What We’re Reading (Week Ending 02 April 2023) - 02 Apr 2023
Reading helps us learn about the world and it is a really important aspect of investing. The legendary Charlie Munger even goes so far as to say that “I don’t think you can get to be a really good investor over a broad range without doing a massive amount of reading.” We (the co-founders of Compounder Fund) read widely across a range of topics, including investing, business, technology, and the world in general. We want to regularly share the best articles we’ve come across recently. Here they are (for the week ending 02 April 2023):
1. Pausing AI Developments Isn’t Enough. We Need to Shut it All Down – Eliezer Yudkowsky
Many researchers steeped in these issues, including myself, expect that the most likely result of building a superhumanly smart AI, under anything remotely like the current circumstances, is that literally everyone on Earth will die. Not as in “maybe possibly some remote chance,” but as in “that is the obvious thing that would happen.” It’s not that you can’t, in principle, survive creating something much smarter than you; it’s that it would require precision and preparation and new scientific insights, and probably not having AI systems composed of giant inscrutable arrays of fractional numbers.
Without that precision and preparation, the most likely outcome is AI that does not do what we want, and does not care for us nor for sentient life in general. That kind of caring is something that could in principle be imbued into an AI but we are not ready and do not currently know how.
Absent that caring, we get “the AI does not love you, nor does it hate you, and you are made of atoms it can use for something else.”
The likely result of humanity facing down an opposed superhuman intelligence is a total loss. Valid metaphors include “a 10-year-old trying to play chess against Stockfish 15”, “the 11th century trying to fight the 21st century,” and “Australopithecus trying to fight Homo sapiens“.
To visualize a hostile superhuman AI, don’t imagine a lifeless book-smart thinker dwelling inside the internet and sending ill-intentioned emails. Visualize an entire alien civilization, thinking at millions of times human speeds, initially confined to computers—in a world of creatures that are, from its perspective, very stupid and very slow. A sufficiently intelligent AI won’t stay confined to computers for long. In today’s world you can email DNA strings to laboratories that will produce proteins on demand, allowing an AI initially confined to the internet to build artificial life forms or bootstrap straight to postbiological molecular manufacturing.
If somebody builds a too-powerful AI, under present conditions, I expect that every single member of the human species and all biological life on Earth dies shortly thereafter.
There’s no proposed plan for how we could do any such thing and survive. OpenAI’s openly declared intention is to make some future AI do our AI alignment homework. Just hearing that this is the plan ought to be enough to get any sensible person to panic. The other leading AI lab, DeepMind, has no plan at all…
…Trying to get anything right on the first really critical try is an extraordinary ask, in science and in engineering. We are not coming in with anything like the approach that would be required to do it successfully. If we held anything in the nascent field of Artificial General Intelligence to the lesser standards of engineering rigor that apply to a bridge meant to carry a couple of thousand cars, the entire field would be shut down tomorrow.
We are not prepared. We are not on course to be prepared in any reasonable time window. There is no plan. Progress in AI capabilities is running vastly, vastly ahead of progress in AI alignment or even progress in understanding what the hell is going on inside those systems. If we actually do this, we are all going to die.
Many researchers working on these systems think that we’re plunging toward a catastrophe, with more of them daring to say it in private than in public; but they think that they can’t unilaterally stop the forward plunge, that others will go on even if they personally quit their jobs. And so they all think they might as well keep going. This is a stupid state of affairs, and an undignified way for Earth to die, and the rest of humanity ought to step in at this point and help the industry solve its collective action problem.
2. The Dismal Art – James Surowiecki
We live in an age that’s drowning in economic forecasts. Banks, investment firms, government agencies: On a near-daily basis, these institutions are making public predictions about everything from the unemployment rate to GDP growth to where stock prices are headed this year. Big companies, meanwhile, employ sizable planning departments that are supposed to help them peer into the future. And the advent of what’s often called Big Data is only adding to the forecast boom, with the field of “predictive analytics” promising that it can reveal what we’ll click on and what we’ll buy.
At the dawn of the twentieth century, by contrast, none of this was true. While Wall Street has always been home to tipsters and shills, forecasting was at best a nascent art, and even the notion that you could systematically analyze the U.S. economy as a whole would have seemed strange to many. Economics, meanwhile, had only recently established a foothold in the academy (the American Economic Association, for instance, was founded in 1885), and was dominated by Progressive economists whose focus was more on reforming capitalism via smart regulation rather than on macroeconomic questions.
Walter Friedman’s Fortune Tellers is the story of how, over the course of two decades, this all changed. In a series of short biographical narratives of the first men to take up forecasting as a profession, Friedman shows how economic predictions became an integral part of the way businessmen and government officials made decisions, and how the foundations were laid for the kind of sophisticated economic modeling that we now rely on. Friedman, a historian at Harvard Business School, also shows how the advent of forecasting was coupled with (and fed on) a revolution in the way information about the economy was gathered and disseminated. Relative to today, of course, the forecasters Friedman writes about were operating in the dark, burdened with fragmentary data and unreliable numbers. But the work they did, flawed as it was, would eventually make it possible for decision-makers to get a much better picture of how the economy as a whole was doing. And even as it’s easy to see how the forecasts of today are much more rigorous and complex than those of Friedman’s pioneers, that only makes one question seem all the more salient: Why, if forecasting has come so far, did so many people fail to predict the crash of 2008 and the disastrous downturn that followed?…
…So why are we not better at foreseeing the future? One answer is that we actually are better. Companies these days are less likely to get stuck with huge inventories of unsold goods, or to get caught short when demand outstrips supply. The volatility of the business cycle, meanwhile, diminished sharply beginning in the early 1980s, a relative calm that lasted until the crash of 2008. There’s plenty of disagreement about why this happened, but one plausible factor was that policy-makers and businesspeople were doing a better job of forecasting. And it’s also true that policy-makers have gotten better at responding once crises do happen. The response of the Fed to the recent financial crisis, for instance, was not perfect, but it was much better than the response of the Fed to past crises, and it was also instrumental in shortening the recession and boosting the recovery. Similarly, while the 2009 stimulus plan should have been much bigger, it was, by historical standards, a substantial response, and it too helped get the economy growing again.
Even so, it’s impossible to look at the forecasting track record of Wall Street and Washington over the last 15 years and not be dismayed. The Federal Reserve failed to see that a massive housing bubble was inflating, and did nothing to stop it, even as the banking sector was, in effect, betting hundreds of billions of dollars on the fact that the bubble would not burst. And even when things started to fall apart, people did not recognize how bad things were going to get—Fed Chairman Ben Bernanke testified to Congress in 2007 that the problems in housing would be largely confined to the subprime sector, while J.P. Morgan, the day before Lehman Brothers went under, issued a forecast saying that the U.S. economy would grow briskly in 2009…
…The failure of forecasting is also due to the limits of learning from history. The models forecasters use are all built, to one degree or another, on the notion that historical patterns recur, and that the past can be a guide to the future. The problem is that some of the most economically consequential events are precisely those that haven’t happened before. Think of the oil crisis of the 1970s, or the fall of the Soviet Union, or, most important, China’s decision to embrace (in its way) capitalism and open itself to the West. Or think of the housing bubble. Many of the forecasting models that the banks relied on assumed that housing prices could never fall, on a national basis, as steeply as they did, because they had never fallen so steeply before. But of course they had also never risen so steeply before, which made the models effectively useless…
…The second problem that forecasters face today is more subtle, but perhaps no less important: that there may actually be too much information out there. This would, of course, sound absurd to Roger Babson. But the reality is that investors and businesspeople are now constantly assailed by a high-volume clang of market info and economic data…
…The real issue here is one that the economist Oskar Morgenstern identified back in the late 1920s—namely, that economic predictions actually end up shaping the very outcomes they’re trying to predict. And the more predictions you have, the more complex that Möbius strip becomes. In that sense, for all the challenges they faced, men like Babson and Fisher had it easy, since forecasts were few and far between. The real irony of our forecasting boom is that as fortune-tellers proliferate, fortunes become harder to read.
3. Don’t Build the Wrong Kind of AI Business – Ben Parr
All this activity in AI has led to a new wave of AI startups and will lead to many more. There are real opportunities to build unicorns—but carelessly slapping generative AI on top of your business model isn’t one of them.
Many apps built right now will fail to attract customers, investors or both. Many venture capitalists I’ve spoken with are waiting to see which companies take off. Others are afraid of platform risk—what if OpenAI builds a competitor to your product and nips your idea before it’s even had a chance to bud?
There are ways to gird against platform risk in generative AI, and they start with understanding the two categories of AI startups out there right now:
- Category 1: Startups building advanced, complex language or machine-learning models (AI infrastructure)
- Category 2: Startups building applications on top of these platforms (OpenAI’s in particular)…
…Platform risk shouldn’t stop you from building on top of an AI platform. For one thing, unless you never build a mobile app and never use cloud computing, it’s impossible to avoid entirely. For another, platforms like Shopify, the iOS App Store or OpenAI can accelerate a product’s growth. And finally, the technology OpenAI and others have developed is so powerful that it’s almost a crime not to utilize it. Even if you won’t use it, your competitors will.
If you do choose to build on top of someone else’s AI platform, I advise you to follow my golden rule of platforms: Build a product the platform you’re built on is unlikely to build for itself. Users tend to choose products built directly by the brands they trust instead of dealing with the headache of yet another login. If the gamble goes wrong, the platform will eat your customer base…
…Founders can avoid this outcome by building something Google or OpenAI are unlikely to build. What are those things? They are:
- Applications requiring a proprietary, niche data set. AI models can train on all sorts of data to customize their outputs, which makes it possible to differentiate your results from ChatGPT’s. If you make a chatbot and train it with a database ChatGPT can’t access (such as medical data, millions of emails and so on), the result will be a specialized chatbot OpenAI can never duplicate.
- A product focused on a specific vertical or use case. AI tools built to serve people in fields like health, parenting, law and government require specialized data, interfaces, compliance capabilities, integrations and marketing, which large public-facing AI platforms are simply never going to provide.
4. David Einhorn – The Long and Short of Investing – Patrick O’Shaughnessy and David Einhorn
Patrick: [00:16:49] If you think about the history of Greenlight and the way that you manage the portfolio, I’d love to understand any evolution you had in your thinking over the full period of managing the firm. Obviously, you’re extremely well known as, like, an incredible analyst, like, a securities analyst and I think that’s really what you did at the start primarily. I’m sure that’s still what drives a lot of your time in investing and thinking. But how is your thinking on portfolio management, portfolio construction overlaying things like macro bets into the portfolio? Describe how that’s changed over time for you.
David: [00:17:21] It’s actually changed a lot. I learned a tough lesson in 2008 during that financial crisis because we kind of understood what was going on and got short a bunch of the banks and rating agencies and financial stuff because that seemed to be where the profit was concentrated. But it then turned out to have a really big impact on our long book, which didn’t have any of that stuff, but it had other things that were then exposed to the tightening credit conditions and the recession that came.
And I didn’t really process all of that as effectively as I wanted to, or I should have. And in many ways, I thought that 2008 was my worst year. We lost 18%. Other people may be lost twice that or something like that. So everybody was very nice and said, “Oh, you didn’t do so bad.” But considering that we kind of saw it coming, I thought it was a completely unacceptable result.
So I have added more macro thinking into what I’m doing, and I try to take a bigger view of all of the positions relating to the top down as opposed to just the bottom up. And then it’s compounded on the long side of the book, where just in the last couple of years, I’ve had the realization that with some of these stocks, nobody’s ever going to care. Nobody is paying attention, nobody is doing the work, nobody cares what the company says. There’s just nobody home.
So we can’t make money by trying to buy something three months or six months or a year before other long-only investors figure it out because they, either aren’t there, or they don’t have any capital or they’re turning into index funds or whatnot. So we’ve had to reconstruct our long book in a way that is designed, at least in theory, to earn a return based upon just what the companies are able to pay us as opposed to relying on other investors to figure it out…
…Patrick: [00:22:52] I remember in periods like that, in the quantitative world, especially feeling these existential crises, like, after a long period of underperformance, just wondering, “Have I just missed a memo here somewhere? I think I’ve done great work, but obviously, the results are what they are.”
What was the psychology for you personally like during that period of time? What sorts of things were you questioning? Weren’t you questioning? How did you get through it? Like, I’ve lived through that kind of hell. Curious what it was like for you.
David: [00:23:19] It was very, very difficult. We weren’t making money on anything. It’s not like you had some winners and some losers. It’s like everything was a loser. So part of it was you can say, “Well, how stubborn do you want to be?” The only thing we really could have done better would have been like liquidate the whole portfolio and go to cash or something like that.
We weren’t going to do that. We had large amounts of investors who left us and understandably so because they’re here because they want to make good returns, and we weren’t making good returns. So your investors, one by one, leave. Friends say, “Why are you still doing this? You made enough net worth for yourself. Why are you fighting this battle?” And I’m sitting here saying, “Well, what am I doing wrong?” Then you start saying, “Well, what are other people doing?”
People say, well what you’re not doing is, is you’re not doing factor analysis. That was the big thing, I think, in 2018. So we said, okay, well, let’s get the factor analysis people in here. We signed a confidentiality agreement and they analyzed our portfolio and they come back and say, “You’re short the value factor.” And you say, “Really? How is that?” And they come back and tell me that my two biggest shorts are value. And that is because they correlate with how value trades, not because they’re actually value.
So I look at it and go, “Well, these things are, like, 100x earnings. How are they valued?” And it’s like, “Well, we don’t know, but this is what the machines tell us.” And I said, “Well, I can’t do anything with this.” If the problem is that I’m short the value factor when I think that I’m a value fund or value-oriented, this is a problem.
So similarly, somebody said, “Well, what you really need to do is technical analysis.” So I said, “Great, I’m going to give you 10 stocks, five of them I’m long, five of them I’m short. I’m not going to tell you which ones are longs and which ones are short. Tell me what they’re going to do over the next three months. Should I buy them? Should I short them? What should I do?”
And he looks at the charts and maps it all out and gives me his recommendations. And three months later, he was right on exactly five of them and wrong on five of them. I don’t know what you do with this. So the point is I would open to trying to figure out better ways to, like, do what we’re doing. But at the end of the day, this was just going to be an impossible environment for what we were doing…
…Patrick: [00:54:26] I was just studying Markel and some of the history of insurance, and it’s always so interesting how old so many of the insurance companies are. The dominant ones were started pre-1950 or something. What have you learned about, within financial institutions, insurance and reinsurance specifically? Because obviously, that’s a place that you’ve built and studied a lot.
David: [00:54:46] We have a reinsurance company. I’m the Chairman of it, which doesn’t mean I’m the underwriter. I don’t actually write the policies, but I’ve watched our teams battle with this for the last decade and a half. And I have to admit that it’s been far more difficult than I thought.
I think we’ve run into numerous examples, which are essentially analogous to the, “What happens when you don’t repossess the car” type of analysis, and losses have sometimes appeared in places that were never even contemplated in underwriting. And I have found it to be a very, very difficult way to make positive risk-adjusted returns.
I used to think initially, we could figure out the stuff maybe better than other people, so we wrote a concentrated portfolio of things that were mostly proprietary deals where we had the whole deal. And the first two or three times, it worked spectacularly, and that led to a lot of confidence. But ultimately, I don’t think that, that turned out to be a sustainable advantage for the company.
So we’ve had to shift entirely where it’s a much more diversified mix. And even then, we’ve had fewer blowups, but it’s still been a real challenge. Currently, today, management is very, very optimistic that the market has finally gotten good, and so we should make some money for a while, so that would be fantastic if it actually materializes. I’m more in the, “I’ll believe it when I see it” camp, which doesn’t mean I disbelieve them. It’s just that this isn’t the first time and it’s been a far more difficult operation than I imagined it would be when we started it…
…Patrick: [01:06:08] What have you learned about early relationship health? That sounds interesting.
David: [01:06:11] We have a program that we have been funding. It’s really fascinating. And what it essentially shows is if you can create a co-regulation relationship with your parent from a very early age, it helps you adjust to people probably throughout your life.
And what we have found is that it’s very important for mothers and fathers, but more mothers than fathers, without getting myself into too much trouble, to actually just hold their children, physically touch and get used to the smell and so forth. And if you actually do that, you find it very common. You can go through a calming cycle.
And if you can learn to calm your baby and if your baby can learn to be calmed by your parent, it enables them to become regulated in their relationships for a long, long period of time. We’ve funded a whole bunch of research that has essentially proved out over a sustained period of time what we’re saying. And now we’re trying to figure out how to implement this as, like, a standard training for new parents, whether it’s with pediatricians or in the birthing center and so on and so forth…
…Patrick: [01:08:22] David, this has been so much fun. I mean, so many interesting topics. The investing world has changed so much in the time that you’ve been doing this. I really appreciate your time. I ask everybody the same traditional closing question. What’s the kindest thing that anyone’s ever done for you?
David: [01:08:36] That is an awesome question. My third-grade teacher one day, grabbed me by the arm as we were getting ready to go to recess. And she said to me, you’re probably smarter than everybody else in this class, but you’d be better if you didn’t tell them that. And that really stuck with me.
Patrick: [01:08:58] What was her name, if you remember her name, teacher’s name?
David: [01:09:01] Yes, it was Ms. Olson. She called herself the Purple Witch.
Patrick: [01:09:04] Why?
David: [01:09:05] That was just her nickname.
Patrick: [01:09:08] What did that change? How did that change you?
David: [01:09:10] It created a self-awareness that I didn’t previously have. How do I come across to other people and how do you behave in the sandbox. It kind of shook me a little bit, but it was really, really kind of her to point that out, and she did it in a nice way where I was able to hear it. That’s particularly important.
5. The Death of Credit Suisse – Joseph Politano
Credit Suisse had been plagued by high-profile issues for years. It lost billions in the failure of hedge fund Achegos Capital and supply-chain financier Greensill Capital back in 2021, had data on $100B worth of accounts leaked to German newspapers in 2022, was probed by the US House of Representatives for its connections to Russian Oligarchs, and was forced to disclose “material weakness” in its financial reporting controls thanks to a last-minute call from the SEC just last week. 7% of Credit Suisse’s total revenue over the last decade went to penalties and fines, leaving the company with a net loss of $3.4B after taxes. The bank was surrounded by rumors of its impending demise for years, bleeding money and confidence while constantly scraping by through a rolling series of disasters…
…In some ways, Credit Suisse’s demise is unique from the problems that plagued Silicon Valley Bank and Signature Bank—the institution met highly stringent European capital and liquidity standards, had been regularly supervised and stress tested over the preceding years, and had fully hedged their exposure to the interest-rate driven shifts in long-term fixed income securities prices that helped bring down SVB—distinctions that may have bought the Swiss government enough time to arrange the shotgun wedding with UBS. In other ways, their demise was much the same—like SVB, Credit Suisse was forced to watch the slow departure of wealthy customers’ funds turn into a rush for the exits as depositors reportedly withdrew tens of billions of Swiss Francs in the days before UBS’s takeover…
…So what happens in the fallout of CS’s demise? Among all Global Systemically Important Banks, CS and UBS were unique for two things: their cross-jurisdictional exposures, thanks to the outsized prominence of Swiss banking in international finance, and their intra-financial system exposures, thanks to the unique nature of the two banks—in short, both Credit Suisse and UBS had prominent relationships with non-Swiss customers and had deep ties to other parts of the global financial system. The risks inherent to those exposures are partly why Swiss regulators decided to force a sale of Credit Suisse before it could collapse, but even a more orderly resolution under UBS could still pose risks to the broader financial system.
Regardless of the potential for direct contagion, the demise of Credit Suisse is likely to shake confidence in other lenders, especially in Europe. In particular, prices for European Additional Tier 1 (AT1) Capital Bonds—debt instruments that usually convert to stock if the bank encounters stress and falls below predetermined capital ratios—have fallen dramatically over the last few weeks. Credit Suisse’s AT1 holders were given nothing in the UBS takeover despite the fact that shareholders got a small payout—which is exactly how Credit Suisse’s specific AT1s were designed, but is highly unusual among the broader AT1 market and not something many investors had evidently appreciated. Other monetary authorities—including the European Central Bank, Bank of England, and Monetary Authority of Singapore—rushed to state that shareholders would absorb losses before AT1 holders under their bank resolution frameworks, but it hasn’t yet been enough to rebuild sentiment for the assets. On net, that will make it harder for European banks to raise money precisely when they may need it most.
6. UBS Got Credit Suisse for Almost Nothing – Matt Levine
After the 2008 financial crisis, European banks issued a lot of what are called “additional tier 1 capital securities,” or “contingent convertibles,” or AT1s or CoCos. The way an AT1 works is like this:
- It is a bond, has a fixed face amount, and pays regular interest.
- It is perpetual — the bank never has to pay it back — but the bank can pay it back after five years, and generally does.
- If the bank’s common equity tier 1 capital ratio — a measure of its regulatory capital — falls below 7%, then the AT1 is written down to zero: It never needs to be paid back; it just goes away completely.
This — a “7% trigger permanent write-down AT1” — is not the only way for an AT1 to work, though it is the way that Credit Suisse’s AT1s worked. Some AT1s have different triggers. Some AT1s convert into common stock when the trigger is hit, instead of being written down to zero; others are temporarily written down (they stop paying interest) when the trigger is hit, but can bounce back if the equity recovers…
…These securities are, basically, a trick. To investors, they seem like bonds: They pay interest, get paid back in five years, feel pretty safe. To regulators, they seem like equity: If the bank runs into trouble, it can raise capital by zeroing the AT1s. If investors think they are bonds and regulators think they are equity, somebody is wrong. The investors are wrong.
In particular, investors seem to think that AT1s are senior to equity, and that the common stock needs to go to zero before the AT1s suffer any losses. But this is not quite right. You can tell because the whole point of the AT1s is that they go to zero if the common equity tier 1 capital ratio falls below 7%. Like, imagine a bank:
- It has $1 billion of assets (also $1 billion of regulatory risk-weighted assets).
- It has $100 million of common equity (also $100 million of regulatory common equity tier 1 capital).
- It has a 10% CET1 capital ratio.
- It also has $50 million of AT1s with a 7% write-down trigger, and $850 million of more senior liabilities.
This bank runs into trouble and the value of its assets falls to $950 million. What happens? Well, under the very straightforward terms of the AT1s — not some weird fine print in the back of the prospectus, but right in the name “7% CET1 trigger write-down AT1” — this is what happens:
- It has $950 million of assets and $50 million of common equity, for a CET1 ratio of 5.3%.
- This is below 7%, so the AT1s are triggered and written down to zero.
- Now it has $950 million of assets, $850 million of liabilities, and thus $100 million of shareholders’ equity.
- Now it has a CET1 ratio of 10.5%: The writedown of the AT1s has restored the bank’s equity capital ratios.
This, again, is very explicitly the whole thing that the AT1 is supposed to do, this is its main function, this is the AT1 working exactly as advertised. But notice that in this simple example the bank has $950 million of assets, $850 million of liabilities and $100 million of shareholders’ equity. This means that the common stock still has value. The common shareholders still own shares worth $100 million, even as the AT1s are now permanently worth zero.
The AT1s are junior to the common stock. Not all the time, and there are scenarios (instant descent into bankruptcy) where the AT1s get paid ahead of the common. But the most basic function of the AT1 is to go to zero while the bank is a going concern with positive equity value, meaning that its function is to go to zero before the common stock does.
Credit Suisse has issued a bunch of AT1s over the years; as of last week it had about CHF 16 billion outstanding. Here is a prospectus for one of them, a $2 billion US dollar 7.5% AT1 issued in 2018. “7.500 per cent. Perpetual Tier 1 Contingent Write-down Capital Notes,” they are called…
…In UBS’s deal to buy Credit Suisse, shareholders are getting something (about CHF 3 billion worth of Credit Suisse shares) and Credit Suisse’s AT1 holders are getting nothing: The Credit Suisse AT1 securities are getting zeroed…
…People are very angry about this… I’m sorry but I do not understand this position! The point of this AT1 is that if the bank has too little equity (but not zero!), the AT1 gets zeroed to rebuild equity! That’s why Credit Suisse issued it, it’s why regulators wanted it, and it would be weird not to use it here.
Oh, fine, I understand the position a little. The position is “bonds are senior to stock.” The AT1s are bonds, so people bought them expecting them to get paid ahead of the stock in every scenario. They ignored the fact that it was crystal clear from the terms of the AT1 contract and even from the name that there were scenarios where the stock would have value and the AT1s would get zeroed, because they had the simple heuristic that bonds are always senior to stock.
That’s the trick! The trick of the AT1s — the reason that banks and regulators like them — is that they are equity, and they say they are equity, and they are totally clear and transparent about how they work, but investors assume that they are bonds. You go to investors and say “would you like to buy a bond that goes to zero before the common stock does” and the investors say “sure I’d love to buy a bond, that could never go to zero before the common stock does,” and the bank benefits from the misunderstanding.
7. I Saw the Face of God in a Semiconductor Factory – Virginia Heffernan
By revenue, TSMC is the largest semiconductor company in the world. In 2020 it quietly joined the world’s 10 most valuable companies. It’s now bigger than Meta and Exxon. The company also has the world’s biggest logic chip manufacturing capacity and produces, by one analysis, a staggering 92 percent of the world’s most avant-garde chips—the ones inside the nuclear weapons, planes, submarines, and hypersonic missiles on which the international balance of hard power is predicated.
Perhaps more to the point, TSMC makes a third of all the world’s silicon chips, notably the ones in iPhones and Macs. Every six months, just one of TSMC’s 13 foundries—the redoubtable Fab 18 in Tainan—carves and etches a quintillion transistors for Apple. In the form of these miniature masterpieces, which sit atop microchips, the semiconductor industry churns out more objects in a year than have ever been produced in all the other factories in all the other industries in the history of the world…
…Employees at TSMC are paid well by Taiwan’s standards. A starting salary for an engineer is the equivalent of some $5,400 per month, where rent for a Hsinchu one-bedroom is about $450. But they don’t swan around in leather and overbuilt Bezos bodies like American tech hotshots. I ask Michael Kramer, a gracious member of the company’s public relations office whose pleasant slept-in style suggests an underpaid math teacher, about company perks. To recruit the world’s best engineering talent, huge companies typically lay it on thick. So what’s TSMC got? Sabbaticals for self-exploration, aromatherapy rooms? Kramer tells me that employees get a 10 percent discount at Burger King. Ten percent. Perhaps people come to work at TSMC just to work at TSMC…
…Two qualities, Mark Liu tells me, set the TSMC scientists apart: curiosity and stamina. Religion, to my surprise, is also common. “Every scientist must believe in God,” Liu says…
…During the pandemic lockdown, TSMC started to use intensive augmented reality for meetings to coordinate these processes, rounding up its far-flung partners in a virtual shared space. Their avatars worked symbolically shoulder to shoulder, all of them wearing commercially produced AR goggles that allowed each participant to see what the others saw and troubleshoot in real time. TSMC was so pleased with the efficiency of AR for this purpose that it has stepped up its use since 2020. I’ve never heard anyone except Mark Zuckerberg so excited about the metaverse.
But this is important: Artificial intelligence and AR still can’t do it all. Though Liu is enthusiastic about the imminence of fabs run entirely by software, there is no “lights-out” fab yet, no fab that functions without human eyes and their dependence on light in the visible range. For now, 20,000 technicians, the rank and file at TSMC who make up one-third of the workforce, monitor every step of the atomic construction cycle. Systems engineers and materials researchers, on a bruising round-the-clock schedule, are roused from bed to fix infinitesimal glitches in chips. Some percentage of chips still don’t make it, and, though AI does most of the rescue, it’s still up to humans to foresee and solve the hardest problems in the quest to expand the yield. Liu tells me that spotting nano-defects on a chip is like spotting a half-dollar on the moon from your backyard.
Beginning in 2021, hundreds of American engineers came to train at TSMC, in anticipation of having to run a TSMC subsidiary fab in Arizona that is slated to start production year. The group apprenticeship was evidently rocky. Competing rumors about the culture clash now circulate on social media and Glassdoor. American engineers have called TSMC a “sweatshop,” while TSMC engineers retort that Americans are “babies” who are mentally unequipped to run a state-of-the-art fab. Others have even proposed, absent evidence, that Americans will steal TSMC secrets and give them to Intel, which is also opening a vast run of new fabs in the US.
In spite of the fact that he himself trained as an engineer at MIT and Stanford, Morris Chang, who founded TSMC in 1987, has long maintained that American engineers are less curious and fierce than their counterparts in Taiwan. At a think-tank forum in Taipei in 2021, Chang shrugged off competition from Intel, declaring, “No one in the United States is as dedicated to their work as in Taiwan.” …
…I put a parting question to Lin: How in the world do you remain undaunted by all these extraordinary problems in nanotechnology? Lin laughs. “Well, we just have to solve them,” he says. “That is the TSMC spirit.”
Disclaimer: None of the information or analysis presented is intended to form the basis for any offer or recommendation. We currently have a vested interest in Alphabet (parent of Google), Apple, Markel, Shopify, and TSMC. Holdings are subject to change at any time.
