TSMC: Losing Leadership

Investment Thesis

TSMC (NYSE:TSM) has gone in damage control mode. The company claims it will maintain leadership, but has nothing to back this up. This seems reminiscent to when around 2017 Intel (INTC) started putting out presentations about its leadership (e.g., how its 10nm had 2x higher density than other 10nm nodes, however, the company neglected that its own 10nm was much later to market, instead competing against 7nm nodes). It is even more reminiscent as my previous analysis had already indicated that TSMC's pace of following Moore's Law at N3 and N2 has almost completely halted, with a pace of shrinking that is even slower than Intel's 5-year cadence from 14nm to 10nm.

Overall, this means TSMC is rapidly losing its leadership position, which goes against the common current consensus as evidenced by the top comment on this recent article. Combined with the deteriorating financials, my thesis remains neutral, as the valuation remains reasonable.

Background

In my previous analysis (TSMC Responds To Intel) I warned that TSM had just reported its last quarter of growth for the foreseeable future.

Q1 Results

TSMC reported Q1 results below the midpoint of its guidance. Gross and operating margins were in line with guidance but marked a sequential decline. Due to flatlining or even negative growth and rising expenses, profitability also decreased, with EPS of 1.31 down meaningfully from 1.82 last quarter.

Technology Symposium: N3, N2

During the earnings call, TSMC was questioned about how it felt about its leadership position, given Intel's claim of regaining leadership in 2025 with 18A. TSMC provided the following (non-)answer:

Let me say that, as usual, we don't comment on our competitors' status, but then we emphasize again on our 3-nanometer and 2-nanometer. Our 3-nanometer is the first in the semiconductor industry to high-volume production. And I believe it is the most advanced semiconductor technology in both PPA and transistor technology.

And for 2-nanometer technology, that was, again, to be the most advanced semiconductor technology in the industry and when we introduce into mass production. And this one, we're fully confident that we will further extend our leadership position well into the future.

This answer reminds me of Intel's response when it first delayed its 10nm from 2016 to 2017. CEO BK at the time said that it would be the first 10nm node in volume production - minimizing the ramp of its competitors. Intel would also continue to maintain that it had favorable PPA (performance, power, and area. Lastly, Intel also said it had the most advanced technology, saying that it was at its third-generation FinFET whereas others were still on their first.

However, the question really was about how TSMC benchmarked this claim, and as is clear from the answer, TSMC did not answer the real question. So let's (try to) answer this question in TSMC's place based on current information, as for example disclosed at its recent tech symposium.

First, in terms of time to market, TSMC is clearly behind. Production in Q4 2025 lags Intel by nearly two years, as Intel has targeted 20A production for early 2024. TSMC is also well over a year behind 18A, which is Intel's 2nm-class foundry offering.

Secondly, in terms of power and performance, Intel has claimed parity with 20A (presumably against N3E), and leadership with 18A, which will deliver +10% performance. TSMC for its part has said that N2 will deliver 10-15% higher performance. This indicates that, within the margin of error, 18A and N2 are equivalently performing nodes, with a slight edge for N2.

Lastly, in terms of area (transistor density), third-party analysis by Angstronomics and WikiChip has indicated that Intel 4 is already quite close to N3. Note that Intel 4 only has a high-performance transistor library, while Intel 3 (the foundry offering) has a full feature set. However, there is no reason (at least a priori) to assume the difference in terms of competitiveness will be different for the other (i.e., high density) libraries. In other words, the expected Intel 3 density of around 200 million per mm2 should be quite close to N3E's 215 million.

What this means in practice is that my forecast remains unchanged. Based on TSMC's disclosure that N2 will improve transistor density by "over 20%", its density will likely remain firmly below 300 million transistors per mm2. On the other hand, 18A was targeting a ~2x increase in density over Intel 3, which extrapolates to a density of 400 million transistors per mm2. A clear victory for Intel's 18A. One risk here, to be sure, is that this information is based on comments by prior CEO Bob Swan. It is not completely inconceivable that targets and specs might have been adjusted since then.

Overall, Intel 18A will be meaningfully earlier to market, it will be roughly on par in terms of power and performance, and it will likely also have a meaningful transistor density leadership. TSMC is losing its leadership rapidly due to moving at a snail's pace from moving to N5 to N3 to N2.

Lastly, TSMC disclosed that it will deliver N2P in 2026 (presumably Q4 2026 assuming it is one year later than N2, implying products will launch in 2027), which will have backside power delivery (BPD). Investors may know that Intel pioneered BPD (which it calls PowerVia), targeted for 20A in 2024. This means TSMC will be nearly three years behind Intel with this feature. This seems reminiscent to the days of tick-tock when Intel was over three years ahead of TSMC with launching major transistor innovations. This further solidifies the claim that Intel is set to overtake TSMC in the next few years.

Valuation

Despite the current revenue and profitability trends, TSMC trades around a 15x P/E multiple, which isn't excessively expensive. The expectation remains that the semiconductor industry is a growing one long-term, which should be a natural tailwind for TSMC.

As such, at first sight, TSMC could be seen a solid and low-risk (given its broad exposure to many top-tier companies such as Nvidia (NVDA) and Apple (AAPL)) investment at a reasonable valuation during this downturn.

However, this ignores the technological trends discussed above.

Investor Takeaway

It is likely that different investors will weigh the different moving pieces with regard to TSMC differently. The general consensus as heard among companies is that the downturn will pass, and semiconductor growth will resume long-term. Given the nature and valuation of TSMC's business, this makes it a fairly low-risk investment.

On the other hand, the foundry landscape is drastically changing. With Intel, there is now a strong third competitor at the leading edge. Moreover, TSMC has been moving at a snail's pace towards N3 (2.75 years) and N3 (3 years). Both nodes also feature a markedly decelerated rate of density improvement.

Hence, as I have argued previously, TSMC is currently moving slower than even Intel during its 3-year delay of 10nm. Meanwhile, Intel is currently moving at arguably the fastest pace in semiconductor history and plans to churn through 4 nodes (including two full, traditional ones that deliver a 2x density improvement) in the next 12 to 18 months. As prime evidence of TSMC losing its leadership, Intel will deliver products with backside power delivery (BPD), a next major innovation, in 2024, whereas it will take TSMC until 2027 to catch up with this feature. By that time, Intel will have moved to 14A.

On the flip side, changes in the market tend to happen slowly (perhaps before gaining steam). The unfolding of the events of TSMC losing its leadership (2025) followed by customers meaningfully moving to IFS as the new leading edge will take years (even in the best case), towards the end of the decade. So while TSMC could continue to perform solid (financially) for the next several years, the stock is not one without worries, as the writing of TSMC losing its (inherited) leadership position is on now firmly the wall.