The Boom-Bust Pattern
The semiconductor industry has been cyclical since its inception. Demand surges, leading to capacity shortages and rising prices. Companies invest heavily in new fabs. By the time those fabs come online (2-3 years later), demand has cooled, creating overcapacity and falling prices. Margins compress, investment slows, inventories deplete, and the cycle begins again.
The major cycles since 2000 tell the story:
- 2000-2001: Dot-com bust cratered semiconductor revenue by 32%. The deepest downturn in industry history at that point.
- 2008-2009: Financial crisis drove a 9% revenue decline. Memory makers were hit hardest — Qimonda went bankrupt, DRAM prices fell 90%.
- 2018-2019: Memory oversupply after a massive build-out. Samsung's memory operating profit dropped 80% from peak to trough.
- 2020-2022: Pandemic-driven demand surge, followed by supply chain chaos. Chip shortages cost the auto industry $210+ billion. Companies double-ordered, creating a bullwhip effect.
- 2022-2023: The correction. Consumer demand crashed, inventories bloated. PC and smartphone chip markets declined 15-20%. Memory prices fell 50%+.
- 2024-2025: AI-driven recovery, but unevenly distributed. NVIDIA and AI chip makers boomed while analog and automotive segments stayed weak.
Why Cycles Persist
You'd think a mature industry would learn to avoid these cycles. It doesn't, for structural reasons:
Long investment lead times — a decision to build a new fab takes 2-3 years to result in production output. When demand is booming and prices are high, the rational response is to invest. But every competitor makes the same calculation simultaneously, leading to collective overinvestment. By the time the capacity arrives, the demand environment has changed.
Commodity memory amplifies the cycle. DRAM and NAND flash are fungible products where pricing is set by spot markets. Small imbalances between supply and demand cause disproportionate price swings. A 3% oversupply in DRAM can cause a 30% price decline because buyers know they have bargaining power and defer purchases, creating a negative feedback loop.
We covered a related topic in Display Driver ICs: OLED Panel Control, High Resolution, and.
Customer behavior contributes. During shortages, customers double-order — placing orders with multiple suppliers for the same chips, intending to cancel the extras once supply normalizes. This inflates apparent demand and encourages even more capacity investment. When normalization comes, the cancellations hit all at once.
Capital intensity creates exit barriers. Shutting down a fab doesn't save much money — the depreciation and debt service continue regardless. It's often cheaper to run a fab at low utilization, selling wafers below full cost but above marginal cost, than to idle it. This means supply doesn't shrink quickly even when demand drops, prolonging the downturn.
Memory Cycles: The Sharpest Peaks and Valleys
Memory pricing is the semiconductor industry's most visible cyclical indicator. Samsung, SK Hynix, and Micron together control 95%+ of the DRAM market and a large share of NAND. Their financial results swing wildly — Samsung's memory division has oscillated between operating margins of 50%+ in boom years to near-zero or negative in bust years.
The memory oligopoly has actually dampened cycles somewhat compared to the early 2000s, when there were six or seven DRAM manufacturers engaged in ruinous competition. With three players, there's more capacity discipline — when prices fall, the big three cut capex more quickly than a fragmented market would. But they still can't prevent cycles entirely because they each have strategic reasons to maintain technology leadership that require continuous investment.
Related reading: Photoresist Chemistry: The Materials That Define Modern Lith.
HBM (High Bandwidth Memory) has introduced a new dynamic. AI accelerator demand for HBM has created a premium memory segment with better margins and more stable pricing than commodity DRAM. SK Hynix and Samsung are prioritizing HBM production, which reduces their available capacity for standard DRAM and helps support mainstream DRAM prices. It's a structural shift that might moderate future memory cycles — or it might just add another variable to an already complex market.
Demand Forecasting: Harder Than It Looks
Semiconductor demand forecasting is notoriously difficult because the industry serves multiple end markets with different dynamics:
- Smartphones — 1.2 billion units per year, relatively predictable but with slowing growth
- PCs — 250-270 million units per year, subject to replacement cycle dynamics and work-from-home effects
- Automotive — growing rapidly (content per vehicle is increasing from ~$500 in 2020 to projected $1,000+ in 2030) but complicated by the EV transition
- Data center — driven by cloud spending and now AI training/inference, which has been wildly unpredictable
- Industrial and IoT — diverse, long-lifecycle, harder to forecast in aggregate
The AI demand spike starting in 2023 illustrates the forecasting challenge. Nobody in the semiconductor industry's planning cycle anticipated NVIDIA's data center GPU revenue growing from $15 billion in fiscal 2023 to over $47 billion in fiscal 2024. That kind of step function demand change breaks every model.
Inventory Cycles Within the Larger Cycle
Superimposed on the 3-5 year capacity cycles are shorter inventory cycles of 12-18 months. Distributors and OEMs build or draw down buffer stocks based on their demand outlook, and these inventory movements amplify or dampen the underlying demand signal that reaches chip manufacturers.
This connects to the ideas in Quantum Computing Chips in 2026: Superconducting, Trapped Io.
During the 2021 shortage, many companies built 6-12 months of safety stock. When demand softened in late 2022, they stopped ordering new chips and consumed existing inventory. Chip companies saw orders evaporate seemingly overnight, even though end-market demand hadn't declined as sharply as the order pattern suggested. The inventory correction lasted through most of 2023 before restocking began.
Tracking distributor inventory levels (reported by companies like Arrow, Avnet, and Mouser) is one of the best leading indicators for the semiconductor cycle. When distributor inventories exceed 8-9 weeks of forward demand, a correction is likely coming. When they drop below 6 weeks, a recovery is usually underway.
Investment Implications
Understanding the cycle matters for anyone investing in or planning around the semiconductor industry. A few patterns tend to repeat:
- Equipment companies (ASML, Applied Materials, Lam Research, KLA) lead the cycle because capex decisions precede production by 1-2 years
- Memory makers have the highest beta to the cycle — biggest gains in booms, biggest losses in busts
- Foundries (especially TSMC) are more stable because they serve diverse end markets and have pricing power at leading-edge nodes
- Fabless design companies vary — those with diversified product lines are more stable, while single-product companies can see extreme swings
The trend toward longer-term supply agreements and more geographically distributed manufacturing (driven by the CHIPS Act and its equivalents) may moderate future cycles somewhat. But the fundamental dynamics — long lead times, capital intensity, commodity pricing in memory, and human psychology around shortage and surplus — aren't going away. The semiconductor industry will remain cyclical for as long as it takes years to build a fab and weeks for demand to change.