In July 2021, widespread shortages made global semiconductor overcapacity look remote. Demand from phones, vehicles, IoT equipment, data centers, and consumer electronics was absorbing output, while advanced-process and mature-node bottlenecks appeared in different parts of the supply chain. Yet “the chip market” is not one capacity pool. A shortage in one process, package, or product can coexist with excess capacity in another.
A 2021 commentary argued that Huawei's reduced chip purchasing had been replaced by demand from other buyers and predicted that overcapacity would remain unlikely for the following decade. Its supporting company results, purchasing data, import values, and foundry-allocation claims belong to 2020–July 2021. The long-range conclusion was an opinion made during a shortage, not a verified description of later years.
What the 2020 purchasing data was used to argue
The source cited Gartner data stating that Huawei's semiconductor purchasing fell 23.5% in 2020 while purchases by the ten largest chip buyers rose 7.3%. It attributed growth to companies including Apple, Samsung, BBK-related brands, and Xiaomi. It also cited China's 2020 integrated-circuit imports at USD 380 billion, 25% above the prior year.
These figures were used to argue that demand shifted rather than disappeared. They do not show identical parts changing hands. One phone vendor may buy a different application processor, radio, memory mix, power solution, or process node from another. Purchase value can also change with product mix, price, inventory, and exchange rates.
Company rebounds were historical snapshots
The article cited Qualcomm's fourth-quarter 2020 revenue rising 62% and net income 165% year on year, and Texas Instruments revenue and net income rising 22% and 45%. These were source-reported period comparisons, not evidence that every US chip company recovered equally or that later growth would persist.
Quarterly results can be affected by acquisition accounting, product cycles, tax, licensing, channel inventory, and the comparison base. A capacity decision should use units, utilization, backlog quality, lead time, customer concentration, and capital plans alongside revenue.
Why one customer rarely determines a whole foundry
The source discussed concern that TSMC could face unused capacity after losing Huawei business, then stated that Apple took its 5 nm capacity and that Intel and AMD demand could tighten the node. Whether each allocation statement was precisely accurate requires dated primary evidence. The broader logic is valid: a qualified foundry can sometimes reallocate capacity when other customers have compatible designs and strong demand.
Reallocation is not instant. Mask sets, design rules, process options, yields, package, test, qualification, and customer schedules differ. A tool or wafer start released by one customer may not satisfy another product's timing or technology.
| Capacity segment | Why supply is not interchangeable | Risk metric |
|---|---|---|
| Leading-edge logic | Few qualified fabs, complex design, expensive masks and packaging | Node-specific starts, yield, advanced-package capacity |
| Mature-node analog and MCU | Special processes, long qualification, embedded memory and high mix | Process-option utilization and qualified alternatives |
| Memory | Cyclical commodity behavior and technology transitions | Bit supply, inventory, price, and capital intensity |
| Assembly and test | Package format, equipment, substrates, and test program differ | Package-specific lead time and substrate availability |
| Automotive-qualified supply | Qualification and change control constrain substitutions | Approved site, package, die revision, and lifecycle |
Why shortage-period forecasts can overshoot
When lead times rise, customers may order earlier, place duplicate demand through several channels, or build safety inventory. Suppliers see a stronger signal and invest. Demand can later normalize while new capacity arrives, producing a correction. Technology transitions may also strand an older line even if aggregate chip demand grows.
IoT and connected vehicles can increase long-term semiconductor content without preventing segment-specific excess. Unit growth, chips per system, die size, yield, price, and process migration determine required wafer capacity. “Exponential demand” is not a capacity model.
A better way to assess capacity risk
- Map the exact die, process, wafer size, fab site, package, substrate, assembly, and test path.
- Separate firm end demand from distributor, channel, and duplicated orders.
- Track lead time, backlog cancellation terms, inventory, utilization, yield, and pricing together.
- Model capital additions with realistic construction, equipment, ramp, and qualification time.
- Build upside and downside cases for product cycles and macroeconomic demand.
- Identify where an alternate node, package, or design can actually be qualified.
- Review geopolitical and customer-concentration risk using dated, verified sources.
Implications for hardware sourcing
A buyer should not assume that a future market correction will make a specific constrained part available. The part may remain single-source, obsolete, or locked to one package even when industry inventories rise. Conversely, purchasing years of inventory at a shortage peak can create excess and aging risk.
From WLconnectivity's interconnect perspective, an electronic redesign also affects connector pinout, power, heat, grounding, and board layout. Semiconductor risk reviews should include the physical interfaces and validation work triggered by any substitution.
A time-bounded conclusion
During 2020–July 2021, reduced purchasing by one major customer did not prevent strong aggregate demand and severe shortages. The source reasonably observed that other buyers absorbed significant supply. It went too far when it treated capacity surplus as broadly unlikely for a future decade.
The defensible conclusion is conditional: semiconductor capacity is slow and segmented, while demand can shift rapidly. Shortage and overcapacity can exist at the same time in different segments. Decisions should be based on the exact manufacturing path and a range of demand scenarios, not a single shortage-era narrative.
