HPSP (에이치피에스피, 403870, KOSDAQ) runs a business that, in theory, should not persist in the equipment industry: it is the world’s only maker of high-pressure hydrogen annealing (HPA) tools to have passed semiconductor production qualification. Its publicly disclosed customer list includes Samsung Electronics, SK hynix, and TSMC at the same time, and its operating margin has held between 51.8% and 53.5% for four consecutive fiscal years. That is the financial fingerprint of a real monopoly.
In 2026 the stock doubled: from a January 2 close of 34,450 won to a June 12 close of 71,500 won, up 107.5% year to date, against an 8.8% gain for the KOSDAQ index. Over the same period, earnings moved the other way. FY2025 revenue fell 4.7% year over year, the first decline in the four fiscal years on record, and Q1 2026 revenue fell another 13.4%. On FY2025 earnings, the trailing P/E expanded from roughly 39x at the start of the year to roughly 81x — still about 71x after rolling in the disclosed first quarter. Nearly all of the gain came from the multiple, not from earnings.
And the monopoly itself is on trial. On May 28, 2026, Korea’s Intellectual Property Trial and Appeal Board ruled one HPSP patent invalid, handing a win to the challenger Yes Tech (예스티). The case the market treats as decisive — on the core patent that underpins exclusivity — will be ruled on by the Patent Court in mid-June. Meanwhile, the largest shareholder, Crescendo Equity Partners, has cut its stake from about forty percent to about 20.11% through two block deals this year, cashing out over 600 billion won, with the exit still in progress.
Earnings falling, the monopoly under judicial review, the largest shareholder selling — and the market paid double. This piece asks what the 81x multiple is actually buying: a long-dated option on 2nm/GAA penetration, or a bet on a patent verdict.
On June 12, HPSP closed at 71,500 won, up 30% on the day — pinned at the KOSDAQ daily price limit — and at a new year-to-date high. Volume reached 16.76 million shares, roughly 3 to 5 times the recent daily average. The trigger: Korea’s Patent Court announced it would rule on all three patent cases between HPSP and Yes Tech together in mid-June, with news of accelerating 2nm investment at TSMC, Samsung, and Intel landing the same day.
The timing is the notable part. The limit-up came two weeks after HPSP lost a round in the patent war — on May 28, the Intellectual Property Trial and Appeal Board ruled its patent #0766303 (a high-pressure gas heat-treatment method and apparatus with a double-wall structure) invalid. And the schedule itself was not new information that day — it had been in the press since mid-April (ZDNet Korea, April 15, 2026). Reading the limit-up as a pure bet on a patent win is too strong an attribution; it looks more like several expectations stacking: pre-verdict positioning for continued exclusivity, plus the same day’s news of accelerating 2nm investment. The only certainty is that buyers chose to pay before the ruling lands.
Flows confirm the turn. From May 27 to June 2, institutions were continuous net sellers, dumping a cumulative 1.98 million shares; from June 9, that reversed on the patent-litigation news. Over the 20 trading days through June 12, institutions were net buyers of about 334,441 shares and foreign investors net buyers of about 2,516,665 shares. Foreign ownership rose from 30.78% on May 14 to 34.15% on June 12, up 3.37 percentage points in a month.
Our earlier piece, Korea’s Memory Industry: One Country, Two Cycles, looked at the chipmaker side. This piece moves upstream: in the 2026 AI rally, how does the market price the equipment suppliers to the chipmakers? HPSP is an extreme specimen. The monopoly is real. So are the cracks.
HPSP’s flagship product is the GENI-SYS series of high-pressure hydrogen annealing tools, which account for more than 90% of company revenue. The process heat-treats wafers in a high-pressure hydrogen environment, repairing interface-state defects at advanced nodes and improving device reliability. According to company IR materials and industry media including The Elec and ZDNet Korea, HPSP has long been the only HPA tool maker in the world with production qualification — a position described as a de facto monopoly. Its main competitor, Yes Tech, only formally entered the market at the end of 2025 and is at the earliest stage of market entry.
A monopoly does not need to declare itself; the financial statements do the talking. From FY2022 through FY2025, HPSP’s operating margins were 53.5%, 53.2%, 51.8%, and 52.0% — four years pinned between 51.8% and 53.5%. That level has almost no peer in the equipment industry. It means customers essentially do not negotiate on price; pricing power sits firmly with the seller.
The customer list explains the demand side. Publicly disclosed customers include Samsung Electronics, SK hynix, and TSMC (The Elec, May 29, 2026); company IR and analyst reports also place Intel and Rapidus in the adoption phase (Lead Economy, 2025). In effect, every major player pushing leading-edge logic is using the machine or preparing to.
GAA is the core of the valuation story. TSMC’s N2 uses a GAA nanosheet structure (originally planned to enter volume production in the second half of 2025), and Samsung’s 3nm/2nm likewise run on GAA — both are significant HPSP buyers. The HPA step is described by company IR and industry media as especially critical in GAA architectures. Push that logic to its end and you have the entire bull case: leading-edge logic’s migration to 2nm/GAA is a structural trend, and an exclusive supplier’s tool demand scales with penetration. Analyst consensus expects 2026 revenue of about 234.1 billion won and operating profit of about 124.5 billion won, roughly 35% and 39% year-over-year growth respectively (Lead Economy, 2025).
The problem is that the story’s three pillars — monopoly, growth, and the share register — all cracked in the first half of 2026.
“Monopoly on trial” is meant literally. HPSP sued Yes Tech for patent infringement in 2023. In October and November 2024, the Intellectual Property Trial and Appeal Board twice ruled in HPSP’s favor — upholding the validity of its patent and rejecting Yes Tech’s scope-confirmation claim; but the two most recent rounds have both gone to Yes Tech.
In April 2025, Yes Tech prevailed in a scope-confirmation trial, which confirmed that its locking-mechanism design does not fall within HPSP’s patent claims. On May 28, the Intellectual Property Trial and Appeal Board ruled HPSP patent #0766303 invalid — a second win for Yes Tech (ZDNet Korea, May 29, 2026 report; The Elec).
Next comes the core case: the validity of HPSP patent #1553027 (a chamber opening-and-closing mechanism), which the Patent Court will rule on in mid-June together with the other two cases. Per the reporting of ZDNet Korea, the market treats this case as the verdict on whether exclusivity survives. Set against the facts above, though, neither direction works like a switch. If HPSP wins, what it defends is the remaining legal barrier — the April 2025 scope ruling has already confirmed that Yes Tech’s redesigned mechanism sits outside this patent’s claims, so the design-around path does not close just because the patent stands. If Yes Tech wins again, what opens is a time window — between legal entry and commercial orders lies a long customer qualification cycle (the next section shows that converting qualification into orders is precisely this business’s own pain point). The ruling will change how fast competition arrives, not flip it on or off that day.
Note the relationship between this crack and the valuation: part of the June 12 limit-up is the market pricing this verdict in advance. The ruling is binary; its consequences are not — and the market has already picked a side.
The flip side of the monopoly is that revenue growth is more fragile than it looks. HPSP’s top line traces a clean deceleration curve: FY2022 to FY2023, up 12.4% (159.34 billion won to 179.09 billion won); FY2023 to FY2024, up just 1.3% (181.40 billion won — effectively stagnant); FY2024 to FY2025, down 4.7% (172.96 billion won) — the first decline in the four fiscal years on record. Operating profit moved in step, falling 4.3% in FY2025 to 89.90 billion won. The first quarter of 2025 extended the downtrend: revenue of 36.90 billion won, down 2.3% year over year; operating profit of 18.71 billion won, down 6.3%.
Net income fell harder (down 15.8% in FY2025 to 72.66 billion won; down 28.8% in Q1), but that is mostly a base effect: FY2024 included roughly 14.4 billion won of one-off other income versus only about 2.2 billion won in FY2025, and Q1 had the same pattern (about 6.54 billion won of other income in the prior-year quarter). The core operating deterioration is not as steep as the net income line suggests.
Nor did the downtrend stop at the FY2025 annual report. The quarterly report disclosed on May 15 shows Q1 2026 revenue of 31.94 billion won, down 13.4% year over year — a steeper drop than FY2025’s full-year 4.7%, and a second consecutive first-quarter contraction; operating profit of 16.14 billion won, down 13.7%, with the operating margin still at 50.5%. Net income of 24.86 billion won was up 63.6% year over year — but the rebound is mostly tax: the quarter booked an income-tax benefit of 5.73 billion won against a 4.84 billion won tax expense a year earlier, and pre-tax profit still fell 4.5%. Samsung Securities, which covers the stock, called the quarter in line, noted the decline from major customers’ fab schedules had been anticipated, and placed the recovery case on broader customer and application adoption in the second half (DART quarterly report, May 15, 2026; Samsung Securities 1Q26 review, May 15, 2026). For this section’s argument, the quarter extends the line rather than bending it: pricing power intact, new orders still absent.
The real issue is the ordering mechanism. Per Lead Economy’s attribution: major logic customers reused existing tools when switching technology nodes rather than buying new units, and DRAM/NAND qualification cycles ran longer than expected — “qualified” has not converted into orders. This is a counterintuitive fragility inside a monopoly business: customers have no alternative supplier, but customers can simply not buy new machines. An equipment monopoly can lock in price; it cannot lock in purchase timing.
That clarifies what the expected 37% growth in 2026 actually requires: node migration must genuinely trigger new tool purchases, and qualification must genuinely convert into orders — precisely the two things that did not happen over the past two years.
The third crack is the share register. The private equity firm Crescendo Equity Partners held about forty percent of HPSP at its peak. Since the start of 2026 it has completed two block deals: in early January, it sold about 10% of the company for roughly 300 billion won, applying the proceeds to acquisition financing (Asia Economic Daily, January 7, 2026); on February 23-24, it sold another roughly 9.05% at a 4.9-7.6% discount, at average prices of 41,600 to 42,800 won, for roughly 325.3 billion won, with the stock falling as much as 7% that day (Seoul Economic Daily, February 24, 2026; Douglas Research). The two rounds cashed out roughly 610-625 billion won in total (media tallies differ slightly), taking the stake down to about 20.11%.
And this is not passive trimming; it is a planned exit. Per Seoul Economic Daily reporting, Crescendo has been working toward a full exit since late 2024 with UBS as lead bookrunner. On June 10, the company announced the convening of an extraordinary general meeting, which the market reads as related to the ownership change and a board reshuffle — the specific agenda items remained undisclosed as of this data snapshot.
There is a price anchor buried here. In early 2025, Carlyle, Blackstone, and MBK Partners bid for roughly 40% of HPSP. Crescendo reportedly asked for about 2 trillion won — effectively demanding a 100% premium — and the talks collapsed (PE Insights). Simple arithmetic on that ask: 40% for 2 trillion won implies a whole-company valuation of about 5 trillion won. In other words, a little over a year ago, the most sophisticated buyers in the market for corporate control collectively decided an implied 5 trillion won was not worth paying. Today the public market’s capitalization is 5.88 trillion won — above the ask that found no takers.
Put the three cracks back into the valuation and the pricing structure becomes visible.
At the start of the year, the market put HPSP on a trailing P/E of about 39x (a market cap of about 2.83 trillion won against FY2025 net income of 72.66 billion won). On June 12, the same earnings carried a multiple of about 81x (a market cap of 5.88 trillion won on 82.3 million shares outstanding); roll in the disclosed first quarter of 2026 and trailing-twelve-month net income is about 82.3 billion won, putting the multiple at about 71x — with that quarter’s net rebound resting mostly on a tax benefit. On either basis, what improved was not operating earnings; only the multiple the market is willing to pay changed. If the 2026 consensus is delivered — revenue of about 234.1 billion won, operating profit of about 124.5 billion won — the roughly estimated forward P/E still sits in the 45-50x range. Growth has to arrive on schedule for the valuation to digest even to that level.
So the market is not buying this year’s profits. It is buying a compound option: monopoly preserved (patent win) multiplied by GAA penetration realized (new nodes triggering purchases). If both legs hold, 81x can be digested by growth over time. If the patent holds but growth does not, the valuation is left hanging. If the patent falls, the pricing basis for the exclusivity premium starts to give way — though, as above, qualification cycles stand between a lost case and commercial competition: the premium erodes as a process, not a switch. The June 12 limit-up says that, at least until the verdict, money has left little discount in the price for the losing scenario.
For the record: foreign investors are the net buyers at this level — ownership up 3.37 percentage points in a month, with net purchases of about 2,516,665 shares over 20 trading days, well above institutions’ roughly 334,441 shares.
Most of the bear case requires no forecasting. It is already on the record.
First, the momentum in the patent war is with the challenger. HPSP won two favorable rulings in 2024, but the two most recent rounds — the April 2025 scope-confirmation trial and the May 2026 invalidation — both went to Yes Tech. The scope ruling also means that even if HPSP wins in mid-June, Yes Tech’s design-around already sits outside the patent’s claims — the payoff to a win is smaller than the entry-sealed narrative suggests. No one can know the mid-June outcome in advance — and the June 12 limit-up implies the losing scenario is currently barely priced at all.
Second, the growth engine has been stalled for two years. Up 1.3% and down 4.7% are not projections; they are numbers on DART filings. What the bulls must explain is not whether GAA will penetrate, but why penetration produced no orders for two years and will in 2026.
Third, the best-informed shareholder is selling. Crescendo has held HPSP for years and sat in the boardroom information seat, and it chose to cut nearly half its stake in two rounds this year, with the exit channel for the remaining roughly 20.11% (UBS as bookrunner) still open. Layer on the roughly 5 trillion won private-market ask: the public market’s current price is above what control buyers collectively walked away from a little over a year ago.
Fourth, structural exposure. About 92% of revenue comes from exports (Lead Economy), concentrated in a handful of leading-edge fabs; the US MATCH Act’s export controls on China could close off expansion into the Chinese market. Customer concentration plus geopolitical constraint means any single customer’s capex timing flows straight through to revenue.
Fifth, the valuation itself. Pushing the trailing P/E to 81x against falling earnings means double optimism — on a growth restart and on a court ruling — is already in the price. The bears do not need to argue the company is bad. The monopoly, the customer list, and the margins are all real. They only need to point out one thing: at the start of this year, the market priced the very same earnings at 39x.
There is no third-party market research giving absolute figures for the HPA market size or HPSP’s share; the de facto monopoly is the framing of company IR and media, not a quantified conclusion. There is no public data on Yes Tech’s actual order volumes or customers since its late-2025 entry. There is no official figure for how many HPSP tools serve TSMC’s advanced nodes. Whether Applied Materials, Lam Research, or other global tool makers are building HPA products: this research found no clear evidence of entry, but also no clear denial — the largest unverified assumption inside the monopoly narrative. There is no current disclosure on the pace of Crescendo’s remaining selldown. FY2026 is disclosed through the first-quarter report (May 15, 2026); Q2 and beyond are pending. Customer concentration detail (top-five customer shares) was not extracted from the annual report notes. The extraordinary general meeting’s agenda is still to be disclosed.
These gaps point to one reminder: this piece describes how the market is pricing this company, not what the company is worth. The former has data. The latter, until the mid-June verdict, is probabilities for everyone.
This report is an independent KSINQ market observation for informational purposes only. It is not investment advice. Data snapshot: June 13, 2026; prices as of the June 12, 2026 close.