How Do You Spot a Semiconductor Cycle Peak Before the Filings Confirm It?

Global semiconductor sales run in pronounced multi-year cycles, and the industry's own data flagged the last two turns: Semiconductor Industry Association billings peaked in early 2022 and fell by roughly a quarter to their 2023 trough, per SIA monthly data, months before most earnings reports conceded it. Peaks are visible early if you watch the right gauges. Market Today publishes information, not investment advice, and this analysis reads the cycle's instruments, not any company's prospects.

What drives the semiconductor cycle in the first place?

Capacity algebra. Building a leading-edge fabrication plant takes years and tens of billions of dollars, while demand for chips can double or halve within eighteen months — phones, personal computers, data centers, and cars ordering in waves that synchronize. When end demand rises, buyers double-order to secure supply, which the supply chain reads as more demand than exists; when demand cools, those same orders cancel simultaneously and inventories pile up. The 2020–2022 surge and the 2023 correction both followed this script, per company commentary and SIA data. Nothing about the mechanism requires bad faith; it requires only long lead times and nervous procurement managers.

Which gauges turn first at a peak?

Three, in reliable order. First, inventory: days of inventory at chipmakers and their customers rise while sales still look healthy — the 2022 peak showed channel inventories building for quarters before revenue rolled over, per company 10-Q disclosures. Second, lead times: the interval from order to delivery, tracked in industry surveys and supplier commentary, collapses from extended backlogs toward normal — a sign the queue of real demand has ended. Third, pricing: spot prices and negotiated annual price cuts appear while contract volumes still grow. Revenue, the gauge most investors watch, is the last dial to move, which is why cycle peaks are always obvious in retrospect and contested in real time.

What did the 2022 peak and 2023 trough actually look like?

The bust was textbook in shape and brutal in size. Worldwide chip billings fell from a record 2021-2022 — annual sales had crossed 600 billion dollars for the first time, per SIA/WSTS data — to a 2023 trough roughly eleven percent below the prior year on an annual basis, with memory chips, the most commodity-like segment, falling far harder: memory revenue dropped by roughly half peak-to-trough, per company filings. Memory leads the cycle in both directions because its products are interchangeable and its buyers can pause purchases entirely. The recovery from late 2023 into 2024–2025 was led by the same segment plus AI accelerators, with industry sales setting fresh records — confirming the cycle's amplitude rather than refuting it.

Where does AI demand fit the cycle framework?

As the force that lifted the 2024–2025 upcycle and the question mark hanging over its duration. Accelerator chips became the industry's largest single demand driver, with the dominant supplier reporting tens of billions in quarterly data-center revenue, per its filings — a concentration of demand that helped push industry billings to new records. The skeptical questions are the cycle's own: how much of accelerator ordering reflects deployment versus inventory building at system makers, and what happens to order books when cloud capital expenditure plans flatten for even two quarters. Neither question is answerable from outside; both have answers visible eventually in inventories, lead times, and the dominant supplier's backlog disclosures. The framework does not predict the turn — it tells you where to look first.

History also supplies the base rate for skepticism. During the 2021–2022 boom, extended lead times were widely read as proof of durable structural shortage; within three quarters, the same order books were being cancelled and memory prices were collapsing. Double-ordering amplifies both directions, which is why survey language about "improved visibility" deserves the same discount at the top that panic about allocation deserved at the bottom.

What are the segment differences that matter?

The cycle is not one wave but several, offset by segment.

Rough as the table is — depths are approximate per company filings and SIA data — the ordering has held across several cycles: memory turns first and hardest, analog turns last and longest.

What is the disciplined verdict?

Semiconductor cyclicality is a fact of capital-allocation arithmetic, and the instruments that reveal it — inventories, lead times, spot pricing, memory revenue — are public and timely. The error to avoid is treating a demand wave, however genuinely transformative, as the abolition of the cycle; every prior "this time is different" in chips, and there have been several, ended with an inventory correction. The honest position is agnostic about timing and vigilant about the gauges: watch inventory days at the big names, lead-time language in supplier commentary, and memory pricing, and you will rarely be the last to know.

Where can readers track the gauges themselves?

The SIA publishes monthly billings, WSTS publishes semiannual forecasts, companies disclose inventory days quarterly in filings on EDGAR, and lead times appear in supplier and analyst commentary every quarter. The cycle is the industry's most documented feature, which makes ignoring it a choice rather than a misfortune.

A final caution on the data itself: SIA billings count sales, not end demand, so a build in channel inventory registers as strength right up until it registers as returns. Cross-checking billings against end-market data — personal computer shipments, smartphone production, cloud capex — separates real demand from pipeline filling, and the divergence between the two series is itself a peak gauge worth naming.