If you’re new to the publication, I recommend checking out the quick start guide. It’ll tell you what to expect from these articles, and get you excited for what’s to come.

👋 Introduction

Welcome back! If you’ve been reading along, you know my articles have focused on reviewing the semiconductor industry and its role in fostering the AI boom. Over the past three months, we have taken a detailed look at three of the main players in the semiconductor space: Nvidia, TSMC and ASML. As part of the articles I provided my sentiment on the company's future prospects. 

Lucky for us, since each article has been written, all three companies have released their quarterly earnings reports. I reviewed these reports to see if any company presented information that would affect my theses. The timing of the release also lined up nicely with the wrap up of the articles in this series. 

So what did each company have to say?

Let’s dive in! 

8/28 - Nvidia’s earnings call

With the release of my Nvidia article on 7/28, I concluded that it’s a fantastic company that unfortunately is overvalued at the moment. People seemed to be expecting incredible growth forever, which made the company a risky proposition to invest in. In the time that has passed, a lot has happened with Nvidia, including a meaningful earnings call. Let’s start with the good news.

The good news

There’s a lot.

• From the earnings call: $30 billion in revenue, a 100% year-over-year growth. $16 billion in net income, a 200% year-over-year growth. Margins maintained in the mid-50%s. These are insane numbers! A 3-trillion dollar company that has 2-3x growth in revenue/income metrics is hard to fathom. It underscores my disbelief at the magnitude of Nvidia’s success.

• The company did a buyback of 63 million shares for a total of $7 billion, the majority happening in July. The board then approved another $50 billion in buybacks. When a company is buying back shares like Nvidia is, it means they’re very optimistic on future growth. In other words, the people that know the most about the company want to own more of the company. That’s encouraging.

• Adoption of the Nvidia ecosystem continues to be extremely strong. It has spread from the CUDA development platform, to now the Ethernet for AI module (whose revenue has doubled quarter-over-quarter). The latter appears to be a focus area for Nvidia, as they’re expecting significant adoption of the networking tool over the next 18 months. It’s good to see them doubling down on expanding their portfolio to cloud providers.

• They continue to move into the software provider territory with their inference microservices (NIMs) that offer pre-trained models for developers to quickly build AI applications in the cloud. In fact, they recently released their own model called Nemotron, which according to open source benchmarks has outperformed both GPT-4o and Claude-3.5. In my article, I predicted Nvidia’s arc to resemble that of a cloud provider’s (AWS, GCP, etc) more closely than that of Cisco’s. With the expansion of their software offerings, that appears to be the path they’re headed down. I think this’ll be great for their long-term success.

• Demand outside of AI, particularly in the robotics and automotive industries, are slowly but surely growing. I talked about robotics being a potentially interesting expansion opportunity, and we may be seeing the very first signs of that taking place. I believe that market could be even bigger than the AI market, and quicker to demonstrate a true return on investment. Key players, like Boston Dynamics, BYD, Siemens, and Teradyne, have already begun to use Nvidia’s robotics platform to drive industrial digitization.

• Nvidia announced a partnership with Accenture, one of the world’s largest consulting companies, to bring AI to the enterprise. This is a big deal. While startups, tech influencers, and engineers are quick to adopt and evangelize new technology, the process for behemoth enterprises is much slower. Accenture bridges the gap between the two worlds. They’ll bring implementation expertise and robust training to help large corporations integrate AI into their daily lives. This will really bolster the staying power of AI, and prevent it from being a passing fad. That’s the difference between what we see with AI today, and what we saw with crypto, NFTs and the metaverse. Walmart didn’t start paying employees with Bitcoin and holding meetings in the metaverse. But they sure are incorporating new AI technology into the marketing, accounting, and engineering departments.

• Blackwell, the latest GPU chip from Nvidia, was originally beset with production issues and delays. Those appear to have been fully worked out. Nvidia is expected to ship 450,000 Blackwell units in 2024 Q4, for approximately $10 billion in revenue. Customers are lining up and willing to pay anything for the Blackwell chips; demand is expected to be very robust into 2026.

The bad news

Most of the above news (except for a couple of points, like the Accenture partnership and Nemotron model) was released during their earnings call on 8/28. After the positive news, the stock price… went nowhere. If any other company in the world announced what Nvidia did, their stock price could have doubled overnight. Instead, we got the usual “priced in” response from investors.

This reaffirms that too much future growth is already baked into the stock price, and that Nvidia is too obsessed over. When retail investors are feverishly excited about a company, a disconnect begins to grow between the underlying fundamentals of the business and the movement of its stock price. The disconnect is currently massive. Nvidia’s share price moves all over the place, with no regard for underlying changes in the company. And when that’s the case, you better tread carefully as an investor.

10/15 - ASML’s earning call

Things didn’t go well for ASML after this call. As predicted in my article, they forecasted weaker-than-expected future growth, and investors quickly punished the share price. ASML was down almost 25% in 2 days, and it dragged down other companies in the semiconductor industry because of investor nervousness. This time, let’s do bad news first.

The bad news

There was plenty of less-than-stellar news to digest. Sales didn’t have a big rebound to start the second half of 2024, as was initially predicted. In fact, they were basically in line with this quarter last year. That’s not good, because 2023 was considered a transition year. That was the time period when the dip in their revenues and free cash flow began, if you remember from the ASML article. Management had predicted a big bounce back in 2024 H2, but it hasn’t come to fruition yet.

What we saw instead was an approximately $300 million dollar increase in sales (compared to this quarter last year). While an 8% increase might seem good, it’s less impressive when you consider that it equates to only a few extra extreme ultraviolet (EUV) and deep ultraviolet (DUV) machines sold. In fact, ASML sold 112 units during last year’s quarter, and 116 units this quarter. Not the massive rebound we were looking for.

Let’s zoom out and compare the first 9 months of 2024 to the first 9 months of 2023. Last year, ASML sold 325 units. This year? 286. We can definitively say, the upwards correction is non-existent.

Why did ASML’s predictions not come true? It all comes back to the big red flag that I raised in the article: China. The US pressured the Dutch to apply further export controls on ASML. The latest round of controls has gutted the China market for ASML, with the company predicting that it’ll get only 20% of its sales from China moving forward, down from 49% last year.

Meanwhile, ASML made good on its promise to build up its inventories. Based on their cash flow statements, spending on inventory has increased 33%. But I have concerns about increasing inventory at a time when they clearly misjudged both the immediate cash flow and the longer-term recovery of demand. If sales don’t demonstrate an upward inflection by the end of 2024, will ASML reverse course and stop stockpiling inventory? Or will they plow ahead and stretch themselves too thin? Either way, if demand doesn’t start growing again, ASML’s bets will put them in precarious situation.

The last bit of bad news - they gave no updates on the development and testing of the next-gen EUV machines (High NA). I would have liked to know how that’s going, but I can understand why that’s a back burner item if the current EUV machines aren’t selling as expected.

The good news

There’s a little bit of good news in all of this.

It looks like ASML has ended the financing terms they were providing to customers last year. From my article, I talked about how ASML was doing this for customers who couldn’t afford to pay hundreds of millions of dollars up front for the lithography machines. The company’s payment terms affected their cash flow, which is why we saw the sudden decrease in 2023. That appears to have turned a corner. Based on financial statements, it looks like they’ve ended further financing, and customers are making payments on what they owe. The second part is key - it means the financial position of their customers is improving. Customers will need to have strong balance sheets if ASML hopes to move DUV, EUV and High-NA EUV units at scale. This could be a leading indicator of improvements, though it’s hard to predict how long it’ll take.

Meanwhile, with the share price taking a 25% hit, the forward-looking ratios begin to look more appealing. While the trailing twelve month P/E ratio is 35x (markedly higher than the S&P’s historical average around 20x), the forward P/E is ~23x. It’s possible, though, that future earnings will be revised lower after the company’s guidance, which will in turn make the forward P/E look less attractive.

Let’s also look at PEG. ASML’s historical earnings growth rate has been 23.7%, but for the sake of argument let’s say that we’ll only get 20% growth. That would give us a PEG ratio of 1.15x, which is pretty close to the fair valuation of 1x. Wall Street analysts are projecting 30% earnings growth, even after the bad news from the earnings call, which would put the share price firmly in undervalued territory with a ratio of 0.77x. I think that’s too optimistic. A worst case scenario would involve the future earnings being revised down (meaning a higher forward P/E, 33x) and a lower earnings growth rate (15%). That would give us a PEG of 2.2x, which is still definitely overvalued.

All in all, the cash flow and share price are heading in a direction that could give us a favorable buying opportunity, but we’re not there yet.

10/17 - TSMC’s earnings call

TSMC had a very positive earnings call on Thursday, especially with investors being jittery about semiconductors after ASML’s call. Driven by the good news, TSMC’s stock jumped ~10% in one day and reached an all-time high. So what were the updates that propelled the share price? Let’s start with the good news.

The good news

• The company posted $23.5 billion in revenue, a 12.9% increase from the previous quarter and a 36% increase from the third quarter last year. That’s a trajectory that makes an investor salivate.

• Earnings-per-share came in at $1.95, beating expectations of $1.79. Wall Street was happy, which is always good for the share price.

• TSMC had gross margins of 57.8%, a 3-4% increase from historical quarters. In spite of using new technology, hitting production roadblocks, and trying to meet huge demand, the company was able to become more efficient. That’s absolutely awesome to see. It’s also indicative of really good management.

• An increasing portion of their revenue is coming from the most advanced chip, the 3nm. A year ago, this chip was 6% of revenue; now it’s 20%. The advanced chips (< 7nm) account for 69% of the revenue. It’s encouraging to see the traction and demand for the newest, smallest chip. It leads me to believe that the 2nm demand will be similarly competitive.

• They shipped 3.3 million wafers, an all-time high, which was a 6.8% increase over the previous quarter and a 15% increase over third quarter last year. Every metric is consistent in reflecting serious growth. I love to see it.

• TSMC continues to invest in new fabrication plants (fabs) throughout the world. Their Arizona plant is expected to begin production in early 2025, which is sooner than I thought it’d be contributing. But it’s good news. The biggest threat to TSMC is geopolitical tensions in the region thanks to China. The more fabs that can be operationalized across the globe, the better. In this manner, TSMC hedges its bets and makes itself more resilient.

• Even while investing in the new fabs, TSMC continues to exhibit excellent fiscal discipline. Their asset to liability ratio has actually increased from 2.1x to 2.6x in the last year. That means they’re increasing the cash they have on hand, while managing how much debt they have on their books. It’s hard to increase your asset-to-liability ratio while you have large capital expenditures on new factories, but they’re doing it.

• In spite of the 10% increase in share price, TSMC continues to be “affordable” relative to other companies in its industry. Its forward PE is 23.6x, which isn’t garish. Its PEG is a 1.68, which isn’t obviously undervalued, but it isn’t grossly overvalued either. These premiums indicate to me that I’m looking at a business whose share price and growth have maintained a tight coupling to its fundamentals. That makes me willing to potentially buy more ownership in TSMC.

At every level, TSMC is trending in the right direction. Their technological and manufacturing supremacy is unmatched, their management team is highly disciplined and competent, and their financials show consistent and healthy growth. But what about the bad news? There must be some bad news, right?

The bad news

Truly, not much. At least, nothing new since we last analyzed the company:

• Geopolitical tension in the region has picked up, with China carrying out increasingly aggressive military exercises in the South China Sea. The new fabs across the world will help, but we’re still years away from those reaching the level of operational efficiency and productivity of the main Taiwan plant.

• By platform, 51% of TSMC’s revenue comes from high performance computing (HPC) and 34% comes from smartphone. Theoretically, an AI bubble burst, combined with the continued decrease in smartphone demand, would decimate the company’s revenue. It’s a risk when revenue is so concentrated amongst a few platforms, regions, or customers.

• TSMC didn’t give any information on the production of the next generation of chips, the 2nm. Rumors are that Apple’s iPhone 18, slated for a 2026 release, will be the first widespread deployment of the 2nm chips. That coincides with TSMC’s previously stated plans to begin production of the 2nm chips in 2025. We will likely have to wait for the 2024 annual report to get more information about how successful they’ve been in getting the 2nm production process up and running.

• Similarly, TSMC didn’t talk much about their new wafer packaging technology, CoWoS-L. If you remember from the article, this was a new process they had developed for Nvidia’s upcoming Blackwell chips, but they’d had issues on the production line. While TSMC hasn’t publicly said anything about the fixes they made, Nvidia has addressed the issue. They’ve said that they’re on track to ship hundreds of thousands of Blackwell chips in 2024 Q4, which would only be possible if the CoWoS-L process was operational. Reading between the lines, we can assume that the new technology has been a success. This may not technically be bad news, but since TSMC hasn’t confirmed first-hand, I’m not ready to move it to the good news section.

Some of this “bad news” is, rather than being truly a negative development, simply a known risk or question for which we don’t have new information. We’ll hope to get clarity at the 2024 end of year annual report.

Besides those questions, TSMC’s 10% jump in the stock market on Thursday was well-earned. I’m encouraged by the good news that builds on the research I did for the article. It’s a great business, simple as that.

Before we get to the takeaways, let’s first take a quick detour…

Playing out possible futures

Let’s play a quick game. I’m going to imagine some hypothetical situations, and try to predict how each scenario would affect the three companies. Normally, I do this exercise by myself so I can develop a mental model of the possible outcomes. It’s helpful to keep track of how many good outcomes are in the scenario list and how many bad outcomes are in the list. You can start to create your own odds machine using these simulations. Here are a few possible futures that have crossed my mind:

1. We need more chips for AI and other high-performance compute (HPC) use cases; we need these chips to be more powerful and more cost effective.

The most bullish case.

• Nvidia has to continue designing new chip architectures and expanding their ecosystem offerings to enable developer productivity.

• TSMC prints record numbers of wafers at record speed at fabs all across the globe.

• ASML operationalizes the High-NA EUV lithography machine to make the design and manufacture a reality.

Everyone wins.

2. Frontier AI models (e.g. Google’s Gemini, OpenAI’s GPT, etc) converge asymptotically and the demand for cutting-edge chips is saturated.

The medium bullish case. What I mean by “converge asymptotically” is that the rate of improvement in the AI models slows down dramatically. All the biggest players on the field are putting out a product that is more or less similar. Innovation resembles the last few generations of the iPhone (minor improvements, slight tweaks) rather than the first few generations (massive changes, disruptive technology).

AI is still huge, but it’s no longer an arms race.

• Nvidia continues to design chips, but they can no longer charge astronomical amounts for their chips. Other chip designers catch up over time as the design process becomes commoditized. Nvidia’s revenue shifts to other parts of their business model, such as their CUDA platform, Ethernet for AI, and other modules. These offerings are high margin and highly profitable, but not quite the same as the glory days of the GPU arms race.

• TSMC doesn’t have a months-long or years-long backlog of demand for wafers, but they continue to print millions of wafers per month to satisfy customers. Revenue balances out between high-performance compute (HPC), smartphones, IoT, robotics, automotive, and gaming. This leads to a more stable growth trajectory.

• ASML can eventually deliver the High-NA EUV lithography machines, which helps set the stage for 2nm chips and beyond. But without the arms race, foundries take their time incorporating the new machines. Sales are slow and only a few customers in the world are interested in spending nearly half a billion dollars on a single machine.

Nvidia probably has a strong correction, but then grows steadily over time once it’s out of the spotlight. TSMC doesn’t miss a beat. ASML acquits itself well in the engineering department, but never becomes a massive compounder in the markets.

3. AI is a bubble.

The most bear case.

• Nvidia immediately loses its cult of obsessive retail investors. People are selling on the way down, trying desperately to unload their shares. Additionally, with the AI bubble burst, demand from the HPC sector evaporates. Nvidia’s revenue, earnings and cash flow correspondingly drop. Nvidia has to begin competing for market share (similar to the cloud providers with each other). Recovery is slow.

• TSMC’s revenue suffers a huge blow. Without uncapped AI spend, the chip demand from the HPC sector is much lower. TSMC slows down its wafer production to prevent a supply glut. Their financials take a short- or medium-term hit, but the next tech phenomenon (whether it be robotics, IoT, or automotive) takes hold eventually. Soon after that, TSMC is back to printing chips and minting cash.

• ASML’s customers (with the possible exception of TSMC) suffer from cash flow issues because of the upstream revenue impacts of the AI bubble bursting. They can’t buy many (or any) High NA EUV machines. ASML eventually begins moving units, but very slowly.

ASML and Nvidia are equal losers in this case. TSMC has a potential way out if the right macro tailwinds occur, but not without taking a beating first.

4. China invades Taiwan.

The darkest timeline case ¹ .

• Nvidia is in huge trouble, because its main producer of chips might not exist anymore.

• TSMC is in huge trouble, because it might not exist anymore.

• ASML is in huge trouble, because its biggest customer might not exist anymore.

We’re all in huge trouble. If this scenario manifests, we’ll have plenty to worry about before we have time to think about investment strategies.

Takeaways

Now for the final takeaways! In this section, I’m going to synthesize the research from my articles and from the earnings calls. I’ll distill it down into a concise explanation of how I’ll be managing my position for each company.

Remember, my takeaways are based on my own financial strategy - you shouldn’t make investment decisions based on my strategy. Use this information, do your own research, and then develop a strategy that works for you.

Nvidia

Nvidia is a great business that continues to demonstrate superb growth and forward thinking. Even before the AI boom, they’ve always been ahead of the curve when it comes to identifying incoming trends. It’s what allowed them to benefit from gaming, cryptocurrency, and now AI. It’s also what gives me confidence that they’ll continue to adapt and innovate ahead of the next wave of trends.

Unfortunately, I find that it’s too obsessed over at the moment. The investor frenzy surrounding Nvidia causes price swings that are completely unrelated to changes in the business. Since I published the article on 7/28, the price dropped from $128 to $99, then went back up to $130, down to $102, before coming up to $137 now. That’s a roller coaster, and you couldn’t possibly identify when the earnings were announced just by looking at those price swings.

From a financial perspective, the company isn’t actually that overvalued. Projections show an annual earnings growth rate of nearly 40%. That’s not unreasonable when you consider the growth trajectory the company looks to be on. Looking out to the end of 2025, Nvidia has a forward PE of 34x and a PEG of 0.85x. As the PEG is in undervalued territory, one could make the argument that now is as good a time as any to buy Nvidia.

But I won’t be the person making that argument. For me, the volatility and attention is too great right now. One of my favorite concepts in investing is that there are no “called strikes”. What that means is, you aren’t limited in how many misses you can have. Even if you choose not to invest in 10 companies that all quadruple in value, you still get to keep swinging. You can find the 11th company and invest in that one. There doesn’t need to be any pressure that you have to get into this one. That’s how I feel about Nvidia at the moment. Sure, it may triple in value again, and if that happens, then so be it. No called strikes.

NVDA TLDR; I’m not selling and not buying. Keeping my position.

TSMC

I went into my article on TSMC with high hopes. Then, the research I did unearthed positive indicators that drove those hopes even higher.

I then took a position in the company, and went into this earnings call with high hopes. The call revealed positive news that drove those hopes higher once again.

TSMC is the whole package. Excellent technology, a huge moat, superb management, and rock-solid financials. They’re doing everything right, which is why I have less to say about this company compared to Nvidia and ASML.

Moving forward, I plan to keep monitoring the share price and buying when opportunities present themselves. At this stage of my research, I’m less concerned with closely monitoring the product or their financials, because I have strong conviction that things are running smoothly. Instead, I’m focusing more on share price and valuation, so I can add to my position when market volatility creates a buying bargain.

Unless the fundamentals of the business change unexpectedly, I’d like to stick with this winner for as long as possible.

TSMC TLDR; I’ll buy when opportunities arise. I’ll keep growing my position.

ASML

ASML is a company that is solving some seriously hard problems. Even with my engineering background, I have a tough time wrapping my head around some of their accomplishments. Warren Buffett and Charlie Munger once famously said that they’re “in the business of solving easy problems”. That’s what we want as investors - a company whose business model is extremely easy to conceptualize and whose moat is impenetrable. ASML has the latter, but certainly not the former. It faces major headwinds:

• What their revenue growth looks like with the increasingly severe China export controls

• Weakening demand from foundries for new lithography machines

• The viability of High NA, Hyper NA and future innovations for EUV lithography

While my long-term view remains unchanged - their lithography machines will be needed and growth will be robust on a 10-year timeline - I don’t feel confident about buying ASML right now. Until these problems are definitively in the rearview, I’ll simply be observing the company from afar. Even if the share price drops into undervalued territory, I would need to see the rock solid business fundamentals before changing my mind.

All that being said, from a technology perspective ASML is one of the coolest companies I’ve ever looked at. The engineer in me is fascinated to see what they cook up next.

ASML TLDR; I’ll be watching from the sidelines.

Semiconductor Farewell, and What’s Next

That brings us to the end of the series of articles on the semiconductor industry² .

It’s time to move on to the next series of companies. So what are we doing next?

I thought about a lot of options. There are so many promising companies that merit a deep dive.

But there’s one industry that has had my attention for a while.

The AI boom and the subsequent development of new products has caused energy consumption to skyrocket. The average ChatGPT query uses 10x the electricity that a single Google search does. Training and deploying each iteration of an LLM uses a massive amount of electricity. Many industry experts have predicted that we’ll be in an energy crisis by 2029.

How will we stay ahead of this escalating consumption? We’ll need new sources of energy, whether it be on-shore fracking, solar power, or wind power ³ .

There’s one other option. One that I’m most excited about, and has been making a comeback lately, in terms of corporate and public sentiment.

Nuclear energy.

Nuclear has a complicated history, littered with catastrophes that make people extremely wary of the technology. But things have changed. The technology has improved drastically, and I believe that many short- and long-term advancements for humanity will be powered by nuclear.

There’s so much to learn about its history, its future potential, and how we can benefit from the technology - both as humans and as investors. As always, I’ll start with the basics and build up to a comprehensive understanding. You won’t need any preexisting knowledge to enjoy the ride.

It’s going to be an awesome next series! Stay tuned!

Agree? Disagree? Would love to hear your thoughts - leave a comment.

1  The darkest timeline.

2  If you want to continue learning about other companies in the space, here are some names you can look into: Synopsis, Applied Materials, AMD, Broadcom, Intel, and Samsung.

3  Or desert power.