the sheer growth of semiconductor content per device have made the sector secular rather than cyclical. Customer concentration among a handful of hyperscale buyers, multi-year long-term agreements, prepayments and state-subsidised capacity all argue that the old boom-bust reflex has been dampened. This view deserves respect, and parts of it are simply correct. Content growth per unit does raise the trend rate. Sovereign subsidy does decouple some capacity decisions from near-term returns. Concentrated, contractually committed customers do give suppliers better forward visibility than a fragmented distributor channel ever did.

But the framework describes a capacity-and-inventory phenomenon, not a demand-growth phenomenon. A faster trend line does not eliminate the second derivative. Capacity still arrives in indivisible, multi-year lumps and is still financed out of cash flows earned at peak margins — which is the classic recipe for over-building. Long-term agreements do not create end demand; they relocate the inventory risk onto whichever counterparty is less able to walk away, and history suggests commitments signed in scarcity get renegotiated in glut. Customer concentration cuts both ways: fewer buyers means better visibility on the way up and a far more abrupt signal when one of them decides to digest what it has already installed. Subsidised capacity, if anything, worsens the eventual supply overhang because it is less responsive to price. The honest conclusion is that cycles in parts of the chain can be longer and shallower than they once were, and that the secular case can be right about the decade while the cyclical framework is right about the next eight quarters. Both can be true; a portfolio has to be positioned for the one that binds first.

Which leads to the valuation trap that follows from all of this. Because margins peak with utilization and pricing simultaneously, trailing earnings at the top of a cycle are the least representative number in the sector, and a multiple calculated on them will look undemanding at exactly the wrong moment. The equity cycle has historically tended to lead the fundamental cycle — multiples compressing while estimates are still being revised upward, and expanding while reported results are still deteriorating. Anchoring on mid-cycle earnings power, and on the through-cycle return on the capital being committed today, is a more durable discipline than reacting to either the print or the guidance.