Over the past few years, governments everywhere have woken up to the reality that semis are a strategic asset. Most famously, this has sparked a brutal trade war between the US and China, as well as a contest where seemingly every country wants to build domestic fabs. These policy ambitions are set against the hard technical constraints of semiconductor fab costs which can only be found in Taiwan, South Korea (sort of) and the US (sort of, looking at you Intel). As a result, when policy people talk about building domestic fabs, they are really talking about trailing edge capacity – 16nm and older, and in most cases much older. Just how much trailing edge capacity does the world need? And how much does it actually have today?
Compounding matters, many of these policy decisions were made at the height of the supply shortages of ’21 and ’22, which means they come built-in with demand assumptions that do not hold anymore as the supply situation has eased considerably. So just as everyone is dolling out subsidy dollars (or Euros) to build trailing edge plants, the companies that actually have much of this capacity are reducing capex. The table below shows announced analog capex changes for 2023 relative to 2022, implying that these companies see no reason to add much capacity any time soon. And after TSMC’s earnings last week, no one really expects them to increase capex next year.
The cynical among us will write off this disconnect between policy goals and commercial reality as misguided politicians burning through tax dollars. There may be an element of this, but we think the situation is actually more complicated. A big motivation behind these policy ambitions is the idea of self-reliance, which given the state of semis today is not unreasonable.
We looked up fab capacity data from Tech Insights. This data paints a nuanced picture. First, trailing edge capacity (16nm and above) still accounts for 90% of all semi manufacturing. Aside from data center, phones and PCs, the vast majority of chips we interact with daily are made on very mature processes. This is still an important, robust market. Second, not surprisingly, Taiwan and South Korea still account for almost 40% of trailing edge capacity. This means that geopolitical problems are a threat beyond leading edge as well. Third, Europe is critically behind accounting for only 6% of trailing edge capacity, and recall they have nothing at the leading edge.
Finally, and probably most importantly, the US accounts for 14% of trailing edge capacity, while China comes in at 16%. China has more capacity than the US. Of course, there are a lot of caveats to that statement. About half of China’s fab capacity is foreign owned, and so somewhat dependent on support from other countries. This data is a bit stale, and the gap is fairly small, so this is not a commanding lead by any metric. And of course, the US has Intel and that TSMC plant in Arizona, which means it has a big lead at the leading edge.
That being said, Chinese companies have been adding fab capacity at a torrid clip for years, with many more fabs under construction than anywhere else in the world. This matters along two axis. First, commercially there is a big wave of trailing edge capacity cresting out on the horizon. We mentioned this on the latest episode of the Circuit, as a vague anxiety looming in the back of our heads. We imagine this is a factor in the capex plans mentioned above. Secondly, from a geopolitical angle, this paints China as being a stronger position than some may think. We imagine there is some war planner in Beijing who envisions leveling TSMC and Samsung foundries on Day One, and thus leveling the field for their semis capabilities. We do not mean to be alarmist, we do think that this scenario will not come to pass. But all of this does explain the motivation behind so many government’s pursing semis manufacturing.