How long will this adjustment cycle last?

Semiconductor products are largely made of silicon; the semiconductor industry is heavily built on numbers and graphs.

Moore's legendary paper published in 1965 has five data points; the latest WSTS Blue Book has 1,784 data points i.e. 37 years and 2 months, or 446 months, with four regions (Americas, Europe, Japan and Asia Pacific), not counting those subtotals, totals, moving averages, etc.

Not sure if I will call it as another form of numerology, this vast amount of data with a long period of consistent measurements is just like a gold mine to statisticians, especially those like me extremely enthusiastic in time series.

The World Semiconductor Trade Statistics offers a wide range of data, analysis and services. The Blue Books tabulate the actual numbers while the Green Books are the graphical representations. For those of us in the industry, these are like monthly scorecards, painfully waiting with full anxiety. How did we perform against our competition?

Three month moving averages (3 MMA's) are included in the Blue Book by default. Given the diversified nature of the semiconductor industry with different holiday and seasonal factors around the globe, I believe it is more appropriate to use the twelve month moving averages (12 MMA's).

I classify a 'Down' cycle when the year-on-year (YoY) growth i.e. the blue line below is peaked from the red marker, heading downwards and crossing the zero growth line. An 'Up' cycle is when the YoY growth bottoms out from the green marker, heading upwards and crossing the zero growth line.

We do not know yet how long this Down cycle will last, i.e. the last red marker on the right hand side. We can, however, observe some 'patterns' if the history can provide some insights.

Out of the previous eight 'Up' cycles, half of them were peaked in September, not to mention that there are twelve months a year. And, out of the previous eight 'Down' cycles, half of them lasted 18-19 months.

For the September phenomenon, I think it can be explained by the consumer spending behaviours such as the Black Friday and year-end shoppings.

For the 18-19 month observation, I am not sure if it is just a coincident to the Moore's Laws, and well, technically, the original prediction by Gordon Moore is indeed that the number of transistors on a chip will double roughly every two years, not 18 months, with a minimal rise in cost. I believe it can also be explained by the notorious cyclical nature of the semiconductor business, swinging up and down, going through the over- and under-invested decisions.

Adding 18 months to March 2022, will the current 'Down' cycle be done by September this year? I do not think so. It will most likely take a much longer period of time. Why?

The following graph compares the global semiconductor revenues and the capital expenditures. From 2000 to 2023, the capital expenditure is on the average 23.1% of the global revenue. As clearly shown on this graph, the investment rate has been above this average of 23.1% for four years, heading into the fifth year in a row. If we add up these 'upsides', it is roughly 20% or one more year of investment.

Not a scientifically proven way of projection, however, my guess is as good as yours. My guesstimate is that this 'Down' cycle will end by the third quarter (Q3) in 2024. Tighten our belts, folks.