Author: Niusike, Deep Tide TechFlow
13.2 billion euros.
This is the total value of orders ASML received in Q4 2025, more than double market expectations of 6.3 billion euros, and a record high for a single quarter in ASML's history.
On January 28th, the lithography giant ASML delivered a report card that made Wall Street sit up and take notice, with its stock surging 5% in pre-market trading.
This earnings report is a signal flare, offering a glimpse into how the AI frenzy is truly transmitting to the upstream semiconductor supply chain.
Orders Explode Overnight
In Q4 2025, ASML achieved revenue of 9.7 billion euros, a net profit of 2.8 billion euros, and a gross margin of 52.2%. These figures were within expectations, even somewhat平淡 (flat).
But the orders—that's the number that truly drove the market wild.
Of the 13.2 billion euros in orders, 7.4 billion came from EUV lithography machines. These machines, costing over 200 million euros each, are the only key for TSMC, Samsung, and SK Hynix to push towards 3nm, 2nm, and even more advanced processes.
More crucially, a structural shift in the orders reveals a key signal: memory manufacturers have entered疯狂 (crazy) mode.
Logic orders were 5.8 billion euros, up 3 billion sequentially, in line with seasonal fluctuations.
Memory orders were 7.4 billion euros,暴增 (soaring) by 4.9 billion sequentially, far exceeding historical levels for the same period.
Memory giants like SK Hynix, Samsung, and Micron are placing orders超季节性 (superseasonally) and疯狂 (frantically). The arms race for HBM (High Bandwidth Memory) capacity is white-hot.
CFO Roger Dassen was blunt: "Customers have become significantly more optimistic about the mid-term market outlook, primarily because they see AI-related demand as more persistent."
This optimism is directly translating into hard cash orders. ASML's order backlog has reached 38.8 billion euros, of which 25.5 billion are EUV orders. Put simply: the next two years' work is pretty much fully booked.
Who is Ordering Like Crazy? South Korea's "Revenge"
Looking at the regional revenue breakdown reveals an interesting detail.
In Q4 2025, Mainland China remained ASML's largest revenue source, contributing 36% or 3.5 billion euros. This figure far exceeded the company's previous expectation of 25%, mainly due to Chinese manufacturers' continued stocking of ArFi (Immersion DUV) equipment.
But the real focus is on South Korea, where Q4 revenue share rebounded to 22%, about 2.1 billion euros.
Behind this number lies the疯狂 (frenzied) expansion by SK Hynix and Samsung. SK Hynix has explicitly stated it will procure 12 EUV machines in 2026, going all out to boost HBM3e capacity, as Nvidia's H100, H200, and B200 are all嗷嗷待哺 (crying out to be fed).
Samsung is even more desperate. They have been beaten badly by SK Hynix in the HBM market, struggling with yield rates. The AI boom now gives them a final window of opportunity; without疯狂 (frenzied) expansion, they risk being彻底出局 (completely eliminated).
Barclays analysts directly named it: SK Hynix will secure 12 EUV machines in 2026. These are big-ticket items at 260 million euros each—12 units mean 3.12 billion euros.
More importantly, this peak in orders from memory manufacturers has only just begun.
Micron just announced 2026 capital expenditures exceeding $20 billion, a year-on-year increase of nearly 40%. TSMC raised its 2026 capex forecast to $52-56 billion, over $10 billion more than originally planned. This money will ultimately flow to ASML.
2026: From "Cautious" to "Aggressive"
A few months ago, ASML was still saying "2026 revenue might be flat or even decline."
Now? They directly provided a revenue range of 34-39 billion euros, with a midpoint of 36.5 billion euros, representing 12% growth over the 32.7 billion euros in 2025.
But this guidance is actually conservative.
Why? Because after TSMC and Micron raised their capex, mainstream institutions have raised their 2026 growth expectations for ASML to over 20%.
ASML's own guidance is "growth of 4-19%," with a midpoint of only 12%. This type of guidance, "below market expectations," is rare among equipment stocks. These companies usually prefer to give a "stretch" target, then exceed it, reaping a stock price surge.
Why is ASML so conservative? Possibly two reasons:
First, uncertainty in the Chinese market. ASML expects China's market share to drop to 20% in 2026, compared to nearly 30% in 2024-2025. What does 20% mean? If 2026 revenue is 36.5 billion euros, the China market would be 7.3 billion euros, more than double the Q4 2025 single-quarter figure of 3.5 billion euros. But considering export controls and the fading stocking demand, whether this number can be achieved is unknown.
Second, the High-NA EUV production ramp. ASML delivered two High-NA machines (EXE:5200B) in Q4 2025 and recognized the revenue. These monsters, priced at 380 million euros each, could easily push revenue higher if more are sold in 2026. But the problem is, besides Intel, other customers are still观望 (waiting and seeing). TSMC believes existing Low-NA EUV plus multi-patterning is sufficient down to 1nm and below, and is in no hurry to adopt High-NA.
So ASML gave guidance that "leaves room." In the equipment stock world, this is actually a positive signal, meaning the company is confident about exceeding expectations but doesn't want to be too boastful.
Subtle Changes in Product Mix
Digging into the product data reveals some interesting points.
EUV: Shipped 14 units in Q4, average selling price (ASP) 260 million euros, revenue 3.64 billion euros, up 22% year-on-year.
ArFi (Immersion DUV): Shipped 37 units, ASP 82 million euros, revenue 3.03 billion euros, up 4% year-on-year.
EUV and ArFi together accounted for nearly 88% of revenue, ASML's absolute cash cows.
But there's a detail: EUV revenue growth was driven mainly by higher ASP, not shipment growth. Q4 EUV shipments of 14 units weren't exceptionally high, but the ASP rose from around 240 million euros previously to 260 million euros.
Why did the ASP increase? Because the product mix is upgrading.
ASML's main sales are currently the NXE:3800E Low-NA EUV, which has reached its target capacity of 220 wafers per hour, even hitting 230 wph in some customer scenarios. It's a "mature to the extreme" money printer.
Meanwhile, the High-NA EUV (EXE series) has a unit price of 380 million euros, almost double that of Low-NA. Revenue from two High-NA units was recognized in Q4 2025, directly pulling up the EUV ASP.
If High-NA shipments increase from 2 units in 2025 to, say, 10 units in 2026, ASML's revenue growth could easily break 20%.
But this depends on when TSMC and Samsung truly place orders. Currently, Intel is the most aggressive buyer, having accepted the first 5200 model tool for high-volume manufacturing, intended for its 14A process.
TSMC is观望 (watching). Samsung is犹豫 (hesitating). Whoever adopts High-NA first gains an advantage in the process race below 2nm. But no one wants to be the first to try the crab, as High-NA yield ramp-up takes time and money.
The Underlying Logic of AI: Compute is Money, Memory is the Bottleneck
The root cause of this order explosion is simple: AI is consuming everything.
But unlike a year ago, the transmission path of AI demand has changed.
A year ago, the focus was on "how many GPUs Nvidia sold" and "whether TSMC's CoWoS capacity was enough." Now, the bottleneck has shifted to memory.
The demand for compute power from large models like ChatGPT, Gemini, and Claude is insatiable. But compute alone isn't enough; you need fast enough memory to feed these compute monsters.
HBM3e is high-speed memory specifically designed for AI. Its bandwidth is over 6 times that of standard DDR5, preventing GPUs from being bottlenecked by memory I/O during training and inference.
The problem is, HBM capacity is extremely tight.
SK Hynix dominates, holding about 80% of the HBM market share. Samsung struggles with yield, holding only around 15%. Micron just entered the game, with mass production expected in 2026.
Nvidia, AMD, Google, Microsoft, Amazon are all scrambling for HBM. The supply-demand gap is expected to last at least until 2027.
So memory manufacturers aren't just expanding capacity; they are "expanding desperately."
Micron's 2026 capex is $20 billion, up nearly 40% year-on-year. SK Hynix wants 12 EUV machines. Samsung has increased its memory business capex by over 50%.
This money will ultimately become ASML orders.
Looking at the application split, logic chips currently account for 70% of ASML's system sales, memory for 30%. But in the order structure, memory is already接近 (approaching) 60%.
This means that over the next 12-18 months, memory will be the biggest engine for ASML's revenue growth.
China Market: From "Stocking Frenzy" to "Normalization"
If there's any less optimistic news, it's regarding the China market.
ASML expects its China market revenue share to drop to 20% in 2026, compared to nearly 30% in 2024-2025, even reaching 36% in Q4 alone.
Why the decline? Because the 2024-2025 China orders were essentially "panic buying" demand.
US restrictions on chip exports to China are tightening. ASML can no longer get export licenses for some high-end DUV equipment (particularly immersion ArFi). Chinese fabs know this: buy as much as you can while you still can.
But this stocking frenzy cannot last indefinitely.
By 2026, Chinese manufacturers will have sufficient equipment on hand, and new procurement demand will naturally decline. A 20% share is actually not low—it equates to an annual scale of 7-8 billion euros.
But for ASML, accustomed to China's "excess contribution," this is a change that needs adapting to.
More troublesome is the gross margin.
The Chinese market primarily buys high-margin immersion DUV equipment (ArFi), which have even higher margins than Low-NA EUV. A decline in these orders will put some pressure on毛利率 (gross margin).
This is also why ASML set its 2026 gross margin guidance at 51-53%, slightly lower than the full-year 2025 figure of 52.8%.
Here's a detail: Q4 gross margin was 52.2%, while the full-year 2026 guidance is 51-53%. This implies ASML expects H1 2026 gross margin to be below 52%, possibly due to fewer high-margin orders from China.
High-NA: The Bet for the Next Decade, or an Expensive "Option"?
Beyond conventional EUV, ASML is preparing a big move: High-NA EUV.
This is the next-generation lithography technology, with a unit price of 380 million euros. In Q4 2025, ASML already delivered two High-NA tools and recognized the revenue.
Currently, Intel is the most aggressive buyer, having accepted the first 5200 model tool for high-volume manufacturing, intended for its 14A process.
ASML is confident in High-NA's performance: "Imaging, performance, and overlay results are all very good."
But widespread adoption of this technology might not happen until 2027-2028.
Why? Because TSMC and Samsung are观望 (watching).
TSMC's logic is clear: Our 2nm is already in production using Low-NA EUV plus multi-patterning. Yields are good, costs are controllable. 1.4nm is also on the way, still using Low-NA. Only below 1nm will we need High-NA.
In other words, TSMC believes Low-NA EUV can fight for another three years.
Samsung's attitude is more微妙 (nuanced). They suffered greatly with 3nm yields; the GAA architecture ramp was too slow, leaving them two generations behind TSMC. Samsung's most urgent task now is to fix 3nm yields and win back lost customers.
High-NA? Let Intel test the waters first.
But if Intel successfully mass-produces its 14A process using High-NA with competitive yields and costs, TSMC and Samsung will immediately follow. Then, High-NA orders will pour in like an avalanche.
This is why ASML dares to say "2030 revenue of 44-60 billion euros."
Because High-NA isn't just an equipment upgrade; it also doubles the unit price from ~200 million to ~400 million euros. As long as shipment volumes don't fall, revenue easily doubles.
Layoffs of 1700: Optimization, or Anxiety?
Amidst this beautiful earnings report, ASML buried some less pleasant news: plans to lay off 1700 people, mainly in the Netherlands, partly in the US.
The reason given: "The company's way of working has in some cases become less agile."
This explanation is微妙 (subtle). ASML isn't short of money or orders but feels the organization has become bloated.
The layoffs focus primarily on technology and IT departments, aiming to "strengthen engineering and innovation capabilities in key areas."
Translation: Cut peripheral departments, concentrate resources on core technology R&D.
This is actually a positive signal. ASML is preparing for larger-scale growth and needs a more flexible, efficient organizational structure.
But there's another interpretation: ASML might be concerned about High-NA EUV's development progress and yield ramp. Laying off 1700 people saves money that can be invested into more critical technology攻坚 (breakthroughs).
After all, Intel has the first High-NA tool. If their 14A process量产 (mass production) goes smoothly, TSMC and Samsung will follow immediately. Then, ASML must have sufficient capacity and technical reserves to handle the order deluge.
12 Billion Euro Buyback: Confidence, or "Stabilization"?
ASML announced a new 12 billion euro share buyback program, to be completed by the end of 2028.
The previous buyback plan (2022-2025) was also 12 billion euros but only 7.6 billion was actually completed.
Why wasn't it completed? Because the stock price rose too fast, making repurchases too expensive; the company found it not cost-effective.
The new buyback plan sends two signals:
First, the company is confident in future cash flow. 12 billion euros is not a small number. ASML's willingness to allocate this money for buybacks shows they believe profitability will continue to strengthen in 2026-2028.
Second, the current stock price is attractive. If ASML thought the stock was overvalued, they wouldn't rush to buy back. Launching a buyback plan suggests management sees current valuation as reasonable or even low.
The dividend also increased by 17%, to 7.50 euros per share. This is ASML's consistent style: make money, give it to shareholders.
One Company, Defining an Era
ASML's earnings are never just about one company's performance.
They are a barometer for the entire semiconductor industry chain, a thermometer for the real transmission of the AI wave to the manufacturing end, and a fuel gauge for whether TSMC, Samsung, Nvidia, and others can continue their狂奔 (sprint).
In 2025, ASML achieved revenue of 32.7 billion euros, net profit of 9.6 billion euros (up 16% YoY). Gross margin 52.8%, net margin 29.4%.
For 2026, the company expects revenue growth of 12% (conservative estimate), the market expects over 20% (aggressive estimate).
By 2030, ASML's target is 44-60 billion euros in revenue, with a gross margin of 56-60%.
This is not a conservative company. This is a company convinced that AI will rewrite the semiconductor industry landscape and is poised to capture the biggest红利 (dividend).
Current market capitalization is 563.5 billion dollars. The 2026 expected P/E is about 39x, sitting in the middle of its historical 30-45x valuation range.
But this valuation might underestimate ASML's true value.
Why? Because two cycles are叠加 (superimposing):
Short-term cycle (2026-2027): AI and memory demand explode. TSMC, SK Hynix, Samsung, Micron expand疯狂 (frantically). ASML's books are full, revenue grows rapidly.
Medium-to-long-term cycle (2027-2030): High-NA EUV begins volume production. Unit price jumps from ~200 million to ~400 million euros. Revenue ascends to another level.
The superposition of these two cycles could make ASML's "super cycle" last until 2030, or even longer.
The record 13.2 billion euro quarterly order intake is just the beginning.
This feast has only just begun.
