Samsung bets big on the machines that print the AI era

The most expensive precision machine on earth just got its most important new customer.

The machine deserves its superlative explained. Extreme ultraviolet lithography prints circuits with light of a wavelength so short it is absorbed by air itself — which is why the machines operate in vacuum, with mirrors polished to atomic smoothness, at a price that makes each unit one of the most expensive manufactured objects on earth. ASML is the only company that builds them. There is no second source, no alternative supplier, no plan B: every advanced chip on the planet — the processors in phones, the accelerators in data centers — begins as light through a Dutch machine. When Samsung deepens its partnership around the next generation of that machine, it is not buying equipment. It is buying position in the only supply chain that matters.

High-NA is the next turn of that monopoly's crank. Numerical aperture — the 'NA' — measures how much light the optics can gather; more aperture means finer resolution, the ability to draw smaller features. Smaller features mean more transistors per chip, and in memory, more bits per wafer. The geometry compounds into economics: finer resolution means fewer manufacturing steps to reach the same density, and in chipmaking, steps are money — each step is time, chemicals, energy, yield risk. Samsung's 2028 target is therefore not a technology demo. It is a cost roadmap: whoever prints finest, prints cheapest, and cheapest wins the memory cycle.

Samsung Electronics has deepened its strategic partnership with Dutch lithography giant ASML around High-NA extreme ultraviolet lithography — the chip-printing technology that draws the finest circuit features in the history of manufacturing. The announcement, carried September 25 by The Indian Eye, is a declaration of intent: Samsung wants to lead the memory industry into the High-NA age.

DRAM is the unglamorous hero of the AI age, and the article's emphasis is precise. Dynamic random-access memory is the working memory inside every data center, laptop and phone — the fast, temporary storage the processors actually compute against. Every layer of AI sophistication demands more of it: larger models need more memory bandwidth, more bandwidth needs more bits, more bits need finer lithography. The logic is cumulative and merciless. The AI boom is usually told as a story about GPUs. It is equally a story about memory — and memory is where Samsung intends to strike first with High-NA.

The memory industry's structure makes the race legible. Three companies make most of the world's DRAM — Samsung, SK Hynix, and Micron — in one of the most consolidated and cyclical markets in technology. Leadership rotates with the manufacturing transitions: whoever masters the next process node first gets the cost advantage, and the cost advantage funds the next transition. Samsung's 'first in the industry' claim for 2028 High-NA DRAM is a bid to seize that rotation — with SK Hynix targeting the same year, the domestic rivalry gives the race its edge. Two Korean giants, one Dutch machine, and a single year as the prize.

The centerpiece commitment is a date: 2028. That is when Samsung plans to introduce High-NA EUV into high-volume DRAM production — the first in the industry, by its account. DRAM is the memory inside every data center, laptop and phone, and making it is a game of printing ever-smaller features ever more economically.

Why it matters is a matter of geometry. Higher numerical aperture — the "NA" in High-NA — means finer resolution. Finer resolution means fewer manufacturing steps to achieve the same density, and steps are money. Fewer steps, lower cost, a roadmap that stretches further into the future.

But machines this precise need new infrastructure around them. Samsung has also joined the industry initiative — originally launched with Taiwan's TSMC — to graduate photomasks from the 6-inch format that has served the industry for decades to 12-inch masks built for High-NA EUV. The timeline runs long: a pilot line by 2031, full production readiness by 2033.

A mask is the stencil through which the light shines; doubling its size is a fundamental retooling of the supply chain. The 2031–2033 horizon tells you how seriously the industry takes this transition: it is planning in half-decades, not quarters.

Meanwhile the race is no longer theoretical. Intel already runs High-NA EUV on select layers of its 18A process. Samsung's memory rival SK Hynix is also targeting 2028 for DRAM. And TSMC — long the technology's most prominent skeptic — has now scheduled volume use from 2030. The skeptics are aboard; the race is on.

The mask initiative is the story's longest fuse, and its most revealing. A photomask is the stencil through which the light shines — the master pattern, printed once and projected millions of times. The industry has used the 6-inch format for decades; graduating to 12-inch masks built for High-NA EUV means retooling the entire mask supply chain, from the blank makers to the inspection tools. The timeline tells you the scale: a pilot line by 2031, full production readiness by 2033. The industry is planning in half-decades, not quarters, because the tooling ecosystem cannot move faster. When Samsung joined the initiative originally launched with TSMC, it was signing up for a decade of coordinated retooling.

That coordination is the quiet miracle of the chip industry. Competitors who sue each other over patents collaborate on the infrastructure they all need — the mask formats, the resist chemistry, the metrology — because no single company can move the supply chain alone. The 2031–2033 horizon is the industry's way of saying the transition is real: you do not plan pilot lines seven years out for a technology you doubt. The skeptics are aboard, as the article notes — TSMC, long the technology's most prominent skeptic, now scheduled for volume use from 2030. When the skeptic schedules, the debate is over. What remains is execution.

The race lineup deserves its scorecard. Intel already runs High-NA EUV on select layers of its 18A process — the American incumbent proving the machines work in production, not just in the lab. SK Hynix, Samsung's memory rival, targets 2028 for DRAM — the same year, the same prize, the fiercest head-to-head in the industry. TSMC, the foundry king, comes from 2030 — later, but with the manufacturing discipline that has made it the world's most reliable chipmaker. Four different strategies, one machine. The Dutch company in Veldhoven sells to all of them, which is the most comfortable position in technology: the arms dealer in a war everyone must fight.

Demand running ahead of supply is the bottleneck the AI era did not expect. ASML's existing EUV capacity was reported nearly sold out through the end of 2027, with AI-driven chip complexity accelerating every customer's expansion plans. The bottleneck of the AI era is not data, not talent, not even GPUs — it is the machines that print the machines, and there are only so many machines. Each High-NA tool takes years to build, ships in dozens of freight containers, and requires an installation effort measured in months. The queue is the market: everyone who wants the future must wait for Veldhoven to build it.

Demand is running ahead of supply. ASML's existing EUV capacity was reported nearly sold out through the end of 2027, with NeoTeo reporting on September 15 that AI-driven chip complexity is accelerating every customer's expansion plans. The bottleneck of the AI era is not data, not talent, not even GPUs — it is the machines that print the machines.

The logic is cumulative. Every layer of AI sophistication demands more memory bandwidth; every memory generation demands finer lithography; every lithography generation demands rarer machines. Samsung's bet is that owning the transition early is worth the enormous capital outlay.

There is also a geopolitical undertow, and everyone in the industry knows it. High-NA EUV sits at the center of the world's most consequential supply chain — Dutch machines, Japanese and American components, Korean and Taiwanese fabs. Who masters the tooling masters the roadmap, and everyone else queues.

The next milestone is 2028. If Samsung delivers High-NA DRAM on schedule, it will have done something none of its memory rivals have: turned the most exotic machine in semiconductor manufacturing into a memory workhorse. If it slips, the race resets — and SK Hynix, Intel and TSMC will be waiting.

The capital intensity is the bet's dark side, and the Western coverage's caution is well placed. These machines cost a fortune — the most expensive precision instruments ever sold — and shipments are measured in units per year, not thousands. The economics only work if the memory market cooperates: DRAM is famously cyclical, swinging from shortage to glut, and a High-NA fab ordered in a boom can open into a bust. Samsung's wager is that owning the transition early is worth the outlay — that the cost advantage of finer printing pays back across the cycle. It is the same wager the industry has made at every node transition. It has paid before. The stakes have never been this high.

There is also the geopolitical undertow, and everyone in the industry feels it. High-NA EUV sits at the center of the world's most consequential supply chain — Dutch machines, Japanese and American components, Korean and Taiwanese fabs — and the technology is too strategic to be merely commercial. Export controls already fence the most advanced lithography away from China; each generation of the technology redraws the map of who can make what. Who masters the tooling masters the roadmap, and everyone else queues — for the machines, for the capacity, for the future. Samsung's partnership is industrial strategy conducted through procurement.

Western coverage — Zacks, TechTimes — frames the partnership as the commercial validation of ASML's biggest bet.

The emphasis: ASML's existing EUV tools are nearly sold out through 2027, and High-NA is the next growth engine. Samsung's 2028 DRAM target gives the technology a memory-industry beachhead, not just a logic one — expanding the addressable market.

The caution, rarely stated aloud: these machines cost a fortune, shipments are measured in units per year, and the economics only work if the memory market cooperates. High-NA is a masterpiece; masterpieces are expensive.

The Western read of the mask initiative: 2031 pilot, 2033 volume — a sign the ecosystem is maturing exactly on schedule for a technology this young.

The Western commercial-validation frame captures ASML's side of the trade. High-NA is the company's biggest bet — years of R&D;, a machine that barely fits in a factory — and Samsung's 2028 DRAM target gives the technology its memory-industry beachhead, not just a logic one. That expansion of the addressable market is what the stock analysts at Zacks are pricing: every DRAM maker that adopts High-NA is a new customer for the most expensive product ASML has ever sold. The sold-out EUV capacity through 2027 is the present; High-NA is the growth engine. The partnership announcement is, from Veldhoven's perspective, the future arriving on schedule.

The mask-transition read completes the Western picture. A 2031 pilot and 2033 volume readiness for 12-inch masks is not a delay — it is the ecosystem maturing exactly on the schedule a transition this fundamental requires. New mask formats mean new tooling orders across the lithography supply chain: blank suppliers, inspection systems, pellicle makers. MarkLines and the Japanese supply-chain press track this secondary effect because it is where the equipment super-cycle lives — a multi-year wave of capital spending that lifts the entire food chain. The machine gets the headlines. The ecosystem gets the revenue.

Eastern coverage — Samsung Newsroom, Korean and Japanese trade press — treats the deal as industrial strategy as much as technology.

The emphasis: Samsung is not merely buying machines; it is co-developing the future of memory manufacturing with the one company that can deliver it. "Industry first" is both a claim and a warning to rivals.

For Korea, the partnership is sovereignty-adjacent: in an era when chip supply chains are geopolitical assets, owning the lithography relationship is owning the memory roadmap. SK Hynix's parallel 2028 target sharpens the domestic rivalry.

MarkLines and the Japanese supply-chain press note the secondary effect: a 12-inch mask transition means new tooling orders for the entire lithography ecosystem — a multi-year equipment super-cycle.

The Eastern industrial-strategy frame treats the deal as sovereignty-adjacent, and the adjective is chosen carefully. In an era when chip supply chains are geopolitical assets — when the ability to make advanced memory is a national capability, not just a corporate one — owning the lithography relationship is owning the memory roadmap. Samsung is not merely buying machines; it is co-developing the future of memory manufacturing with the one company that can deliver it. 'Industry first' is both a claim and a warning to rivals: the warning is the point. For Korea, the partnership secures the country's position at the top of the memory ladder for another generation.

The domestic rivalry sharpens the strategy. SK Hynix's parallel 2028 target means the two Korean giants will race each other into the High-NA age — and internal competition has always been the Korean industry's accelerant. Samsung Newsroom's framing, and the Korean trade press's, presents the ASML partnership as the decisive move in that race: deeper co-development, earlier access, the inside track. Whether the inside track converts to market leadership will be decided in the fabs, one wafer at a time. But the announcement has already done its strategic work: it told the industry, the rivals, and the governments that Samsung intends to lead.

The South — The Indian Eye, regional tech press — reads the announcement as a map of who builds the AI era and who buys it.

The emphasis: the value chain of artificial intelligence runs through Veldhoven, Suwon and Hsinchu. The countries racing to build AI data centers — India prominent among them — are downstream of these partnerships, buyers of the memory the machines print.

The question is whether any of the value escapes upstream. India's semiconductor ambitions are precisely an attempt to stop being a permanent customer; Samsung's announcement is a reminder of how steep the ladder is.

The South's quiet hope: every memory generation that gets cheaper also makes AI cheaper to run everywhere else. The machines print in Korea; the benefits, eventually, print everywhere.

The Southern map-of-who-builds reading is the story's widest angle. The value chain of artificial intelligence runs through Veldhoven, Suwon and Hsinchu — the Dutch machines, the Korean memory, the Taiwanese foundries — and the countries racing to build AI data centers are downstream of these partnerships, buyers of the memory the machines print. India, prominent among them, is pouring concrete for the AI era while importing every critical component of it. The question is whether any of the value escapes upstream, or whether the downstream stays downstream forever.

India's semiconductor ambitions are precisely an attempt to climb out of the permanent-customer trap — fabs planned, incentives offered, the long climb begun. Samsung's announcement is a reminder of how steep the ladder is: the frontier keeps moving, and each generation demands rarer machines, deeper partnerships, longer timelines. But the South's quiet hope has its own logic: every memory generation that gets cheaper also makes AI cheaper to run everywhere else. The machines print in Korea; the benefits, eventually, print everywhere. The 2028 milestone will be judged in Suwon. Its consequences will be felt in every data center on earth — including the ones India is building now, with memory it still must buy.