Are semiconductor stocks undervalued after a selloff?

A cyclical looks cheapest exactly when it is most expensive. Here is why price-to-earnings breaks down in semiconductors, what to use instead, and how to tell a genuine cyclical bargain from a peak-earnings trap.

By Nathan Wickham-Hurd · Founder, Oak Growth · First-class Economics & Finance, MBA · Last reviewed August 2026

Context, August 2026. Semiconductor and memory shares have sold off sharply from their June highs as investors reduced exposure to the AI trade on concerns about the scale of capital spending, with money rotating into healthcare, financials and real estate. Several memory names are down by half or more from their peaks even as reported earnings remain strong. Market conditions change; the framework below is meant to outlast them.

Why a chip stock halving does not make it cheap

Semiconductors are the most cyclical major industry in the market, and cyclicals break the usual valuation rules in a way that catches out careful investors. The trap has a name: a cyclical looks cheapest exactly when it is most expensive.

At the top of a cycle, earnings are at a record, so the price-to-earnings ratio looks low. Then demand normalises, earnings fall by 60–80%, and the same share price implies a multiple three times higher than it appeared. A chip company on eight times peak earnings can be on thirty times normalised earnings without the price moving at all.

Peak year: price 100 / EPS 12.5 = 8× Normal year: price 100 / EPS 3.5 = 29×

This is why P/E is close to useless in this sector without knowing where you sit in the cycle, and why a 50% fall tells you nothing on its own.

What to look at instead

Normalised earnings, not this year's

Average operating profit across a full cycle — ideally seven to ten years, covering at least one downturn — and value the business on that. If the current year is more than roughly double the average, you are looking at peak earnings and should assume they mean-revert.

Capital intensity

Leading-edge fabs cost enormous sums and depreciate on a fixed schedule regardless of demand. Check capital expenditure as a share of revenue over several years and whether free cash flow survives a bad year. Fabless designers and equipment makers have very different economics from memory manufacturers — treating “semiconductors” as one sector is a mistake.

Whether the moat is design or scale

Some chip businesses have genuine moats: instruction-set ecosystems, decades of process leadership, tools nobody else can build. Commodity memory has almost none — it is a capacity race where high prices invite new supply that eventually kills the price. Same sector, completely different durability.

Customer concentration and the capex question

When a handful of hyperscale buyers account for the bulk of demand growth, the question is not whether AI is real but whether those buyers keep spending at the current rate. Their capital budgets are discretionary and can be cut in a quarter. Work out how much of the company's revenue depends on that continuing.

Inventory

Rising inventory days ahead of a downturn is the classic tell in this industry. Compare inventory growth with revenue growth — when inventory grows materially faster, a price correction usually follows.

So is the selloff an opportunity?

Sometimes, and the way to know is not the size of the fall. Ask what the business earns in an average year, what that stream is worth discounted back, and whether today's price sits below it with room to spare. If the answer depends on the current year's earnings persisting indefinitely, you don't have a valuation — you have a bet on the cycle.

The honest position: nobody knows whether AI capital spending sustains at these levels. That uncertainty is precisely why the margin of safety matters more here than almost anywhere else. See also our take on whether AI stocks are overvalued.

Value chip stocks on what they earn through a cycle

Oak Growth estimates intrinsic value and margin of safety across roughly 1,000 companies, with the moat, cash flow and debt checks that separate a cyclical bargain from a peak-earnings trap.

Explore Oak Growth

Common questions

Are semiconductor stocks undervalued after a selloff?

A large fall doesn't establish value in a cyclical sector. Chip companies look cheapest on price-to-earnings at the top of the cycle, when earnings are at a peak that won't persist. Value them on average earnings across a full cycle instead, and check capital intensity, inventory and customer concentration.

Why is P/E misleading for chip stocks?

Because the earnings figure is unstable. At a cyclical peak the E is unusually large, so the ratio looks low; when demand normalises earnings can fall by more than half and the same share price implies a multiple several times higher. A low P/E in this sector is often a warning rather than a bargain.

Are memory stocks different from other semiconductor companies?

Substantially. Commodity memory is a capacity business with weak pricing power — high prices attract new supply which eventually collapses the price. Fabless designers, equipment makers and analogue specialists have very different economics and far more durable competitive positions.

How do you value a cyclical company?

Estimate what it earns in an average year rather than the current one, ideally averaging operating profit across seven to ten years including a downturn, then discount that stream. Insist on a wide margin of safety, because the timing of cycles is not forecastable.

Also see: Are AI stocks overvalued? → · Is AI a bubble? → · What is a value trap? →