There Is No AI Race
Without This Machine

You can spend your time arguing about which AI chip company wins the next three years. Nvidia vs. AMD. Custom silicon vs. merchant. TSMC vs. Samsung. Those are fine debates. But they all share one assumption: that someone will keep making the machines that allow any of these chips to exist at all. That someone is ASML. There is no other someone.

Every logic chip fabricated below 7nm — every H100, every Blackwell, every Apple A18, every AMD EPYC, every Snapdragon — was manufactured using EUV lithography equipment made exclusively by a company headquartered in Veldhoven, Netherlands with 43,000 employees and a stock that trades on NASDAQ. Intel tried to build a competing EUV platform. It gave up and placed its leading-edge future in ASML's hands. China has spent billions on domestic EUV development. It has produced nothing that works at scale. ASML filed over 100,000 patents across the technology and owns a 24.9% stake in Carl Zeiss SMT — its exclusive optical supplier. The moat isn't a moat. It's a continent.

The stock bottomed at $683 in October 2024 after an accidental early release of Q3 2024 preliminary results showed orders of €2.6 billion — about half of what the street expected. The market panicked. ASML clarified immediately that the miss was due to customer order timing, not a change in demand. They were right. Q4 2024 came in with a record €7.1 billion in new orders. The stock has more than doubled since.

Today at $1,606, you're paying roughly 57x trailing earnings for the machine at the center of the global semiconductor supply chain. That's not cheap. But "monopoly that every chipmaker on Earth cannot function without" has historically supported a significant quality premium, and there is no scenario in which the AI infrastructure buildout — which requires leading-edge logic and HBM at unprecedented scale — does not require more ASML machines.

The October 2024 low is important to understand because it was caused by a disclosure accident, not a business accident. ASML's IR team posted preliminary Q3 2024 results a day early. Orders: €2.6 billion, vs. consensus of ~€5.6 billion. Within hours, the stock dropped over 20%. ASML held an emergency call, explained the order timing dynamic, and noted that €36+ billion in backlog hadn't changed. The market didn't immediately care.

Then Q4 2024 earnings dropped. Orders: €7.1 billion, the largest quarterly booking in company history. The stock retraced sharply, and the narrative flipped from "demand collapse" to "the AI capex cycle is real and every fab is ordering." What followed was a 19-month grind higher from $683 to today's $1,606 — through tariff scares in April 2025, multiple bouts of macro uncertainty, and a Q1 2026 orders miss that barely registered given the backlog context.

The 200-day MA at $1,186 tells you where the stock was roughly seven months ago. The 50-day at $1,446 is where it was six weeks ago. The stock has outrun both meaningfully — the kind of distance that usually invites near-term consolidation but doesn't change the long-term picture.

ASML reported Q1 2026 on April 15, 2026. Revenue came in at €8.77 billion — above the guided range of €7.5–8.0 billion. Gross margin landed at 53.0%, toward the top of the 51–53% band. Net income was €2.76 billion, EPS €7.15 diluted. There is no ambiguity in any of these numbers: Q1 was a clean beat.

The conversation was dominated by orders: €3.94 billion of new intake against a street consensus of €5–6 billion. Sell-side coverage briefly lost its mind. Then people remembered that ASML's backlog was ~€36 billion entering the quarter — about 1.1× annual revenue — and that quarterly order intake is one of the noisiest single data points in semiconductor equipment. Customers make multi-year commitments; they don't re-order every quarter. One light quarter means almost nothing when the visibility horizon is 12–18 months.

Management held full-year 2026 guidance at €30–35 billion revenue and 51–53% gross margin — maintained in the face of Liberation Day tariff uncertainty, which was still actively rolling through markets during the call. That maintenance is confidence, not wishful thinking. CEO Christophe Fouquet pointed specifically to accelerating AI-driven fab investment as the demand underpinning the guidance range.

Q2 2026 guidance: €7.2–7.7 billion revenue at 51–53% gross margin. Back-half loading is normal for ASML — Q4 is consistently their strongest quarter, and the trajectory implies a meaningful second-half step-up to hit the midpoint of full-year guidance.

Extreme ultraviolet lithography uses 13.5nm light to etch transistor patterns at dimensions that would be physically impossible with conventional laser-based tools. The previous generation — DUV immersion using 193nm ArF light — can be pushed via multi-patterning to approximately 5nm, but beyond that, you need EUV. There are no workarounds. The physics is non-negotiable.

What ASML built to generate and control 13.5nm light is genuinely one of the most complex machines humanity has ever made. A CO₂ laser fires 50,000 pulses per second at molten tin droplets, each smaller than a human blood cell. The resulting plasma emits EUV radiation, which is then focused through mirrors coated in alternating layers of molybdenum and silicon — each layer precisely 3.5 nanometers thick — made exclusively by Carl Zeiss SMT, in which ASML owns a 24.9% stake. The entire system operates in a near-perfect vacuum because EUV photons are absorbed by air. At full production throughput, it processes 160 wafers per hour. The machine ships in 40 freight containers, takes months to install, and costs approximately €200 million.

Building this required over two decades of development and more than €6 billion invested before ASML shipped the first production-ready EUV system. The company holds over 100,000 patents across the technology stack. Nikon builds credible DUV systems. Canon builds lithography tools. Neither has shipped an EUV machine to a customer — and after watching ASML spend 25 years getting to commercial EUV, neither is close.

EUV (NXE series) revenue grew from €2.8B in 2019 to €10.4B in 2025. The 2024 dip to €7.9B combined EUV was driven by China export restrictions removing a major customer, not by a falloff in underlying demand. The 2025 recovery to €11.6B combined confirms that. Every leading-edge foundry and IDM — TSMC, Samsung, Intel, SK Hynix — is expanding EUV capacity aggressively to meet AI workload demand, and all of that expansion flows through ASML's order book.

Standard EUV (the NXE series) has a numerical aperture of 0.33 — a measure of the system's ability to resolve fine detail. ASML's next generation, the EXE:5000, pushes NA to 0.55, enabling chip features below 2nm. The machine is larger, more complex, and approximately €330–380 million per unit — nearly double the price of its predecessor. ASML shipped its first EXE:5000 to Intel in 2023. As of Q1 2026, only a handful of units have been delivered globally.

That handful generated €1.16 billion in FY2025 revenue — up from €465 million in FY2024. This is not a rounding error. High-NA EUV grew 149% year-over-year and will continue growing as the industry's most advanced fabs require sub-2nm capability by the end of the decade. TSMC has committed to High-NA for its N2P and A14 nodes. Samsung is committed. Intel's 14A process is explicitly dependent on it.

The economics are straightforward: if ASML ships 15 EXE units per year at €350M average ASP, that's €5.25 billion from one product line that didn't exist three years ago. The 2030 targets imply 25–40+ EXE units annually. Even at the low end, this is a business-within-the-business that alone would rank among the top 100 global technology companies by revenue.

And the competitive situation in High-NA is even more one-sided than standard EUV. Nobody is attempting to build a competing High-NA system. ASML has no credible timeline for any competitor, because there is no competitor. The EXE:5200 — the second generation of High-NA — is already in development. Customers who want to manufacture 1nm-class chips in the early 2030s are already in conversations with ASML about it.

Since October 2022, ASML has been prohibited from shipping EUV machines to China under Dutch export control regulations implemented under sustained US government pressure. Since late 2023, advanced DUV immersion systems (ArF immersion) have also required licenses that are not being approved. China represented approximately 10–15% of ASML's revenue before restrictions. That's real money — roughly €3–4 billion per year on a €32 billion base — and the market correctly assigns a risk premium to its loss or further escalation.

What gets less attention is the structural implication. The restriction creates a permanent ceiling on Chinese semiconductor capability. SMIC cannot produce chips below approximately 7nm equivalent without access to EUV. Hua Hong cannot. CXMT can build DRAM, but not at HBM-grade density. The domestic Chinese lithography program — SMEE — is years behind and has produced nothing competitive at advanced nodes. The gap between what Chinese fabs can make and what the global AI supply chain requires is not narrowing; it's widening.

This means every AI chip, every HBM stack, every advanced logic process that matters for the next decade flows exclusively through TSMC, Samsung, SK Hynix, and (increasingly) Intel — all of whom are ASML customers with no alternative supplier for leading-edge equipment. The machines excluded from China are the machines every other customer wants more of.

The bear case is escalation — specifically, a further restriction on all DUV equipment (including older KrF and ArF Dry systems), which would add another €1.5–2.0 billion in potential annual revenue risk. This is a real risk. But even in that scenario, ASML's core franchise — EUV and High-NA for the world's most advanced fabs — is unaffected.

Most ASML coverage focuses on new system orders. The installed base story gets almost no attention, and it is one of the most durable, high-margin revenue streams in enterprise technology.

ASML has shipped approximately 8,000+ DUV machines and 130+ EUV machines globally as of 2025. Every one of those machines requires ongoing maintenance contracts, software upgrades, optics calibration, spare parts, and remote monitoring services. These contracts are non-discretionary: a fab that lets an ASML machine go uncalibrated or unpatched produces defective wafers. The customer can't afford to skip service.

Service revenue in FY2024 was €6.49 billion. In FY2025, based on the total-less-product breakdown, service grew to approximately €8.19 billion — a 26% increase year-over-year. That growth rate exceeds total company revenue growth (15.6%) and is driven purely by installed base expansion. As more EUV machines ship and age into higher service intensity, this line should compound at double-digit rates for years.

Service gross margins are estimated at 60%+, materially above new system margins. This means the mix shift toward service as the installed base grows is also a margin tailwind. By 2030, with 300+ EUV machines installed globally, service revenue from EUV alone could approach €4–5 billion annually. The total service business could be a €12–15 billion revenue line by then — more than ASML's entire company revenue was in 2018.

ASML grew revenue from €18.6 billion in FY2021 to €32.7 billion in FY2025 — a 76% increase in four years, at consistent 51–53% gross margins throughout. This is a company executing its own structural advantage at scale, not a company riding a product cycle.

The quarterly picture shows Q4 2025 at €9.72 billion — the largest single quarter in company history — followed by a normal Q1 2026 seasonal step-down to €8.77 billion. The FY2026 guidance implies a back-half ramp, which is consistent with ASML's historical pattern. If they hit €8.5–9.0 billion per quarter in Q3-Q4 2026, the midpoint of guidance is very achievable.

Free cash flow has been remarkable: €10.65 billion in FY2025 (FCF yield ~1.7% on current market cap, low but consistent with monopoly-quality compounders). ASML returned €8.17 billion to shareholders in FY2025 through buybacks (€5.72B) and dividends (€2.45B). The balance sheet carries €7.97 billion in cash against €2.71 billion in debt — net cash of €5.26 billion. There is no financial leverage risk here.

ASML's range of outcomes is unusually bounded on the downside by backlog coverage and on the upside by how aggressively the AI buildout accelerates fab investment plans. The bear case requires something to go wrong, not just slow.

The probability-weighted outcome is approximately $1,780 — about 11% above today's price. That's not a ripping trade setup from a pure return standpoint. The argument for holding (and adding on weakness) isn't that it's cheap. It's that the duration of the structural thesis is measured in decades, not quarters, and the opportunity cost of being out of the world's most important hardware monopoly during an AI infrastructure buildout is high.

The tight consensus-to-price spread matters. When a stock trades near consensus, it typically means: (1) the obvious bull thesis is already priced in, or (2) the stock is being systematically undervalued and will eventually re-rate meaningfully higher. For ASML, I lean toward the latter — the 2030 targets aren't fully priced at current multiples, and any indication those targets get raised would send consensus scrambling upward.

ASML reports in euros. The stock trades in dollars on NASDAQ. Using an assumed EUR/USD rate of approximately 1.09 throughout.

The math is clearest at Method 3. If ASML hits even the low end of its 2030 targets and the market grants a 40× terminal multiple (conservative for a monopoly compounder with 56%+ margins), the stock doubles from here. If they hit the midpoint at 45×, it's 2.3× from here. The question is whether you believe in those targets.

The targets are ambitious — they require ~10% annual revenue growth from 2025's €32.7B base and meaningful margin expansion from 52.8% toward 56–60%. But the demand driver — AI infrastructure requiring advanced logic and HBM at scale — is secular, not cyclical. And ASML is the only company that can supply the production equipment. I don't need the high end to materialize. I just need the low end to be directionally correct.

At $1,606, you're paying 57× trailing earnings for a monopoly growing at 10–15% annually with net cash, 50%+ ROE, €10.6B in FCF, and a 2030 roadmap that makes the current valuation look reasonable in hindsight if they execute. The risk is you're paying up in the near term for duration. The opportunity is that duration is exactly the right thing to hold in an AI infrastructure buildout that's just getting started.