TSMC (NYSE:TSM) is and will likely remain for years to come a prudent choice for investing in the semiconductor industry, which has been and is expected to remain a growing market over time. As (pure play), TSMC derives its revenue from hundreds of customers, including a dozen large ones.
However, for what would otherwise be a relatively low-risk business with a very digestible valuation (aside perhaps from some geopolitical concerns), a major risk that is becoming increasingly evident is the rise of competition at the cutting edge, which is both the largest and fastest growing part of the market. This could ultimately lead to a loss of market share and pricing power, or even the (almost) complete loss of customers in the event that TSMC is unable to provide them with the technology that enables them to succeed in their own markets (i.e. loss of process leadership).
My previous coverage on the stock for over six months remains relevant (as the thesis described therein continues to unfold):
Apple (AAPL), as its largest customer, has contributed significantly to TSMC’s success over time. Both in terms of revenue (perhaps on the order of 20-25% of total revenue), but also as the primary user of virtually all of TSMC’s latest nodes over the last decade, with whom the companies work closely. According to this sourceApple accounted for $17.5 billion of TSMC’s revenue in 2022.
However, recently, Johny Srouji, head of Apple Silicon, and John Ternus, senior vice president of hardware engineering, said some pretty disturbing things in an interview (emphasis mine).
“We rely on TSMC for a lot of our internal chips… It’s actually very complicated. These transistor technologies are very advanced and complicated, but I boil it down to a few principles: We always want to deliver and build the best chips on the market. the planet for the best products. So this is the North Star. And that means we need access to the best tools, including the best transistor technology.. So it’s extremely important.” – Srouji
Srouji said Apple “needs” the best process technology. The problem here, for TSMC, is that it is on a rapid pace to lose leadership it inherited due to Intel (INTC) stumbles around 10/7nm, back to Intel (around 2nm).
Coincidentally, Intel also launched a foundry business. Some have already speculated that the prepayment received by Intel to accelerate the construction of its 18A manufacturing plant in Arizona even came from Apple.
While I wouldn’t go that far, these developments mean that Apple’s foundry business is open to anyone who can deliver to Apple what it considers its “North Star.” Indeed, later in the interview he confirmed that Apple is not religious about a partnership with TSMC.
8/11 Srouji said that while supply chain diversification is important, its first principle is essential: “where can you get the best technology and where can you get the scale.”
“…if a foundry has this and meets our requirements, we are always exploring – including outside Taiwan.”
Note that even though Intel is new to the foundry business, there certainly shouldn’t be any concerns about scale. As a manufacturer of (primarily) processors, most of Intel’s manufacturing capacity has always been cutting edge, and given its foundry ambitions (and growth in general), Intel has planned and built many new factories, notably in Arizona ($30 billion). ), Ohio ($20 billion initially), Germany ($30 billion initially), and Israel ($25 billion).
As another recent indication of how TSMC appears to be struggling to keep pace in the next era of full-gate transistors, this recent roadmap from TSMC shows how the A10 node is planned for 2030. On the surface, this appears to follow rhythm. Moore’s law, which slows down from a rate of 2 years to 3 years, and indeed this roadmap has gaps of 2 or 3 years between nodes.
However, the goal of a monolithic chip with more than 200B of transistors just seems weak/unambitious, since even on N5 (2020) there are already monolithic chips with almost 100B of transistors. Even noting that the introduction of high-NA lithography will cut the (maximum) size of a monolithic chip in half, it remains to be seen whether TSMC can re-accelerate its shrink rate from the slow rate of N3/N2 (similar to Intel on its 10/7 nm nodes).
For example, if 100B is the goal for N2 (although this is already possible with N3, although TSMC revealed that N2 would be a very minor shrinkage), then A10 with 200B transistors on half the area would require shrinkage of 4x on two nodes. , which is aligned with Moore’s law.
Like many (including Intel), TSMC expected a recovery in the second half. However, third-quarter revenue declined 15% to $17.3 billion. EPS was $1.29, a decline of 25%.
More generally, beyond the Apple risk, the importance of cutting-edge nodes is demonstrated to the extent that nodes of 7nm and below represented 59% of turnover. Although TSMC already has a competitor in this segment with Samsung, increased competition from Intel Foundry Services only adds further pressure, especially given TSMC’s increased margins in recent years.
This was indeed recognized by Pat Gelsinger, who said that TSMC’s margins are so high that there is enough room to reduce TSMC’s prices and still be left with a lucrative business. Note that this contrasts with some popular comments we could have read. (The (high-tech) foundry industry is certainly large enough to support a few players, but it is also so advanced that no additional players can compete, thus ensuring the overall profitability of all competitors.)
So, in a few years, TSMC is poised to lose its (foundry) process leadership and potentially become a more expensive foundry in general. At this point, there doesn’t seem to be much (if any) reason for TSMC to maintain such a large market share, which would result in a loss of pricing power as well as market share. However, contrary to the above, a price war from below should not be expected, since both companies are very focused on achieving high margins.
Looking ahead, even at the high end of the fourth-quarter guidance, revenue would decline by about $1 billion, marking the fourth consecutive quarter of decline. Note that this means that the rise of GAI did not benefit TSMC much, since most of the gross profit dollars went to Nvidia (NVDA), the BOM (bill of materials) for H100 silicon (a few hundred dollars at most assuming on the order of 75 chips per wafer and $15,000 per wafer) being roughly a rounding error from the sale price real (well within 5 figures).
Despite the forward-looking uncertainty regarding TSMC’s business as described, the company is actually very investable, which remains consistent with my previous analysis. This is evidenced by the forward P/E of just 20x, although the price is at the high end of its one-year range.
One can only guess why the stock doesn’t have a higher valuation. It may be the Asia/China effect that has also affected stocks like Alibaba (BABA). Perhaps it is the Taiwan effect, in particular the risk of invasion from China. This may just be actual financial performance as investors wait for a new up cycle.
The risk of literally losing a quarter of one’s business, although as described seems very real, in reality is probably not yet taken into account by investors. Even if it were to happen, like Intel’s transition to Apple silicon, this process would take several years even after startup, with startup itself likely still years away. By then, the rest of the market might have risen so much that TSMC’s finances might remain fairly stable from where they are today.
Overall, although, at least in principle, the risk of TSMC losing its largest customer is real, how to factor this into the valuation is quite tricky since the actual probability is unknown.
Takeaways for investors
Regardless of black swan events, TSMC is an investable stock. At this point, perhaps even more than its new competitor Intel, since the former’s valuation has not yet increased much. As trends such as GAI show, the semiconductor industry remains a long-term growth market, so investing in a leading foundry such as TSMC, which is the undisputed leader in terms of market share with hundreds of customers (majors and minors), would be a prudent choice for investors.
Of course, as we have described, one of the possible black swan events would be the rise of new competitor Intel. In fact, there already appears to be notable initial activity, as evidenced by the prepayment Intel received from an undisclosed 18A customer to expedite construction of the new Arizona manufacturing plants.
The possibilities range from finding an alternative supplier for at least some of their chips to completely losing the business of their largest flagship chips as they move to the best technology on the market. Ultimately, at the cutting edge of technology, the only thing that matters to TSMC’s customers is how their chips compare to those of their competitors. Using a foundry/process that is further along the Moore’s Law curve for this purpose is a decades-proven strategy for outperforming competitors.
So ultimately, as cited, while Apple hasn’t made any official announcement, it has stated that it is not religious about the foundry it uses. The company only aims to create the best possible chips. As TSMC loses its process leadership is increasingly guaranteed (with Intel already supplying the 0.9 18A PDK to its customers), this is a key risk that investors will need to address, at least as a possibility.
Beyond Apple, these competitive dynamics, in general, could result in a loss of market share and/or pricing power, putting downward pressure on revenues and profits, which could counterbalance any general market growth, of which TSMC has been one of the main beneficiaries in recent years (and decades).