Semiconductors are one of the most cyclical industries in the public markets, and the engine of that cyclicality is the mismatch between slow, lumpy, capital-intensive supply and fast, volatile end demand. Building a fab takes years and billions of dollars; once running it must operate near full utilization to cover its enormous fixed costs, so supply is inelastic in the short run. Demand for chips, by contrast, swings with the consumer-electronics, datacenter, auto, and industrial economies — and those swings get amplified on the way up the supply chain through inventory behavior. The result is a recurring boom-bust pattern that swamps the fundamentals of even well-run companies and makes "where are we in the cycle?" the single most important question for a semiconductor stock.

How the cycle is formed

Two distinct cycles overlap, operating on different clocks:

• The capacity cycle is the slow loop. Fabs are committed 18–36 months ahead of output (TMT IB Guide), and a new leading-edge or DRAM fab can run $15–20B and 2–3 years (UncoverAlpha). Because competitors read the same demand signal during a boom, they tend to commit capacity simultaneously — overshooting in unison and seeding the next glut. Peak-to-peak, the full cycle is commonly described as 3–5 years, though durations vary widely.

• The inventory cycle is the faster loop, and it is where the bullwhip effect lives. When lead times stretch, downstream buyers double-order and pad safety stock; modest end-demand changes get magnified at each link. One illustrative framing: a ~10% rise in smartphone demand can become ~20% more orders to chip suppliers, ~30% to foundries, and ~40% in equipment orders (TMT IB Guide). When demand softens, the same mechanism runs in reverse: customers stop ordering and burn down stock, so chipmaker revenue falls far more than end consumption does.

The four phases — upcycle → peak → correction → trough/recovery — show up in the standard indicators. Lead times extend from roughly 8–12 weeks to 20–50+ weeks into a peak; book-to-bill, capacity utilization, and ASPs roll over into the correction.

How it's used in practice

Analysts and traders read cycle position through a small set of leading metrics, not reported earnings (which lag):

• Book-to-bill ratio — bookings ÷ billings over a period. Above 1.0 signals more orders than shipments (tightening); below 1.0 signals weakening (Wikipedia). The SEMI equipment book-to-bill is a widely watched upstream tell.
• Days of inventory (DOI) — the most direct gauge of the inventory cycle. In the 2022 correction, blended industry DOI rose to ~212 days at Q4 2022 from ~167 days a quarter earlier (Synovus, citing the data). Falling DOI toward historic lows (memory makers reportedly near ~2–3.3 weeks at 2018/2025 tightness, per UncoverAlpha) signals the setup for a price recovery.
• Capacity utilization, lead times, and ASP/pricing trends round out the dashboard.
• Capex and WFE spending confirm where management believes it is — Gartner forecast a ~19% drop in both capex and wafer-fab-equipment spending for 2023 (Gartner).

The professional discipline is counter-cyclical: valuations look cheapest (high trailing P/E or losses) near the trough and richest near the peak, so trailing multiples invert the usual signal. Importantly, semiconductor stocks tend to lead the fundamentals by roughly 6–12 months — the SOX index bottomed in October 2022 (down ~45% from its January 2022 high, per Synovus) while inventories were still rising, then returned roughly +58% in calendar 2023 (price return, per Investing.com SOX history) even as full-year 2023 industry sales still fell. Waiting for "good numbers" means buying after the move.

Adoption, debate & evidence

The existence of the cycle is not seriously contested — it is documented across decades, and the 2023 downturn was widely described as roughly the seventh since 1990 (cycle-count is industry commentary, not an audited figure). The downturn's depth is well documented: global semiconductor sales fell 8.2% to $526.8B in 2023 (from a record $574.1B in 2022) with memory the hardest hit, dropping to ~$92.3B (about a −29% year-over-year decline), per the Semiconductor Industry Association / WSTS. (Early-2023 forecasts had been steeper — roughly −9% overall and −35% memory — illustrating how forecasts overshoot the realized trough.) What's genuinely debated is whether the cycle is structurally dampening. Bulls argue diversification into autos and industrial smooths aggregate demand; Gartner's framing in 2023 was the opposite — "cyclicality is back" — and the data showed individual end markets (PCs, phones) still correcting violently even as autos held (Gartner). The honest read: aggregate volatility may be moderating, but segment-level cyclicality has not disappeared.

Cyclicality is also highly uneven by segment:

• Memory (DRAM/NAND) is the most cyclical. It is a true commodity ("a bit is a bit"), made by an oligopoly (Samsung, SK Hynix and Micron together hold roughly ~90% of the DRAM market, per Counterpoint Research) that legally cannot coordinate output. ASPs and margins swing violently peak-to-trough — memory makers posted heavy operating losses through the 2023 trough (SK Hynix, for example, reported a deep negative full-year 2023 net margin), then rebounded sharply on the AI/HBM upturn.
• Analog and embedded are the least cyclical. TI states product life cycles of 10–15 years (TI.com); Analog Devices cites industrial sockets lasting ~17 years on average with steady pricing (techinvestments.io). Sticky designs and slow obsolescence dampen the inventory whip.
• Logic/foundry sits in between, with leading-edge nodes more exposed to consumer demand.

Strengths & limitations

Cycle analysis is powerful because the leading indicators are real, published, and physically grounded in capacity that cannot turn on a dime — unlike many technical signals, the bullwhip has a causal mechanism. Its limitations are timing and noise: the metrics tell you direction and extreme, not the exact turn. The #1 misuse is trailing-valuation anchoring — buying because the trailing P/E looks low at the peak (earnings are about to collapse) or refusing to buy at the trough because the company is losing money. A second trap is treating "the cycle" as monolithic; a memory glut and an analog shortage can coexist. Secular demand shocks (e.g., an AI-datacenter build-out) can also temporarily override the normal inventory rhythm for some products while others still correct.

Sources

• TMT IB Guide — The Semiconductor Business Cycle (capacity vs inventory cycle, bullwhip, lead times)
• UncoverAlpha — Every Memory Cycle Ends the Same (memory cyclicality, oligopoly, fab cost/time, inventory weeks)
• Book-to-bill ratio — Wikipedia (definition)
• Gartner — Cyclicality Is Back (Feb 2023) (capex/WFE forecast, debate)
• Synovus — Has the Semiconductor Sector Bottomed? (Feb 2023) (DOI figures, downturn count, SOX Oct-2022 bottom, stocks-lead-fundamentals)
• Semiconductor Industry Association / WSTS — 2023 sales fell 8.2% to $526.8B (final 2023 industry + memory sales)
• Investing.com — PHLX Semiconductor (SOX) historical data (~+58% calendar-2023 price return)
• Counterpoint Research — Global DRAM market share (big-three ~90% DRAM concentration)
• TI.com — Product life cycle and techinvestments.io — Industrial analog semis (analog 10–15yr / 17yr socket life)

Dispute flagged: whether semiconductor cyclicality is structurally diminishing is genuinely contested — aggregate volatility may be moderating while segment-level cyclicality persists. The 2023 sales figures are the final SIA/WSTS numbers (which came in shallower than early-2023 forecasts). Some narrative figures — the cycle-count "seventh since 1990", specific inventory-week lows, and individual-company margins — are industry commentary and should be treated as commonly-cited illustration, not audited precision.