UPDATED: This is an updated version of our post to reflect TSMC’s announcement this morning.
This has been a busy week for the semiconductor industry. This topic has been eating away at us all week starting when the Wall Street Journal published a piece titled “Trump and Chip Makers Including Intel Seek Semiconductor Self-Sufficiency” last weekend. This was followed on Friday by the announcement that Taiwan Semiconductor Manufacturing (TSMC) would build a leading edge fab in Arizona, and invest $12 billion to get the plant up and running by 2024. While not specified in their announcement, there is a belief in the industry that the US government will be subsidizing this to some degree.
In this note, we want to look at what is going on and what the US government seems to be trying to accomplish. Then in a following piece we will examine an alternative approach for semiconductor Juche.
Now we are as patriotic as anyone. We recognize the gift that being born in America provides. We want to see this country prosper, it’s just that we are not sure this idea is actually going to help in any meaningful way or achieve its intended outcome without a whole host of unintended (or intended) consequences.
Part of our objection is that the US has a history of shunning explicit industrial policy, and for good reason. Industrial policy done poorly ends up as a form of cronyism paid for by companies, consumers and taxpayers. Without dipping into partisan waters, this administration has not demonstrated that it can, or even really wants to, follow through on the detailed work needed to pull off such a plan. A few months ago, they floated the idea of nationalizing telecom equipment providers, an idea which has since thankfully sunk out of sight.
The root of the problem with this article is right there in the headline. Intel wants to help bring semiconductor manufacturing back to the US. This is odd as Intel is already the largest manufacturer of semis in the US. It stands to reason that they would like to see the government support manufacturing semis in the US. Or as we put it on Twitter:
The heart of the piece are “letters reviewed” by the authors that show the US government has been in communication with in Intel and Taiwan’s TSMC. TSMC’s intentions are now clear, but Intel’s role in this is not entirely clear. As noted above, they want help, but probably do not want to be seen as asking for a direct hand out from the US government.
On its face, this sounds like a reasonable idea. If the US is the biggest end-consumer of chips, shouldn’t some of that production take place in the US? But here the details really matter. Recall that TSMC is not the first Taiwanese company to seek US government support for building a next generation plant in the US. Foxconn promised something very similar in Wisconsin, but that plan does not appear to be fulfilling early expectations.
Admittedly, TSMC has a better track record than Foxconn and are likely not making this announcement lightly. That being said, there are are a hole host of operational details involved in this which make a big difference in the outcome. Crucially, the amount of government subsidy for the project is not clear. It seems likely that the US government is supporting this somehow, either directly with funds and tax breaks, or indirectly through guaranteed orders.
There are also significant operational details to consider. How will TSMC manage a fairly complicated coordination plan to bring the fab online? Splitting its team ahead of a big ramp-up would be resource taxing even in normal times, let alone now when international travel often requires 14 day self-quarantine. Would the plant be economically feasible? Would it be a Trusted fab that can supply the US Military? How would it that be governed.
There is an important side note here. TSMC has an embedded currency advantage which give them a 20%-30% cost advantage. Here is a pretty remarkable Twitter thread from people who know a lot more about this then we do.
TSMC is the only company capable of producing chips using the leading edge technology at 5nm, and they will almost certainly lead the way in whatever comes next. Rather than use their currency advantage to undercut everyone on price, they have used that advantage to corner the market in expertise. (We are leaving Samsung out of this, long-story short, the same macro-economics apply to them as well, and they are a bit further behind on the technology curve.)
Shifting operations to the US would likely present significant cost adders to TSMC, and for a company that periodically spends more on capex than they generate in revenue, this is not a simple problem.
Mos importantly, the geo-political situation is highly complex. For decades, TSMC has been avoiding pressure from China to open a fab there. Opening a fab in the US will likely cause the company problems in China, which is almost as big a market for TSMC.
When the news first hit, there was a common quick assessment that by opening the fab in the US, TSMC would get a pass to continue producing chips for Huawei, who is one of their largest customers. However, this view was quickly squashed when a few hours later the US government announced new restrictions on supplying Huawei. These are lengthy and we are not legal experts. But our initial understanding is that fabs (i.e. TSMC) are still restricted in what they can provide to Huawei.
The irony in this is that TSMC is a pretty decent company. All of our interactions with them, and their general reputation, has been that of solid engineers and managers trying to run a complicated business. They are now stuck between the rock and the hard place of China, Chinese customers, the US and US customers.
So TSMC opening a plant in the US is complicated and something and we have likely not heard the end of this saga.
And then there is Intel. They may still have a role to play in this, but it is not clear what that would be. Intel has been trying to sell its foundry services to outside customers for close to a decade. They have largely failed to do so. Not only have they signed up essentially no customers, but their reliance on TSMC has probably grown as Intel acquired assets like Infineon’s mobile business, FPGA maker Altera and many others who use TSMC. The foundry business requires a huge amount of customer service and that competency appears to be beyond Intel’s reach.
Building a leading edge foundry is probably one of the most challenging commercial endeavors out there. There are no easy options.
In our next piec, we will examine some alternatives.