Semiconductor capex will grow by 35% in 2021 and 15% in 2022, with the biggest cuts coming from memory companies, which will drop by 19%. SK Hynix’s capex will drop by 50%, and Micron’s capex will drop by 42%. Samsung’s capex will only increase by 5% in 2022, but will remain roughly the same in 2023. Foundries will cut capex by 11% in 2023, led by TSMC with a 12% cut. Among major IDM vendors, Intel plans to cut prices by 19%. TI, ST and Infineon will increase capital expenditures in 2023 against the trend.
Companies that slash capital expenditures are typically associated with the PC and smartphone markets, which are set to slump in 2023.
IDC forecast in June that PC shipments would drop 14% in 2023 and smartphone shipments would drop 3.2%. The downturn in PCs has largely affected Intel and memory companies. The weakness in smartphones mainly affects TSMC (Apple and Qualcomm are two of its largest customers) and memory companies. IDMs (TI, ST and Infineon) with increased capex in 2023 are more closely linked to the automotive and industrial markets, which are still healthy.
In 2023, the top three spenders (Samsung, TSMC, and Intel) will account for approximately 60% of total semiconductor capex.
Years of high growth in semiconductor capex tend to be years of peak growth in each cycle of the semiconductor market.
The chart below shows the annual change in semiconductor capex (green bar on the left scale) and the annual change in the semiconductor market (blue line on the right scale).
Since 1984, every significant peak in semiconductor market growth (20% or more) has been matched by a significant peak in capital spending growth.
In almost all cases, a significant slowdown or decline in the semiconductor market within a year of the peak leads to a decline in capital expenditures within a year or two of the peak. The 1988 peak was an exception, and capex did not fall the following year, but was flat for two years after the peak.
This pattern has exacerbated volatility in the semiconductor market. In good years, companies aggressively increase capital expenditures to increase production. When the boom crashes, companies cut capital expenditures.
This pattern often leads to excess capacity that follows boom times. This excess capacity could lead to lower prices and further exacerbate the market downturn.
A more logical approach would be to steadily increase capex each year based on long-term capacity needs.
However, this approach can be difficult to convince shareholders. Strong capex growth in good years is usually supported by shareholders. But sustained capex growth in a weak year will not.
Since 1980, semiconductor capex has accounted for an average of 23 percent of the semiconductor market. However, that percentage ranged from 12% to 34% per year, and from 18% to 29% on a five-year average.
The 5-year average shows a cyclical trend. The first 5-year average peak was in 1985 at 28%. The semiconductor market fell 17% in 1985, the largest decline on record at the time. The five-year average rate then declined for nine consecutive years.
The average eventually returned to its peak of 29% in 2000. In 2001, the market experienced its worst decline ever, at 32%. The 5-year average has since declined for 12 consecutive years, reaching a low of 18% in 2012.
Since then, the average has been rising and will reach 27% by 2022. Based on the 2023 forecast, Semiconductor Intelligence expects this average to increase to 29% in 2023.
2023 will be another major downturn for the semiconductor market. Semiconductor Intelligence predicts a 15% drop.
Others forecast a drop as low as 20%. Could this be the start of another decline in capex relative to the market?
History suggests this will be the likely outcome. A severe downturn in the semiconductor industry tends to scare companies into slowing capital expenditures.
The factors behind capital expenditure decisions are complex. Because fabs currently take two to three years to build, companies must forecast capacity needs for years to come.
Foundries account for about 30% of total capex. Foundries must plan their fabs based on estimates of their customers’ capacity needs over the next few years.
The cost of building a new large fab can run as high as $10 billion or more, making it a risky proposition. However, based on past trends, capital expenditures in the industry are likely to be lower than those in the semiconductor market in the coming years.