Our original TSMC thesis summary, posted on September 24th, 2018 was this:
“Quickly becoming “everyone’s semiconductor fab” with a top-shelf client list at the heart of the AI-5G-IoT revolution. The biggest risk is the US-China trade war, but competition is limited from both US and Chinese firms. Effects should be muted.”
Our valuation at that point was around $60 per share. As of writing this, TSMC trades close to $80. We’ve let it run for a while because, frankly, the sudden price bump in the last 2 months was a pleasant surprise. We were in no hurry to sell TSMC.
Why did this happen? In this “Covid-Quarter”, TSMC’s Q2 revenues grew by 29% from this time last year. Their Gross Margin increased by 10% - from 43% to 53%. That is a big jump. This big uptick in TSMC’s fundamentals was not surprising because many of their famous clients have been growing at a healthy clip. We’re talking about companies like AMD, Qualcomm, Nvidia, Apple, Xilinx and so on. These chipmakers get their chips actually made at TSMC. If they do well, TSMC does well.
We’ll spare you some of the suspense. We won’t sell out of TSMC. You can read about our “sell discipline” here – the gist is that we prefer not to sell out of Global Dominators that have a sustainable competitive advantage. Sure, if facts change, we change our minds. Alternatively, if we have a much better investment idea and no other way to fund it, then we will (reluctantly) sell some of our best companies.
In TSMC’s case, we hope we never have to sell. But now, the question is whether we should trim our position because of the stupendous price appreciation of its stock this summer. That’s what this analysis is about.
Our initial TSMC thesis played out almost perfectly. Their dominance in semiconductor manufacturing for clients who focus on designing specialized chips is still intact. The dominance showed in the company’s results over the last 2 years. Check out how their business has changed.
What do these charts tell us? Smartphones are becoming smarter. One of the obvious drivers is Apple, which started designing its own custom chips – “fabbed” at TSMC. Why did Apple do that? Because off-the-shelf CPUs just don’t cut it anymore. Apple sees a confluence of 2 massive phenomena, which drove it to take matters into its own hands: AI and 5G.
It may sound awfully reductive, but the big windfall for TSMC over the last few years has been…Artificial Intelligence. AI is the reason why computing is shifting away from the CPU. GPUs, FPGAs and ASICs are encroaching upon the CPU’s dominance. We’ve covered this topic in some detail in the following worldview articles:
- Artificial Intelligence is Real
- The Grand Inflection Point
- Is Apple killing Intel?
To keep it short and sweet, here’s what happened: For decades, CPUs were dominated by Intel. Then AMD started competing and encroaching on their territory. This accelerated over the last 2-3 years. In the meantime, Nvidia had cornered the Graphics and Gaming market with their GPUs. Accidentally (and happily) for Nvidia, it turned out that their GPUs outperformed CPUs in the type of logic and computing involved in Machine Learning. The promise of AI suddenly looked achievable to computer scientists. Other specialized chipmakers like Xilinx (the market leader in FPGAs) saw their moment as well, offering an alternative to GPUs for AI processing. But deep pockets like Google and Apple had even grander ideas. They decided to design their own chips, which would be AI or 5G-ready. Through this whole drama, TSMC suddenly realized its founder’s dream: “to become the world’s semiconductor fab”.
Cue in a juicy subplot called ARM. ARM, a British semiconductor company founded in 1990, took on the mighty Intel by offering a totally new CPU architecture. ARM’s RISC-based architecture proved to be much more power-efficient than Intel’s bread-and-butter X86 designs, which are more powerful processors but also more power-hungry (to put it simplistically). ARM really had its moment when smartphones came into being. These mini-computers didn’t need desktop-level “general-purpose” CPUs. They needed something leaner. Their need for power-efficiency ruled out Intel’s X86. The thing is: ARM doesn’t manufacture chips; they simply designed them. This is big. ARM didn’t just offer up an alternative CPU architecture (RISC-based), but it presented a whole new business model. Unlike Intel, which designs and manufactures its chips from start to finish, ARM simply licenses its intellectual property – its chip designs. It’s up to the licensee (like Qualcomm or Apple) to figure out how to actually manufacture them.
This is where TSMC steps in. Their competition is limited because Intel (out of hubris or whatever) refuses to “fab” ARM-CPUs. Today, TSMC manufactures chips for almost everybody that’s using ARM designs, and even for some who use Intel designs (like AMD). The model has been working like clockwork – companies either design or license designs for their specialized semiconductors, and outsource the painstaking process of manufacturing them to TSMC. Nvidia has done it. Qualcomm has done it. Apple has done it. AMD has done it.
AMD is an interesting case. For the first time ever, AMD has beaten Intel in the “node wars”. They’ve outpaced Intel in making smaller, faster chips at the 7-nanometer (7nm) node, thanks to TSMC. Intel, on the other hand, already had manufacturing delays with the 10nm node a couple of years ago, and just last week in their earnings call CEO Bob Swan revealed that they now have issues with the 7nm release. Now, these node-wars are exciting, but their relevance is often exaggerated. Some engineers claim that Intel’s 10nm process is comparable to TSMC’s 7nm process. That may be the case, but the point is that it’s working out great for AMD – they’ve been able to move fast and gain market share.
In a somewhat veiled but stunning revelation, Intel CEO Bob Swan hinted in the company’s last earnings call that they are open to outsourcing their 7nm process! This is stunning because throughout its existence, Intel took immense pride in being the one and only vertically integrated chipmaker. That was their big Moat. But with AMD breathing down their neck, Intel is forced to swallow its ego.
There are 2 major shifts in computing that are creating massive headaches for Intel but simultaneously galvanizing TSMC’s raison d’etre:
- Heterogenous Computing
- The increasing adoption of ARM designs
Heterogenous Computing refers to the “decentralization” of computing from a CPU in a desktop or server to a world in which computing is optimized amongst different devices (Cloud Server, desktop, smartphone, robot etc.) and amongst different kinds of chips – CPUs, GPUs, FPGAs, and ASICs.
Because of the power-efficiency of ARM chips, they’re finding new fans. The latest famous fanboy is Apple. Recently, in their Worldwide Developer Conference, Apple announced that they will transition to custom ARM-based chips even in their Macbooks. Previously, Apple had already furnished iPhones and iPads with custom ARM-based chips. Of course, these chips will be fabbed at TSMC. We discussed this issue in detail, after which we dumped our Intel shares – not because of Intel’s immediate revenue loss but because of what this decision could mean for the future of computing and, consequently, Intel.
The subjective case is strong. But how does all this translate to numbers for TSMC?
Let’s start with how TSMC has performed vs. our expectations:
The question now is: will TSMC keep growing at this pace? For how long? This is where we revert to our practice of Expectations Investing. We answer the question: what do we need to believe? The subjective case, in our view, is strong. The phenomena of Heterogenous Computing and new Chip Architectures are just getting started. The combined tailwind will likely be big over the next few years. It’s hard for us to put some numbers around this view but here are some near-term datapoints – forecasts provided by TSMC’s clients and suppliers:
In a strange year like 2020, that’s quite a statement. Now, if you’ve been a long-time reader of The Buylyst, you know that we don’t care much for quarterly swings in revenue. But all we’re doing here is taking this as one datapoint of evidence that betting on the compounding of data and insights appears to be a pretty good one to make.
Back to Expectations Investing: What revenue growth do we need to believe to buy into TSMC now? We need to work backwards.
TSMC’s current Market Value is $408 billion.
A 30% upside would mean that TSMC’s market value would be about $530 billion.
Normally, we slap on a 20X multiple on Free Cash Flow to evaluate a company. Working backwards, a $508 billion valuation would imply Free Cash Flow of about $26.5 billion. Is this achievable? TSMC’s Free Cash Flow over the last 12 months ending June 30th, 2020 was about $13 billion by our calculations. That’s a big leap to make.
Now, using $26.5 billion in Free Cash Flow as a reference point, let’s back into the Revenue number. Let’s assume that TSMC’s operating cost-structure remains similar to what it is now. Specifically, we mean:
- EBITDA Margins remain roughly where they have been over the last 3 years – about 63%. This is a conservative assumption because we expect some economies of scale and scope as the business expands.
- Capital Expenditure remains roughly at the same level.
Some people may argue that Assumption #2 is too generous – that it should be higher. But we’d like to point out that TSMC has already increased its Capex significantly over the last few years. They’ve anticipated the momentous shift in computing we’re witnessing now and have invested capital in order to harness this tailwind.
It’s a high run-rate: nearly $20 billion. Just for comparison, Intel’s Capex is about $16 billion. Intel, of course, also spends about $13 billion on R&D (which they consider to be an operating expense in the Income Statement). TSMC doesn’t need to spend so much on R&D (it spends about $3 billion), considering it’s not an original designer of chip architecture; Intel is. TSMC’s sport is converting its clients’ designs to reality – it’s about process control and efficiency. While it’s important for TSMC to be on the cutting-edge of technology to maintain its supremacy, $20 billion a year seems like plenty of money to do so.
Back to our “Expectations Investing” estimates: what level of revenue will trickle down to about $26.5 billion in Free Cash Flow? Our safe bet is about $65 billion in revenue. That’s roughly 60% over TSMC’s current annual revenue run-rate. Again, it’s a big leap.
Let’s break this down in terms of “implied revenue growth” and stock price. Here’s how our “Expectations Map” looks:
In order to hold TSMC stock, we need to believe that revenue can grow at least 30% from this point on. Said another way, that translates to about 8% of revenue growth per year for the next 3 years. This, to us is totally believable.
Now, if we were NEW buyers into TSMC, we’d need to believe that revenue can grow by 60% from this point on – in order to expect a 30% upside in the stock. Said another way, that translates to about 20% of revenue growth for the next 3 years – also believable, but less probable.
Has it happened before – has revenue ever grown so rapidly in a short span of time? Yes, TSMC doubled its revenue from 2014 to 2019 – in about 5 years.
So, the bet we would need to make is that TSMC can repeat that performance. We would bet on yes rather than no. Why? For all the factors we discussed above, and more:
- The continuing quest to fight the impending end of Moore’s Law.
- The rise of Heterogenous Computing because of the proliferation of AI.
- The increasing relevance of ARM designs in both smartphones AND desktops.
- The continued growth of Cloud Computing.
- 5G has yet to blossom.
- IoT has yet to blossom, especially in digitized factories.
- Autonomous Vehicles have yet to go mainstream.
In short, there are 2 growth vectors that are massive tailwinds for TSMC:
- The exponential growth in data.
- The exponential growth in insights from that data – i.e AI.
Compared to the last 5 years, we’re willing to bet that the next 5 years will be even more lucrative for TSMC. We’re willing to take a chance on that positive optionality by continuing to hold TSMC at 5%.
TSMC Thesis Busters
TSMC is a posterchild for our preference for Global Dominators. It’s playing the right sport, has high barriers to entry, very little competition, and a sustainable competitive advantage. But no investment is without risks. We see 5 major risks to our TSMC thesis:
- Intel cleans up its act and beats ARM at its own game by making even more power-efficient chips.
- Intel starts fabbing chips for other companies, like TSMC.
- GlobalFoundries comes back from oblivion and starts manufacturing chips for fellow American companies
- Samsung steals away some of TSMCs clients.
- The US-Chine “tech war” worsens.
Among these, the biggest risk is #5. If that happens, the probability of #1-#4 increase significantly. TSMC is Taiwanese, not Chinese. But Xi Jinping’s intentions are far from laisezz-faire. Taiwan could be a pawn in the great US-China trade war, which is increasing about technology leadership.
Of course, if there is any sort of “takeover” from China, American chip-making giants like AMD and Nvidia will look for an American or non-Chinese manufacturer. Cue in GlobalFoundries or Samsung.
But it’s not so simple. Chipmaking is one of the most complicated manufacturing processes in the world. Only a handful of companies can do it. At the moment, most of the chip manufacturing in the world is being done by Intel, TSMC and Samsung. It is in the interest of many American companies to keep TSMC “alive and kicking”. AMD, for example, would be reticent to use rival Intel, should Intel start its own “fab” business. Apple, for example, would be reticent to use Samsung, one of its main rivals (same for Qualcomm).
TSMC also recently opened up an American manufacturing facility in Arizona. It’s a small start but we think that if they can close the cost-gap, they will expand their American operations. The risks of pandemics and trade wars seem temporary, but companies have now accepted that these risks can back any time.
TSMC is in a very complicated business. Consequently, its competition is naturally limited – down to 2 other players in the world. While political and pandemic risks are not controllable, technological superiority and being indispensable to clients is a more controllable outcome. That’s all TSMC can do. Their massive investment - $20 billion a year – in technology leadership suggests that TSMC Management thinks the same way.
In our view, the upside far exceeds the downside. It’s hard to attach a probability to each event. If the facts change, we’ll change our minds. But until then, we’ll be holding on to one of the most prolific Global Dominators in the world today.
We’re holding on to TSMC at an approximate weight of 5%.