SK hynix High-NA EUV Plan Targets 2028 as Investors Demand More Cash

SK hynix has set a 2028 target for High-NA EUV in DRAM mass production, even as investors press the company to return more AI-generated cash. The SK hynix High-NA EUV plan places an expensive manufacturing transition beside an increasingly urgent capital-allocation debate.

High-NA EUV, or extreme ultraviolet lithography with a higher numerical aperture, prints smaller circuit features than current EUV systems. SK hynix says it is evaluating a related consortium focused on larger photomasks, according to its reported 2028 production target.

The timing matters because SK hynix is no longer defending investment during a downturn. It is deciding how to divide record earnings between new factories, advanced equipment, financial reserves, dividends, and share repurchases. Micron’s more aggressive return commitment gives shareholders another benchmark, while Samsung faces similar pressure in South Korea.

This is not simply a story about a better lithography machine. It is a test of whether SK hynix can preserve its manufacturing lead without reinforcing the old memory-sector habit of retaining cash throughout the cycle.

SK hynix High-NA EUV Moves From Research Toward a 2028 Factory Target

The important change is SK hynix’s decision to attach a production year to a technology that has largely remained in evaluation.

SK hynix says it aims to use High-NA EUV processes in DRAM mass production by 2028. That target turns a research program into a manufacturing deadline, although the company has not publicly detailed the first DRAM generation involved.

Current EUV systems use optics with a numerical aperture of 0.33. High-NA systems raise that figure to 0.55, improving the resolution available in a single exposure. Better resolution can reduce the need to split one intricate pattern across several lithography steps.

That distinction matters for advanced DRAM. Every added patterning step introduces more process time, alignment work, inspection, and potential yield loss. High-NA EUV promises a way to print smaller features without allowing that complexity to grow at the same rate.

SK hynix already has relevant production experience. It introduced EUV into mass production for its fourth-generation 10-nanometer-class DRAM, known as 1a DRAM, in 2021. The 2028 plan extends that progression rather than starting an entirely new manufacturing path.

The target also fits the supplier’s schedule. ASML said in January 2026 that it had shipped eight High-NA systems to several customers, with six operating. Its first second-generation EXE:5200B system was running at a customer site and meeting full specifications.

ASML expects the platform to satisfy high-volume manufacturing requirements by the end of 2026. Its published High-NA roadmap points to customer insertion during 2027 and 2028.

“Insertion” does not mean an instant transition to profitable mass production. A memory manufacturer must qualify the equipment, integrate materials and masks, stabilize process control, and reach an acceptable yield. Those steps can take longer than a successful demonstration.

SK hynix has therefore disclosed a destination, not a completed route. The date gives customers and investors a useful marker, but it leaves open which layers will use High-NA and how quickly production volume will increase.

The company is simultaneously considering participation in an ASML and TSMC-led effort involving 12-inch photomasks. Photomasks carry the circuit pattern that a lithography system projects onto a wafer. Most semiconductor production currently relies on 6-inch masks.

That work addresses a physical compromise in the first generation of High-NA EUV. The system can print finer features, but its reduced exposure field limits the size covered in one pass. That limitation is especially relevant when manufacturers print large logic dies.

DRAM dies are generally smaller than the largest data-center processors. Still, mask format, equipment productivity, and manufacturing cost will influence how broadly SK hynix deploys the technology. A credible 2028 target requires progress across that supporting system, not only delivery of a scanner.

AI Memory Profits Have Changed the Capital-Allocation Argument

SK hynix must now justify each major investment against cash returns that shareholders believe the AI memory boom can support.

In 2024, SK hynix established a shareholder return framework for 2025 through 2027. The company raised its annual fixed dividend by 25 percent, from 1,200 won to 1,500 won per share.

It also retained the principle of using half of accumulated free cash flow as the resource for shareholder returns. SK hynix estimated that annual cash dividends would reach 1 trillion won under that framework.

The company reserved 5 percent of free cash flow for strengthening its financial structure. It also set a capex discipline target that places average annual investment in the mid-30 percent range of revenue.

Those rules were created to balance financial resilience with the capital demands of advanced memory manufacturing. The company’s official shareholder return program also allowed an earlier additional return if stronger performance produced a meaningful increase in free cash flow.

The financial backdrop has since become much stronger. SK hynix reported 2025 revenue of 97.1 trillion won, operating profit of 47.2063 trillion won, and net income of 42.9479 trillion won. The company linked those results to AI memory demand and announced additional dividends and share retirement.

HBM, or high-bandwidth memory, stacks memory dies to feed AI processors data at far higher rates than conventional memory arrangements. Demand for HBM has lifted both earnings and the strategic value of advanced DRAM capacity.

That success changes what investors consider a reasonable balance. Cash retention looked defensive when the memory market was recovering from a downturn. It looks more conservative when AI demand produces record profit and customers pursue longer supply commitments.

Investors are also questioning whether the 50 percent return framework still matches the scale of the company’s cash generation. A shareholder policy can be generous in absolute terms while remaining cautious relative to earnings.

The dispute is not simply between spending and saving. Shareholders want SK hynix to establish how much liquidity it actually needs for factories, equipment, research, and the next industry decline.

Management has a legitimate reason to avoid treating one strong cycle as permanent. Memory prices have historically moved through sharp expansions and contractions. Building fabrication capacity also creates long-lived costs that remain after product prices fall.

However, stronger customer commitments can reduce some uncertainty. Multi-year supply agreements offer better visibility than memory manufacturers received during older commodity cycles. They do not eliminate risk, but they weaken the assumption that every AI-related profit surge will disappear immediately.

That is why the SK hynix High-NA EUV plan has become part of a wider credibility test. The company must show that investment has a defined technical purpose and measurable schedule. Otherwise, shareholders can characterize retained cash as an inefficient precaution rather than necessary funding.

The Main Contest Is Investment Discipline Versus Investor Patience

The decisive question is whether SK hynix can finance its next manufacturing transition without asking shareholders to wait indefinitely for larger returns.

Investors are not demanding that SK hynix abandon advanced manufacturing. High-NA EUV, new DRAM capacity, advanced packaging, and HBM production are central to the earnings they want the company to distribute.

Their objection concerns scale, timing, and commitment. If management keeps the option to spend broadly while limiting long-term payout promises, shareholders carry most of the uncertainty. The company keeps the cash whether the cycle strengthens or weakens.

That tension intensified after SK hynix and Samsung reported record AI-driven profits. Reuters found growing capital-return pressure from investors seeking larger dividends or repurchases.

SK hynix and Samsung were expected to hold a combined $263 billion in net cash by the end of 2026, based on LSEG data and Reuters calculations. That estimate was more than double Nvidia’s projected $102 billion.

Both Korean memory companies targeted returns equal to roughly half of free cash flow at the time. Micron had committed to returning 100 percent, giving investors a direct industry comparison.

Richard Clode, a portfolio manager at Janus Henderson Investors, argued that retaining the 50 percent approach would create an inefficient balance sheet. He called for SK hynix to return at least 80 percent of free cash flow.

His criticism also contained a broader market argument. A company that refuses a larger long-term commitment can appear uncertain about the durability of its own earnings. That perception matters because investors still treat memory as a cyclical business.

SK hynix responded that its record cash generation would let it meaningfully expand shareholder returns while maintaining investment and financial soundness. The company said during an earnings call that it was considering additional measures and planned to provide details within 2026.

That statement acknowledges the pressure but does not resolve it. Investors still need to know whether expanded returns mean a temporary distribution, a new percentage of free cash flow, or a durable repurchase framework.

The High-NA program strengthens management’s case for retaining some flexibility. Lithography tools are only one part of the spending required for advanced DRAM. New clean rooms need deposition, etching, metrology, inspection, and process-control equipment.

SK hynix also committed to a large conventional EUV procurement. A March 2026 regulatory filing said the company would purchase 11.95 trillion won of ASML equipment by December 31, 2027.

Analyst David Dao of Bernstein estimated that the order represented about 30 EUV machines over two years. The disclosed EUV equipment order was described as the largest publicly reported single order from an ASML customer.

That equipment is expected to support both HBM and advanced DRAM at facilities including Yongin and M15X in Cheongju. It shows that the 2028 technology target sits within a much larger expansion program.

Yet large investment commitments make allocation transparency more important, not less important. Management must connect spending to capacity, product schedules, customer demand, and expected returns. A large order alone does not show that every retained unit of cash creates more value than a repurchase or dividend.

The strongest answer would be a framework that funds clearly defined strategic projects while automatically returning surplus cash. Such a mechanism would make shareholder returns the result of investment discipline, not a competing request considered after spending decisions.

High-NA EUV Has a Productivity Tradeoff That the Headline Cannot Resolve

The 2028 target is technically plausible, but commercial success depends on throughput, mask infrastructure, yield, and the number of layers that benefit.

High-NA EUV improves optical resolution. It does not automatically lower the cost of each finished DRAM die. Manufacturers must evaluate the entire production sequence.

A High-NA scanner uses anamorphic optics, which magnify the pattern differently along two axes. This design supports higher resolution but produces a smaller exposure field than existing EUV tools.

For a chip larger than that field, a manufacturer may need stitching, which joins multiple exposed sections. That creates alignment challenges and can affect productivity or yield.

The semiconductor industry’s proposed move from 6-inch to 12-inch photomasks is intended to relieve that limitation. A larger mask can support a wider usable exposure area and potentially improve system productivity.

ASML has said the larger format could raise productivity by 40 percent if the industry completes the transition. However, the timetable extends far beyond SK hynix’s initial DRAM target.

The reported larger-mask consortium expects a pilot line by 2031 and high-volume readiness by 2033. SK hynix is evaluating participation rather than presenting the mask transition as a requirement for its 2028 launch.

That distinction prevents two separate timelines from being confused. SK hynix can introduce High-NA EUV on selected DRAM layers in 2028 while the industry continues developing larger masks for later applications.

Samsung is also targeting High-NA EUV for DRAM high-volume production by 2028. TSMC has indicated that it plans to use the technology in advanced-node high-volume production from 2030.

Intel moved earlier. It has already used High-NA equipment in work related to its advanced logic processes. Its experience gives the industry practical data, but logic production and DRAM manufacturing do not present identical process requirements.

Samsung’s matching 2028 goal creates the most relevant competitive comparison. If both Korean memory producers qualify the technology on similar schedules, High-NA could become a necessary cost of maintaining process parity rather than an exclusive advantage.

Micron also matters even without an identical public deadline. It can pursue improvements through conventional EUV, materials, cell architecture, packaging, and process integration. High-NA adoption must outperform those alternatives economically, not only produce finer patterns.

SK hynix therefore has at least four hurdles to clear. The tool must sustain production-grade availability. Its process must deliver adequate yield. The selected layers must reduce enough multipatterning to justify the change. The resulting dies must offer attractive cost or performance.

The public target does not answer those questions. SK hynix has not disclosed expected wafer throughput, layer count, yield, or unit-cost improvement for its 2028 implementation.

That absence is normal several years before a production transition. It still requires careful language. The company has announced an objective, not confirmed a manufacturing advantage.

There is also a timing mismatch between capital spending and technical evidence. Equipment procurement, facility preparation, and engineering teams require funding before high-volume results exist. Shareholders must decide whether the company’s process record justifies that early commitment.

Management can reduce that uncertainty by publishing milestones. Tool installation, initial wafer processing, customer qualification, and yield progression would make the SK hynix High-NA EUV plan easier to evaluate.

Without those markers, 2028 risks becoming a symbolic date. With them, investors can distinguish a controlled technology insertion from an open-ended spending program.

Samsung, Micron, and ASML Set the Boundaries of SK hynix’s Choice

SK hynix controls its capital policy, but competitors and one critical supplier control how much flexibility that policy retains.

ASML is the only commercial supplier of EUV lithography systems. Its delivery schedule, service capacity, and platform maturity therefore constrain every leading semiconductor manufacturer considering High-NA adoption.

The company reported eight High-NA shipments through the end of 2025, with six systems operating. That installed base provides a foundation for development, but it remains small beside the mature conventional EUV fleet.

A limited population of production tools can slow learning. Engineers need operating hours across multiple sites to identify reliability issues, improve components, and standardize process methods.

SK hynix’s 2028 deadline consequently depends on ASML reaching its end-of-2026 high-volume manufacturing readiness target. Delays would compress the period available for qualification and yield improvement.

Samsung applies pressure from the other side. Its own 2028 DRAM target means SK hynix cannot postpone adoption solely to protect near-term free cash flow. A competitor that achieves stable High-NA manufacturing first might gain a process-complexity or density advantage.

However, a simultaneous target does not guarantee an identical deployment. Samsung and SK hynix can choose different layers, DRAM generations, process flows, and ramp volumes. The first public production date will reveal less than the resulting yield and cost.

Micron represents another route. Its manufacturing progress can test whether aggressive High-NA adoption is essential. If Micron remains competitive through conventional EUV and other process changes, investors may question the urgency of early High-NA spending.

The comparison also extends to shareholder policy. Micron’s commitment to return 100 percent of free cash flow raises expectations for capital efficiency across the memory sector.

That commitment should not be copied without considering different investment schedules and balance sheets. Still, it gives investors evidence that a memory manufacturer can pair large technology spending with a clearer distribution formula.

SK hynix must therefore satisfy two constituencies that evaluate different clocks. Customers care about reliable HBM and DRAM supply across future product cycles. Shareholders care about whether extraordinary current profits become durable value today.

The company cannot serve customers by stripping its investment budget. It also cannot persuade investors by treating every future technology possibility as a reason to retain cash.

The Korea discount adds urgency. This term describes the lower valuations often assigned to South Korean companies because of concerns including governance, capital allocation, and shareholder treatment.

A transparent payout policy would address one part of that valuation gap. A credible High-NA roadmap would address another by showing that investment supports defensible manufacturing capability.

Those objectives can reinforce each other. Stronger technical execution creates cash, while a predictable return framework assures investors that successful execution will reach them.

The conflict emerges when neither side is measurable. An unspecified technology budget paired with an unspecified future payout leaves management maximum discretion but gives investors minimal visibility.

That is the outcome SK hynix needs to avoid. Its next capital-return announcement should explain how the 2028 manufacturing plan fits within a defined investment ceiling and surplus-cash mechanism.

Three Signals Will Show Whether the 2028 Promise Holds

The next evidence should come from capital policy first, manufacturing milestones second, and competitive validation third.

The first signal is SK hynix’s promised update on shareholder returns. Investors need more than a statement that distributions can expand. They need a percentage, formula, timetable, or repurchase commitment that can be measured against free cash flow.

A policy above the current 50 percent benchmark would strengthen the argument that SK hynix can fund advanced manufacturing without building an oversized cash reserve. A temporary special distribution would provide cash but offer less evidence of lasting discipline.

The second signal is a concrete High-NA qualification milestone. SK hynix should identify when equipment enters its development line, when test wafers begin, or when a specific DRAM process completes qualification.

That disclosure would turn the SK hynix High-NA EUV plan into a sequence of testable steps. A schedule that remains limited to “by 2028” would preserve uncertainty about the depth of preparation.

Investors should also watch whether the company defines the first use narrowly. A limited number of critical layers would suggest a controlled insertion designed to reduce multipatterning where the economics are strongest.

A broad deployment claim without throughput or yield information would deserve more skepticism. High-NA is valuable only when better resolution produces an acceptable manufacturing result.

The third signal is evidence from Samsung, Micron, Intel, and ASML. Samsung’s progress will show whether the 2028 DRAM date reflects an industry transition or an aspirational race.

Micron’s process competitiveness will indicate whether manufacturers can postpone High-NA without losing ground. Intel’s production experience can expose reliability and integration issues, while ASML’s shipment and uptime figures will show whether the platform is ready to scale.

These signals should be read together. A better payout framework without technical milestones could indicate underinvestment. Rapid equipment spending without better capital rules could deepen investor concerns.

The stronger outcome is a paired commitment. SK hynix funds a documented path to 2028, then returns cash beyond that requirement through a predictable mechanism.

For semiconductor customers, this determines whether future memory capacity arrives with stable yields and manageable costs. For developers and AI product teams, it affects the supply, performance, and availability of the memory behind accelerators and data-center systems.

For investors, the question is more immediate. Does SK hynix treat record AI profit as permanent capital under management, or as value that must be divided transparently between future production and present owners?

Watch the next shareholder policy, the first named High-NA qualification milestone, and Samsung’s competing 2028 progress. Together, they will show whether SK hynix has established a disciplined manufacturing plan or merely placed two ambitious promises on different calendars.