If there has been a strategy to the Trump administration’s actions on semiconductors, it has been to try to lessen China’s dominance in advanced communication technology by denying China’s national champions access to state-of-the-art semiconductors that are either: manufactured in the US or that contain US intellectual property or that are manufactured by machines that use US intellectual property.

At the same time, however, Trump has substantially weakened the domestic technology industry by denying skilled industry workers entry to the US, based on his white racist ideology. He has also undermined the finances of domestic research universities by encouraging violence against Asian students, thereby discouraging them from coming to the US and enrolling. My point is that there seems to have been no long-term strategy to strengthen US tech and to close the 5G gap with China (in fact, quite the opposite), but rather only the simple-minded notion of doing near-term harm to firms like Huawei.

TSMC has already agreed to send lagging-edge machinery to a new foundry it intends to construct in Arizona and Intel has promised two more. Samsung is also planning to open a combination fab (for itself)/foundry in Texas. Plans for both are under threat, though, by current Trump-inspired legislative proposals in AZ and TX to suppress voting in Democratic areas of those states. Intel and AZ faced a similar situation in 2014, with large companies threatening to leave the state after the legislature passed a bill permitting merchants to refuse to serve LGBTQ customers. Under immense pressure from businesses, which the legislature shrugged off, the governor vetoed it.

How events will play out over the next ten years is hard to say. As an investor, however, one thing that interests is the idea that the semiconductor manufacturing industry has been made substantially less efficient today than it was even a year ago. China must take much more seriously than the lip service it has given to date to the need to develop its own chip manufacturing capabilities. Non-China firms that want to sell to China must now have both their research and the fabs they access located outside the US. The US, in turn, is suddenly aware of how dependent it is on TSMC and demanding to have state-of-the-art manufacturing facilities located domestically.

To me, this implies there’s going to be very strong demand, likely for years, for new semiconductor manufacturing equipment to fill fabs that will be built in both China and the US. Yes, there’ll be excess capacity, and yes the plants won’t be humming along at profit optimizing speed. But that will be a secondary concern, I think. The primary objective will be for plants to be either under control of the China camp or the US–because neither camp is likely to make its most advanced chips available to the other side.

What happens to TSMC, the world’s undisputed technology leader, is unclear, although recent reports say that, given Biden administration incentives, it is planning on opening as many as six new fabs in AZ.

This has never been my favorite area of tech and, I think, has typically not had the zip of the chip companies themselves. But I think circumstances have changed. I’ve been learning on the fly and already have established positions in AMAT, ASML and LRCX. These aren’t recommendations, just disclosure that I have an interest in this area.

About 25 years ago, the semiconductor business began an evolutionary change, one that encouraged separation of semiconductor design from semiconductor manufacturing.

Four reasons for this:

—cost. As semiconductor designs became more complex, the cost of a semiconductor fabrication plant (fab) began to breach the $1 billion level. This became a substantial barrier to entry, not only because of the capital needed for construction but also because its efficient operation required it spew out $3 – $4 billion of annual output. What startup had that kind of money or that kind of sales.

—sharing not a thing. No one wants to share manufacturing space with a rival who operates a fab, for fear its designs will be pirated. At the very least, the subcontractor is supporting a rival’s operations. Apple’s withdrawal from Samsung’s fabs is a prominent example.

—work in a bureaucracy? There’s a tendency for companies to lose their entrepreneurial spirit as they get larger and begin to be controlled by professional managers who don’t have product knowledge. Think: GM, Intel, Microsoft under Ballmer, Disney under Eisner. (An aside: I find that how well I’m treated when I call a company, say I’m a shareholder (i.e., a part owner) and ask for information is a good indicator. The bad ones tend to support only brokerage analysts, who then try to sell info about “our” companies to you and me.)

In any event, entrepreneurs tend to have little interest in office politics.

—alternative solutions. One is the emergence of trusted third-party fab companies, called foundries, the best of which is Taiwan Semiconductor Manufacturing Company (TSMC). A second is the development of Arm Holdings, a UK-based company now owned by Softbank (but, assuming regulatory approval, being sold to Nvidia), which provides extensive software templates for use in semiconductor design.

The prime example today of the older integrated model is Intel (INTC). It remains wedded to an architecture introduced in 1978. Formerly the world leader, it has recently been passed in fab capability by TSMC and in chip performance by smaller rival Advanced Micro Devices (AMD), a TSMC and ARM customer.

INTC’s simpler chips for server farms are now also being supplanted by proprietary designs (ARM + TSMC) by Google, Microsoft and Apple.

where we stand today

To summarize, the most advanced semiconductor chips are being designed by US firms, using UK-based ARM architecture and manufactured by TSMC in Taiwan.

A state-of-the-art fab now costs well over $10 billion. One can be built just about anywhere there’s land, skilled workers, reliable electricity and water. As a practical matter, the decision about location comes down to tax breaks.

Oddly, but on reflection maybe not so (think: Trump Steaks, or his Atlantic City casinos), the Trump administration announced its policy of denying Chinese companies access to US designs, despite there not being a cutting-edge fab located in the US. That has belatedly become a priority. TSMC has agreed to build an older-generation fab in Arizona, Samsung a big fab in Texas. We’ve since seen, however, how unreliable the power grid is in Texas. Probably more important, pending voter suppression legislation in both states threatens the viability of these projects.

Investment implications tomorrow.

I’ve just updated my Keeping Score page for the month of March and 1Q21.