In the past two weeks, we received quite some inquiries about the recent events involving TSMC, Huawei, and, no surprise, the increasing tension between the U.S. and China. We laid out our thought process in the last week's post and we decide to do a special report this week to walk you through what's really going on.

Q1: TSMC announced its plan to build a semiconductor fab in Arizona. Who is TSMC, what does it do, and who are the customers?

• TSMC is the largest foundry in the world. Companies like Apple, Qualcomm design their own processors and turn to TSMC to manufacture them. Apple is the largest customer, followed by Huawei. Overall customers in the US contribute ~60% of the revenue for TSMC, and China ~20%.

Q2: Who does TSMC compete with?

• TSMC is a dominant player in the market with over 50% of market share, and a near-monopoly in the most advanced part of the foundry market. Some of the major competitors are Intel and Globalfoundries in the US, Samsung in South Korea, and SMIC in China. Comparing to TSMC, they all have different issues:

  • Intel has a small foundry operation to make chips for other processor designers. But for the most part it only makes processor for itself. The potential conflict of interest is a big problem.

  • Samsung has a similar conflict of interest problem as Intel and is not as reliable as TSMC as a contract manufacturer.

  • Globalfoundries wasn’t able to keep up with TSMC and gave up the most leading-edge market, its technology is now ~2 generations behind and not good enough for the high-performance processors today.

  • SMIC is still trying to catch up to TSMC but technology-wise it’s 2 generations behind.

  • That leaves TSMC pretty much the only choice for companies designing the most advanced processors for all kinds of different markets. 

Q3: Why is TSMC’s plan to build a factory in the US significant?

• One of the trends we see over the past couple of years is that processors are no longer just designed by traditional semiconductor companies like Intel and Qualcomm. The large tech companies like Google and Amazon are also designing their own processor to better support their core business, especially cloud computing, which is the critical infrastructure of the entire digital economy, and the semiconductor processor is at the heart of it. But these companies still rely heavily on TSMC to actually manufacture the processor. The problem is that all the leading-edge capacity TSMC has is located outside of the US, which is not ideal in terms of security and supply chain resilience, especially in the current geopolitical environment. That is why the US government has been pushing TSMC to bring factories back to the US soil. 

 Q4: Why hasn't TSMC done that before and what changed now?

• The biggest reason TSMC didn’t do it before is the cost and supply chain ecosystem. Its existing fab 18 located in Taiwan has a planned capacity ~100K wafer/month on 5nm technology, and cost ~$17 billion. By comparison, the announced Arizona factory has a total cost of $12 billion but only have 1/5 of the capacity of 20K wafer/month on the same 5nm technology. In addition, Taiwan has a very complete ecosystem for other parts of the manufacturing process like processer assembling and testing. On the earnings call in April, TSMC Chairman Mark Liu said the cost to build a fab in the US does not make sense for them. It is possible the federal or local government offered some incentives to close the gap, but the cost of building this Arizona factory is likely still higher.

• However, we think one of the reasons TSMC decided to do it anyway now is that the US government appears to be pushing hard for a local fab, with or without TSMC’s help. We can think of 3 possible ways the US can do it without TSMC. 1) having Samsung build a similar fab in the US; 2) let Intel spin off the foundry operation and create a pure commercial foundry without conflict of interest; 3) build a brand new foundry operation with the support of TSMC’s existing large customers like Apple and Qualcomm.

• There is no guarantee any of these options will succeed. But if it does, it will be a big risk for TSMC’s current business from US customers. By having a foot in the US, TSMC is trying to get ahead of these potentially negative scenarios and hedge the risk. For the US government, it is also a less risky choice and much more likely to be successful given TSMC’s leadership position and decades of experience operating a foundry business.

Q5: Why does the US, all of a sudden, want a leading-edge fab domestically so bad?

• One easy answer is the consistent trend of moving manufacturing back to the US, but that’s probably not the most important one. The $12 billion investment only creates 1600 direct jobs anyway.

• The more important reason is probably China. China is pushing hard to build a self-sustained semiconductor supply chain. It is currently not competitive, especially for equipment and manufacturing. But it is now one of the top strategic priorities with massive investment into the industry. If China successfully built a local supply chain, not having a leading-edge manufacturing capability domestically will be a big disadvantage for the US. What the US is doing now is on one hand filling in the missing piece (foundry), and on the other hand, keeping China from building a strong supply chain. The latest restriction put on Huawei is one example.

Q6: What is this latest restriction on Huawei?

• Huawei was put on the Entity list last year, so access to US semiconductor components is already limited. The latest restriction is one step further, and more precisely targetting Huawei’s semiconductor design unit HiSilicon. If any foundry in the world makes processors for HiSilicon and uses equipment or software from the US to do so, it is now required to get a license from the US government before it can legally ship anything to Huawei.

Q7: How does that impact Huawei?

• The US companies have a very strong position in the equipment used to make processors. It is almost impossible for any foundries in the world to make processors without any equipment from the US. By enforcing the latest rule, the US can effectively control what kind of chips that Huawei is allowed to use, even if the chips are designed by Huawei itself.

Q8: How does that impact the local semiconductor supply chain in China?

• SMIC is the largest and most advanced semiconductor foundry in China, and Huawei is the largest customer. SMIC still relies heavily on the equipment from US companies to make those chips. If it can’t get a license from the US government, two things could happen, either it loses the largest customer, or risk being punished by the US if it keeps making chips for Huawei. We wouldn’t be surprised to see the US grant licenses to foundries outside of mainland China for certain products but stay more restrictive to the local ones.

Q9: What would be the response from China?

• We haven’t seen any concrete measures yet, but there are reports that some US tech companies might be included in China’s unreliable entities list. It’s a list started in May 2019, a direct response to the US putting Huawei on its own version of the entity list. Not clear what kind of measures may be taken but could include things like limiting market access etc.

Q10: What would be the impact on the semiconductor supply chain?

• China will only try harder to accelerate the localization process and the supply chain may bifurcate in the long term. Companies may be forced to pick sides in this process. We are likely just seeing the beginning of a decade long de-globalization of the semiconductor supply chain.