Deep Dive: $ASML

Grab a cup of coffee for this one. Arguably, this is the most in-depth research on ASML you can find anywhere :)

The chip in the phone you’re reading this on is very likely made with machines from a single Dutch company. So is every advanced chip in the world.

This company is ASML.

ASML is the only company that can manufacture EUV lithography machines. Each of these sells for ~€230m, and without them, no AI chips can be manufactured. Even at this price, ASML’s customers can’t get machines fast enough, and the company can't meet demand through 2028.

ASML's journey is one of the best business stories, and it nearly didn’t happen.

ASML started working on EUV in 1997. 15 years later, despite the best efforts of some of the smartest engineers in the world, the technology was still years behind schedule.

That year, the smartphone boom drove a sharp rise in chip demand, and ASML's largest customers decided they couldn't wait any longer. Intel, TSMC and Samsung bought 23% of ASML for €3.85bn and committed another €1.38bn to fund its research.

Think about how unusual it is - multi-billion-dollar companies don’t usually fund their suppliers' research. They shop around for alternatives. But Intel, TSMC and Samsung all knew there was no alternative to ASML.

Even with funding and technology support from some of the world's most advanced tech companies, the first EUV-made chips didn’t ship until 2019. It took nearly 20 years and ~$9bn in R&D before EUV chips went into production.

But that persistence paid off. Today, the most advanced chips in the world, including the ones that run AI models, are printed on ASML’s EUV machines. No other company has brought one to market.

To put the demand in context, ASML is booked out till 2028, and customers are asking for allocations in 2029 and 2030!

Welcome to Rebound Capital. If you are new here, we conduct in-depth research on beaten-down stocks and study companies that have made successful comebacks. Subscribe for free and join 26,700 other investors so you don’t miss our next briefing.

Here is a quick order in which we will analyze the business:

Let’s dig in.

An EUV machine has close to 100,000 parts. The system is so large that it ships in 3 cargo planes. This huge machine prints patterns too small to see.

The finest patterns or lines on circuits today are ~13nm wide. The latest high-NA EUV machines can even print patterns ~8nm wide. For comparison, a human hair is ~80,000 nm in width!

One thing to note is that the industry’s node names don’t describe actual feature sizes. Nothing on TSMC’s 2nm node is 2nm wide. Transistor gates sit 48nm apart and the closest wires 26nm apart. The names are generation labels, and every chipmaker uses them that way.

Lithography enables these small patterns on silicon. The key is that the smaller a circuit is, the more of them you can print on a wafer. This directly influences a chip's computing power and memory capacity.

Nvidia’s Blackwell AI chip, made on an older TSMC process, already holds 208bn transistors across 2 dies.

How it works: A lithography machine works like a projector. Light is sent through a stencil (using this analogy to keep it simple) of the chip’s design, called the mask. The light then hits a silicon wafer coated with a light-sensitive material. Patterns are created in the wafer due to the material reacting with the light. This process is then repeated multiple times layer by layer (up to ~100 times). Also, each layer must line up with the one below within a few nanometers.

To draw smaller patterns, you need shorter-wavelength light. The older DUV technology used light with a wavelength of ~193 nanometers. EUV machines use light with a wavelength of just ~13.5 nanometers. This is what makes EUV so hard.

There is no lamp or laser that gives off EUV light. So the machine has to generate its own. To do this, the machine uses molten tin. The machine melts Tin, then lasers (Trumpf is a specialized supplier for these lasers) hit 50,000 tin droplets a second, each about a third the width of a human hair. This superheats the tin droplets to temperatures many times higher than the Sun's surface. This creates a tin cloud/gas that emits EUV light. Of course, now you have to deal with the molten tin and clean it away.

The next challenge is that almost everything absorbs EUV light, including air and glass. So the light travels through a vacuum, and the machine can't use lenses. Instead, it uses mirrors made by Zeiss. Each mirror has more than 100 layers, each only a few nanometers thick, and is smoother than an atom's width. If one mirror were the size of Germany, its biggest bump would be ~0.1 mm high.

The mirrors are able to reflect a maximum of ~70% of the light. On top of that, light reflects off of upto 11 mirrors before reaching the wafer resulting in only ~2% of the light reaching the destination. This requires a powerful light source to compensate for the intensity loss.

Then there is motion. The wafer must keep moving so the correct patterns are drawn. Since this printing happens at the atomic level, the wafer sits on a stage that starts and stops within a fraction of a second, and lands within a few nanometers of where it should be every time. Sensors check the wafer’s position up to 20,000 times a second. Any misplacement, even at the atomic level, is caught.

This is only a glimpse of the complexity. EUV stacks one extremely difficult process on top of another, making it hard to compete. The capital intensity of the R&D process, and the effort required to create another Lithography competitor, makes it very likely that no other company in the Western world will compete with ASML in the future. Only China may eventually build its own EUV, but those machines will be locked out of the Western world.

We see no threat to ASML’s monopoly.

ASML sells 2 kinds of lithography machines:

EUV systems: Only ASML sells these systems. The chips they print run AI systems. The standard EUV machine (called low-NA) sells for ~€230m. The latest machine, called the High-NA, can print even finer patterns. It costs ~€350m and weighs ~165 tons!

EUV system sales are estimated to account for a high-30s % of ASML’s 2026 sales.

DUV systems: These older machines print most of the layers on every chip, including AI chips, and all of the chips in cars, appliances and most electronics.

The most advanced DUV machines pass light through a thin layer of water to print sharper patterns, and ASML sells ~90% of them. Together with wafer-measuring and inspection tools, DUV sales are estimated to account for ~35% of 2026 sales.

Installed Base Management: This is ASML’s service business. This section generates revenue from maintenance, upgrades, spare parts, and other customer services. One key element here is the upgrades business. Upgrades let customers get more out of a machine they already own (for example, improving an EUV machine's efficiency). In today’s environment, where new machines are booked out for years, customers pay for upgrades as they would for new capacity.

This business is expected to grow 50% in 2026 and reach >25% of sales.

In 2025, China accounted for ~33% of ASML’s sales, and that share is expected to fall to ~20% in 2026.

We think ASML probably would be on any top 5 list of companies with the widest moat in the equity world.

This is a key risk many investors cite. Many see this as the biggest threat to ASML’s monopoly. We will not brush aside these concerns, but from what we can see, China is some way away from building a comparable EUV machine.

Even if China develops something comparable, the Western world won't use it. China will use it domestically to meet its own EUV demand, and as of today, no EUV machines are sold in China because of export bans.

The real risk is how fast China can stop buying DUV machines from ASML. Or how many years will it take for ASML’s China revenue to go to zero? In our base case, this happens in the next 5-10 years. ASML's revenue gains from the AI revolution will overshadow this loss.

Currently, China does have a state-owned company, Shanghai Aishengna, making DUV machines for the 28nm node. This is where ASML was about 15 years ago. Chinese commentators themselves say it lags ASML by about four generations. Reports say ~5 of these machines will ship this year and ~20 in 2027. ASML will ship ~130 immersion machines in 2026 alone. China’s machines still depend on imported parts, too.

An unverified 2025 story claims a Chinese firm reportedly took apart an older ASML DUV machine to reverse engineer it. They tried to break it apart and reverse-engineer its features but couldn't put it back together, so they had to call ASML’s engineers. The story is unconfirmed (it comes from a single US commentator), but it is a fun reminder of how hard these machines are to copy.

The biggest risk to any AI investment today is timing. We are in the third year of the AI supercycle, and the question is whether it is nearing its end. Capex digestion could stall spending, or even cause it to fall, over the next few years. The bear cases are well known, from low returns on investment to outright capex overspending.

Here is what we see in the overall AI trade. The world’s largest and best-run firms say demand is voracious, and they are betting hundreds of billions of dollars on it. Use cases are proliferating (Meta’s Muse has found product-market fit for consumer agents), and open-source models are so good that companies can unlock significant productivity while spending far less than they did with frontier models. This bodes very well for AI infrastructure. The frontier labs may be in trouble. But that’s good for overall AI adoption.

Having seen how well AI is doing in math, investment research, coding, and many other fields, we at RC are willing to bet we are in the middle of the AI revolution.

Now we will present the demand signals for ASML’s business. All signs point to strong revenue and profit growth in the coming years:

ASML’s sales follow what its customers spend on new fabs. So we built our own estimate of what its 5 biggest customers will earn and spend out to 2030.

Here is what we expect ASML’s 5 biggest customers to earn and spend ($B):

ASML’s shares are up ~2x since our June 2025 report. Yet the stock trades at ~25x next year’s earnings today, below the ~28x it traded at back when we first wrote about it.

Earnings estimates rose faster than the share price. This is what we like most about owning a business like ASML. The value it creates can outrun even a big stock move.

Next, we dig into why the market is giving us this opportunity and why we think the market’s forecasts are too low. Then we present our Excel model and ASML’s intrinsic value estimates.