Why Semiconductor Capacity Cycles So Sharply

How long lead times, product-specific capacity, inventory amplification, and concentrated equipment supply turn modest demand changes into severe semiconductor shortages and gluts.

A chip shortage is never just a shortage of “chips”

A semiconductor supply chain contains designs, wafers, process nodes, packaged devices, test capacity, substrates, and qualified suppliers. A fab running below its nameplate output may still be unable to make the exact automotive microcontroller or memory part that a customer needs. A large wafer-start number can coexist with a shortage of packaged or tested components. The unit of analysis must therefore be the qualified product and process, not aggregate capacity.

Cyclicality appears when decisions made at one boundary arrive after conditions have changed at another. Customers increase orders, distributors add inventory, manufacturers expand, and equipment suppliers deliver months or years later. When the end market turns, the same layers unwind in reverse, often faster than new capacity can be cancelled.

Investment arrives in large, slow steps

Fabs require specialized buildings, clean rooms, utilities, process tools, recipes, staff, and qualification runs. Construction and tool installation take time, and the investment is difficult to repurpose for another industry. A company can delay a project, change its equipment mix, or run an existing fab below capacity, but it cannot turn a new leading-edge fab on in response to next quarter's order book.

The delay creates a coordination problem. Several firms may see the same prices and lead times and commit at once. Each decision can be rational for the individual firm while the combined additions exceed future demand. When the capacity arrives, utilization and pricing fall, but the fixed costs and depreciation remain. This is a capital cycle, intensified by product-specific qualification and by the fact that one node cannot automatically replace another.

The U.S. Government Accountability Office describes semiconductor shortages as involving manufacturing capacity, materials, equipment, workforce, and geographic concentration rather than a single missing input. Its review is a policy analysis, not a forecast of every cycle, but it supports the broader point that capacity is distributed across multiple constrained stages. GAO's semiconductor supply-chain report documents those interacting vulnerabilities.

Orders can amplify a modest end-market change

Each tier orders from the tier before it. A chip buyer may order extra units to protect production; a distributor may add safety stock; an assembler may place orders with more than one supplier during an allocation period. The upstream fab observes orders, not final consumption. When the end market softens, cancelled or delayed orders can make the apparent decline much larger than the change in finished-device demand.

This is the bullwhip mechanism, not a claim that every semiconductor order is fictitious. The useful evidence is the relationship between end-market units, distributor inventories, lead times, cancellations, and wafer starts. During the 2020–2022 shortage, companies and governments reported shortages across selected products and stages, while later inventory corrections affected different segments at different times. The episode does not prove a single cause; it illustrates why an order book is not the same observation as consumption.

Leading-edge tools are a shared dependency

At the most advanced logic nodes, extreme ultraviolet lithography is a critical process step. ASML is the commercial supplier of EUV lithography systems, and its own annual report describes the company's dependence on a limited set of specialized suppliers and the difficulty of expanding system output quickly. This is a real equipment concentration, but it is not a claim that all semiconductor capacity passes through EUV: many memory and mature-node products use other lithography and process routes. ASML's 2024 annual report describes EUV systems and the supplier dependencies around them.

The bottleneck changes timing rather than eliminating the cycle. A fab can have capital and demand but wait for tools, optics, chemicals, masks, or trained staff. Conversely, an EUV shipment does not create qualified output until the process is installed, tuned, tested, and connected to a customer design. Capacity is a chain of conditions, not a single machine count.

Technology transitions create choices, not an automatic compulsion

Leading-edge manufacturers invest in new process generations because performance, power, density, and customer roadmaps can make older nodes less competitive for particular products. But the decision is economic and strategic, not a universal physical law. Some firms stay focused on mature or specialized nodes; many products do not need the newest geometry. A company that skips a generation may lose certain opportunities, but it can also avoid a capital burden that its customers will not pay for.

This distinction matters in a downturn. A firm may keep research and qualification work alive while delaying volume capacity. It may redirect equipment to another product if the process and design are compatible, or it may be unable to do so because the materials, masks, software, packaging, or customer approvals are specific. The flexibility must be demonstrated for the product in question.

Why aggregate utilization misleads

Industry utilization is an average across nodes, products, regions, and stages. High utilization at advanced logic does not prove availability of analog chips, power semiconductors, sensors, or automotive-grade components. Low utilization at one fab does not mean the fab can produce a shortage item without a new process qualification.

Financial statements add more boundaries: bookings, backlog, inventory, and revenue can move at different times. Inventory may be raw wafer, work in progress, finished goods, or stock held by a distributor. A release certificate establishes compliance with a defined test; it does not establish field reliability under every use condition.

How to analyze the next cycle

Specify the product first. Identify its node, package, test, substrate, qualification, and customer concentration. Then ask which stage is constrained, how long it takes to add or remove capacity, who finances the inventory and qualification, and whether the reported order signal comes from final demand or precautionary stocking.

Model at least two paths: a demand slowdown before new capacity is complete, and a shortage that persists because the missing stage cannot be substituted. Check whether the company can slow spending, redeploy tools, or preserve key staff without damaging the next process generation. The conclusion should be conditional: the cycle is a property of interacting commitments and information delays, not a clock that predicts its own turning point.

Inside CompanyGraph

CompanyGraph tracks the heavy-investment phase live: companies whose capital spending runs high against operating cash flow relative to industry peers while exceeding depreciation, the statement shadow of capacity being added faster than it wears out.