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

Hello readers! We’re picking back up with ASML, a Netherlands-based equipment manufacturing company that supplies TSMC and other chip foundries with the machines needed to print ever-smaller transistors on the latest GPUs. Last week we started with an analysis of the industry and its history:

• Understanding the industry

What did we learn? ASML builds a lithography machine that uses extreme ultraviolet light (EUV), a feat of engineering that many scientists and engineers thought was impossible. The EUV machine is required for foundries to be able to print 7 nm and smaller chips, and is therefore widely credited with having saved Moore’s Law. In fact, many argue that the present-day AI boom is only made possible because of ASML’s EUV machines.

How did ASML achieve the impossible? That’s what we’re going to find out today.

Let’s dive in!

Understanding the product

What were the challenges in creating EUV lithography machines?

Let’s talk about the requirements for an EUV machine to be considered a success. The Rayleigh Criterion says that the minimum size of a wafer’s transistor is:

• directly proportional to the wavelength of the lithography machine

• inversely proportional to the size of the numerical aperture on the lithography machine (more on this later)

In order to print the smallest chips ever, EUV needed to have a much smaller wavelength than DUV machines. Remember from the last article that DUV used 193 nm wavelength light. How big was the step to EUV?

13.5 nm wavelength light. Over 14x smaller than the previous iteration. It’s invisible to the human eye, and is absorbed by air. It’s 5 DNA strands wide.

In order for this to work, there were problems that needed to be overcome:

1. The EUV light needed a high conversion efficiency. In other words, the machine needed to produce enough light that could be collected and used for printing, without having absurd power requirements. Previous attempts at EUV had a very low efficiency, around 1%.

2. There needed to be an effective debris collection system within the machine. During creation of the UV light, debris was an unavoidable side effect. This had to be filtered out somehow to maintain the chip’s integrity.

3. DUV used projection lenses to focus the laser down from the reticle size to the actual transistor size. That wouldn’t work for 13.5 nm wavelength light - it would simply be absorbed by the glass lens.

4. Because the EUV light would also be absorbed by air, the machine required a vacuum for the light transmission.

This is why it took billions of dollars of R&D and close to 20 years before the EUV lithography machine was commercially viable. The anticipated rollout of the machine was delayed many times, until finally ASML cracked the code.

How did ASML solve those challenges?

What follows is a series of incredible facts that will make you appreciate how unlikely it is that this machine even exists. Here’s what ASML developed over the course of their two decades of R&D:

1. The double laser solution. To create EUV light, ASML developed a system that fired a laser at a drop of molten tin. The reaction would release EUV light, which would then be captured and used for printing. To solve the low power output problem, ASML designed a double laser solution. The first shot, or prepulse, would hit the suspended molten tin droplet and turn it into a concave sheet. It had to hit perfectly, otherwise the second shot would miss or rebound or damage the machine. Then the second shot, or main pulse, would hit the concave sheet and release the EUV light. The plasma released from this reaction is hotter and denser than the plasma from lightning, but falls slightly short of the sun’s core¹. As I said, literally out of this world. They’re making EUV light that doesn’t exist on this planet. This double laser solution solved the power output problems, allowing them to produce 125 wafers an hour (the amount requested by their customers)² .

   

2. The most precise lasers ever. In order to hit the molten tin droplet twice, the laser needed to be unbelievably precise. No ordinary laser would do. ASML led development of a custom laser for this job. The result: a laser, if placed on the moon, could hit a penny on the Earth’s surface. Every. Single. Time. That level of accuracy is not superfluous, either - it’s required to hit the infinitesimally small tin droplet.

3. The flattest mirrors ever. Instead of projection lenses, ASML used mirrors to reflect the EUV light down through the reticle onto the chip. Even the slightest imperfection would cause the EUV light to be absorbed or refracted, thereby destroying the entire process. How flat is this mirror? If the mirror was enlarged to be the size of the Netherlands, the largest bump on the surface would be less than 1 millimeter tall. For a mirror that’s only a few feet in diameter, that’s as close to perfectly flat as possible³ .

   

4. Hydrogen gas to handle debris. The gas is dual-purposed: it catches the debris and prevents it from ruining the chip. It also automatically etches the mirrors clean and keeps them from developing imperfections because of the debris. Nicely done.

5. The clean room. ASML’s work happens in a “clean room”, with air that’s 10,000x cleaner than outside air. I don’t know what that feels like, but it sounds amazing. More importantly, it maintains the integrity of the components of the machine and allows them to prep the vacuum for the EUV light transmission.

   

   Engineers in the clean room.

To tie all these bits together, ASML’s EUV lithography machine:

• fires a laser (which is accurate to within a penny from the moon to the Earth’s surface)…

• … at a suspended droplet of molten tin, hitting it bullseye twice…

• … and capturing enough energy to create EUV light that is invisible to the human eye, absorbed by air, and 5 DNA strands wide…

• … reflecting that EUV light off of the flattest, most perfect mirrors known to humankind…

• … to print transistors that are less than 1/20,000th the thickness of a human hair…

• … in a room that is purified to be 10,000x cleaner than air.

There’s one more detail I forgot to mention. It’s doing this to 50,000 droplets per second! Each one of these steps requires precision to an unfathomable degree, and yet they all keep working perfectly billions of times a day.

Even a few minutes of mis-operation can cost ASML’s customers millions of dollars in lost revenue. And yet this machine never misses! For hours, days, months at time - it just chugs along and keeps Moore’s Law alive.

Holy crap. This is so damn impressive. ASML accomplished this through top tier engineering and a ton of time and money. You also have to give them credit for maintaining confidence in this idea through 2 decades of R&D setbacks, while the rest of the world dismissed EUV as impossible. When you’re right, you’re right.

What was the result of this breakthrough?

TSMC was the first foundry to use EUV light to successfully print the 7 nm chip. After that demonstration, every foundry was lining up to get their hands on a machine. ASML couldn’t produce these machines fast enough to satisfy demand.

In totality, ASML has delivered 140 EUV lithography machines, each costing up to $200 million. Each finished machine requires 40 freight containers, 20 trucks, and 3 Boeing 747s to be transported to the customer. There are over 100,000 parts in the final product. That’s both incredibly complex and incredibly lucrative.

Because of the cost of these machines, not everyone could afford it. GlobalFoundries, once a powerhouse, bowed out of building advanced chips because they didn’t have the capital to afford hundred million dollar machines. On the flip side, TSMC, Samsung and Intel make up around 84% of ASML’s business.

ASML is a monopoly when it comes to EUV lithography, and accordingly they can set their prices however they wish. Foundries can’t object unless they’re no longer interested in printing advanced chips.

ASML’s breakthrough provided them with a lot of power and influence on the global stage.

How do they maintain the moat around their EUV technology?

No one has been able to replicate ASML’s EUV technology for a number of reasons:

1. It required long term investment and R&D. Remember it took ASML billions of dollars and almost 20 years to commercialize the technology. During this time, other equipment manufacturers were investing time and money in alternative solutions. Intel, as a foundry, pushed its suppliers to invest in quad-patterning, penta-patterning, and so on (which never worked)⁴ . Now, equipment manufacturers might be working on EUV, but it could take them 10-20 years to get it right. But even then…

2. It’s a problem of “extreme technological complexity”. There’s no guarantee that competitors will solve the problem if they just pour enough money and time into development. The last section should have given you an appreciation for how delicate and precise the process is. To solve the EUV problem, competitors will need the smartest people in the world in a field of very specialized knowledge. Unfortunately for them, almost all of those people work at ASML. And why wouldn’t they? They want to work on the cutting edge technology with the best of the best. It’s a self-fulfilling prophecy.

3. Complex, unique deals with suppliers. For decades, ASML has been purchasing subsidiaries to protect their supply chain. Often times in their investor letters they talk about bolstering their “integrated supply chain”. Amongst the many suppliers they’ve acquired, a couple standouts are Carl Zeiss SMT (makes the flattest mirrors) and Trumpf (makes the lasers). These are critical parts of the EUV operation, and other competitors don’t have access to these companies at all. In total, ASML has over 5,000 suppliers in their total base, accounting for over 80% of the components of their machine. They place great emphasis on supply chain risk management.

4. Protected intellectual property. ASML has a vast portfolio of patents, on subjects ranging from light sources and optics to machine design and alignment. Competitors can’t just copy line-for-line.

5. Industry inertia in their favor. Semiconductor manufacturing is a notoriously risk-averse industry. All major players understand that the slightest hiccup can have long ranging impacts to customers and the financial bottom line. So they like to stick with what they know. And what they know is that ASML’s EUV lithography machine works exceptionally well, is battle tested, and is backed by the brightest minds in the world. It would take a lot for foundries to switch to another solution.

What do they have planned next?

At the top of today’s article, we talked about the Rayleigh Criterion, which says that the minimum size of a transistor is:

• directly proportional to the wavelength

• inversely proportional to the size of the numerical aperture

I promised there’d be more on the second bullet later, and later has arrived. ASML’s next innovation won’t be focused on reducing the wavelength of the light, but instead on increasing the size of the numerical aperture (NA).

I won’t get into the gory technical details this time. All that’s important is that we need to increase the value of the NA ratio. Today’s EUV machines have a value of 0.3; the next gen machines, aptly called High-NA EUV, will have a value of 0.55 ⁶ .

The higher NA will create a better resolution on the light, thereby allowing finer features to be printed on the wafer. Notably, this will enable printing the 2 nm chips. It’ll also streamline the process of 5 nm and 3 nm chips. EUV machines are optimized for 7 nm, so foundries use double patterning with EUV to produce 5 nm and 3 nm chips. With High-NA EUV, they’ll be able to produce those smaller chips in a single pass.

Compared to the current EUV generation, High-NA EUV will be able to create about 3x more structures in the same area.

Meanwhile, the UV light might be getting smaller, but the machine size is getting bigger. It’ll take 7 planes, 50 trucks, and 6 months of assembly to deliver one machine to a customer. It’ll cost $380 million ⁸ ! And yet, every single existing customer has already placed orders for the new machine. The first High-NA EUV machine is currently being delivered to Intel, who will work on refining the production process with the new machine ⁵ .

But before we get too excited, there are challenges that must be overcome for High-NA.

What are the current risks to their continued supremacy?

You’d think this company has an ironclad position, but that’s not the case. I was actually surprised to discover some meaningful concerns about their future growth:

1. High-NA is not guaranteed. Although the first machine is being field tested, it’s no guarantee that it’ll work as expected. The NA increase required new mirrors from Zeiss; it required new reticle technology; with the smaller resolution, it reduces throughput. Will ASML be able to iron out issues when operationalizing the new machine? Will they be able to iron out issues soon, or will it take another decade? Will High-NA be worth roughly double the cost of the current-gen EUV machines?

2. Low turnover of machines in the field. Here’s a crazy stat: over 96% of machines ever sold - ever! - are still in operation. These machines simply don’t break. ASML didn’t have a single product recall from 2018-2022. Remember there’s over 100,000 parts per machine - and not a single one broke for 4 years. While this is awesome engineering, it also means that customers don’t have a defined cycle on which they need to replace these machines. If new technology doesn’t mandate that foundries buy new lithography machines, it seems like the old ones will keep working forever.

   

3. Leadership changes introduce instability. Peter Wennink, the CEO since 2013 who oversaw the largest growth in ASML’s history, left in April 2024. Martin van der Brink, the CTO and co-president alongside Wennink, retired in the spring of 2024 also, after almost 40 years at the company. I don’t like it when there are meaningful changes at the top of the org chart, especially at an inflection point in the company’s growth. The new CEO, Christopher Fouquet, has been with the company for 15 years and oversaw the EUV production lines, so it’s not like he’s a stranger they plucked off the street. But will he be able to navigate ASML through their next period of growth? Only time will tell.

4. Canon has developed a new paradigm of lithography. The technology, called nanoimprint lithography (NIL), will do away with EUV light and instead work like a stamp. It presses the chip design into the silicon wafer; think of it like a printing press, but at the nanometer level. This approach means there’s no need for a multimillion dollar laser light system. But there are serious challenges: the reticle needs to be a perfect 1:1 replica of the chip design, since it’s being stamped directly; the reticle has repeatedly been damaged during production, causing thousands of broken chips to be produced until the reticle damage was noticed; the “press” approach created incomplete designs, where parts of the reticle wouldn’t get transferred to the wafer. Canon continues to work on these issues, and it’ll likely be slow growth as NIL tries to be an EUV challenger. But it’s worth keeping an eye on.

5. The catch-22 of needing chips to build their machines. In order to build the EUV lithography machines, ASML needs a lot of advanced machinery. This advanced machinery requires advanced chips made by foundries. So you have this loop where, in order for the foundries to make more chips, they need more lithography machines, which need more chips to be made, which require… you get the idea.

6. Quantum computing will revolutionize the world. It’s a fundamentally different way that computers will operate. It deserves its own deep dive, and it’s on my list to get to. What’s important to know is that producing quantum chips will be a completely different process from our current chips. Different materials and different fabrication methods too. Is quantum computing 5 years away, or 20 years away? I don’t know. But it’ll change things for sure. Who’s to say whether ASML will be the tech leader for those machines?

   

   A quantum computer prototype.

7. Focus on ESG will complicate innovation. In the last few annual investor letters, over 50 of the 200+ pages were dedicated to ESG (environmental, social, and governance) issues. It’s clearly a priority for ASML as a company; they want to get to zero-footprint by 2030. I bring this up not as a political statement in either direction. Instead, I wonder how it’ll affect their ability to innovate. R&D is notoriously wasteful and inefficient. If ASML is constrained by ESG targets, will they be able to create the same breakthroughs that led them to the top?

8. Each advance becomes exponentially more difficult to achieve. This is the thing that concerns me more than anything else. It took close to 20 years and $7 billion to get EUV commercialized, which then produced 7 nm chips. By the time foundries were producing 5 nm and 3 nm chips, they needed to combine EUV with double patterning ⁷ . Now, with 2 nm on the horizon, EUV might not even be able to achieve that. All of that development, which led ASML to the very boundaries of physics, worked for 3 chip sizes and about 10 years? So, in order to keep advancing, ASML has to venture even further into the realm of heretofore unexplored physics? That’s a hell of a bet to make.

You don’t have to take my word for it, either. Here are some direct quotes from their investor letters:

There’s a couple ways to look at these risks. On one hand, I can say that ASML’s monopoly on EUVs means that these risks are footnotes and not a real threat. On the other hand, I can say that EUV is the only thing allowing ASML to hold on in spite of these risks. It depends which perspective you take.

Either way, ASML might be enjoying the fruits of its labors, but there’s no guarantee this’ll continue indefinitely.

How come there was no mention of Applied Materials or LAM?

Before doing my research, I thought those companies were competitors of ASML. They’re not. They both make a lot of tools that are required for wafer production, such as deposition, etching, metrology, and cleaning, but they don’t build lithography machines. Since the focus of ASML was their lithography technology, there wasn’t much to say about Applied Materials or LAM.

On the other hand, Canon and Nikon turned out to be the direct competitors of ASML. Canon is betting on their nanoimprint lithography, but they generally focus on niche areas and therefore own about 5% of the DUV market. Nikon used to be the behemoth in the industry, but was steamrolled by ASML’s success and is now down to 10-15% market share. Most of their market share is in older, more mature nodes since they haven’t been able to ramp up production of cutting edge machines.

I didn’t spend as much time on competitors today simply because ASML is the definition of a global monopoly. No one is playing in the same league as them.

Takeaways

No one can question the engineering excellence of ASML. From a purely technological standpoint, their innovations have changed the world. They’ve redefined physics and manufacturing. They saved Moore’s Law. They made it possible for today’s AI boom to happen. They are a key reason why TSMC and Nvidia have enjoyed so much success.

But the future might not be as rosy. There are speed bumps, potholes, and black holes along the way. This doesn’t diminish the excellence of their products, but it does bear keeping in mind.

Like last week, let’s summarize what we’ve learned so far:

Clearly, there are many good things going for ASML but it isn’t a slam dunk case. What we learn next week with the financials will go a long way in shaping our investment thesis.

Thanks for reading, have a great week, and I’ll see you then!

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

1  It’s good to leave room for improvement.

2  The efficiency of their double laser solution was a bit above 6%. My initial reaction was that this is still really low, but experts all lauded the improvement and were blown away. So, I guess it’s good enough.

3  True level, anyone? https://youtu.be/-MwCJpEuC44. Lambs to the cosmic slaughter! It’s what I kept thinking of, anyway.

4  Some would say that Intel’s backing of the wrong conceptual horse is what compromised their position as a competitive foundry. If Intel had been all-in on EUV, maybe they would have purchased the first machine and would have been where TSMC is today. To their credit, they’re trying to make amends for that. More on that later.

5  This is Intel trying to make amends, and this time, be the leader.

6  The successor to High-NA EUV is already planned. It’s called Hyper-NA EUV, and will have a numerical aperture of 0.7. ASML is targeting 2030, but it’s too theoretical right now to base any investment decisions off of.

7  This is what I was referring to in the last article. It only took one advancement of chip size to need to “stitch” together solutions again, even after all the work that went into developing EUV. I would imagine that was a bit disappointing.

8  This is what I imagine ASML’s pricing meetings are like: https://www.youtube.com/watch?v=EJR1H5tf5wE.