Semiconductors are the most cyclical major industry in public markets, and the current moment illustrates why better than any textbook.
The forecast
In its spring 2026 forecast, published in May, World Semiconductor Trade Statistics projected the global semiconductor market at $1.51 trillion in 2026, a rise of 90% on the year. The memory segment alone is forecast to surge around 250%, to more than $800 billion.
The rest of the industry grows far more modestly on the same forecast: logic 37%, microprocessors 20%, analog 10%, discrete semiconductors 8%, sensors and optoelectronics 3%. Regionally, WSTS sees the Americas up 112%, Asia Pacific 87%, Europe 58% and Japan 28%. For 2027 it projects about $1.9 trillion, a 27% gain, with memory growth cooling to 32%.
For scale, the Semiconductor Industry Association reported global sales of $791.7 billion in 2025, itself up 25.6%. A 90% year would be the largest expansion in the industry's history, driven by AI infrastructure and high-bandwidth memory.
The share prices
Meanwhile the Philadelphia Semiconductor Index fell 10% last week and sits about 20% below its June record high. South Korea's chip-heavy market lost 4.1% on Monday after falling nearly 9% the previous week.
An industry heading for its best year ever, with its shares in a 20% drawdown. This is not a contradiction. It is the normal behavior of the sector, and understanding why is the single most useful thing an investor can know about chips.
Why the cycle exists
The root cause is a mismatch between how fast demand moves and how slowly supply can respond.
A leading-edge fabrication plant costs billions of dollars and takes years to plan, build, equip and qualify. Demand can shift in a quarter. So when shortage appears, prices rise, margins expand, and manufacturers commit capital to new capacity. That capacity arrives years later, frequently into a market that has already cooled. Supply is added in enormous, indivisible steps; demand moves continuously. The result is alternating shortage and glut, structurally.
Two features make it worse.
The bullwhip. When chips are scarce, customers do not order what they need. They order what they hope to be allocated, and they order from several suppliers at once to improve their odds. When the shortage breaks, those customers are sitting on months of inventory and stop ordering entirely until it clears. A modest wobble in end demand therefore arrives at the fab as a collapse in orders. Order books swing far more violently than actual consumption, and the fab is at the end of the whip.
Memory is a commodity. This distinction is routinely lost and it matters enormously. DRAM and NAND are largely interchangeable between suppliers, so they price on spot markets and margins move with breathtaking speed in both directions. Foundry and logic work, by contrast, is sold on contracts negotiated in advance with named customers. That is why memory makers can swing from spectacular profitability to losses within a few quarters while foundry margins move far less. WSTS's own forecast makes the point: memory up 250%, logic up 37%.
Why the shares move first
Equity prices reflect expected future earnings, not reported ones. Chip stocks therefore tend to peak while earnings are still climbing and bottom while losses are still deepening.
That is precisely what makes the sector so confusing to watch. By the time a memory maker reports a record quarter, the market has usually been trading the next turn for months, working from inventory levels, spot pricing and capex announcements rather than the income statement. A stock falling on a blowout quarter is not irrational; it is the market concluding that the peak is visible.
Applied to today: a 90% growth forecast is a statement about 2026 revenue. A 20% drawdown in the index is a statement about what investors expect to happen after that, and about whether the capital being committed now will meet demand or overshoot it.
What to watch instead
Traditional cycle indicators have thinned out. SEMI discontinued the widely followed North American book-to-bill ratio in 2017, and now publishes equipment billings instead. Those billings reached $36.55 billion in the first quarter of 2026, up 14% year on year, which measures how much capacity is being built and is therefore a forward read on future supply.
Equipment orders are the cleanest early signal available: they reveal the capacity that will arrive in two years and, historically, the point at which they peak has tended to precede the price downturn that follows the new supply.
Is AI different?
This is where informed people genuinely disagree, and the honest answer is that it is unresolved.
One view holds that AI infrastructure represents a structural reallocation of corporate capital rather than a conventional inventory cycle, that demand comes from a handful of very large buyers with multi-year plans, and that high-bandwidth memory is more contracted and less commodity-like than traditional DRAM. Memory manufacturers have also moved toward longer supply agreements at fixed pricing, which if it holds would dampen the amplitude of the swing.
The other view is that every prior boom also felt structural at the time, that a 250% forecast revenue increase in one segment is the definition of a supply response waiting to happen, and that the buyers concentrating today's demand are precisely the ones capable of pausing spending abruptly.
What the record supports is narrower than either camp: the industry has never repealed its cycle, and the largest upswings have preceded the largest corrections. Whether the amplitude is smaller this time is a claim about the future. The 90% forecast and the 20% drawdown are both facts about the present, and they are not in conflict.
