Search HPSP in English and you will read that it is the world’s only maker of high-pressure hydrogen annealing equipment β a monopoly hiding inside the chip supply chain. Then you open the company’s own Korean filing, and it says something quieter and stranger: its market share cannot be calculated. Both sentences are in the public record. Only one of them is in English.
π Key Takeaways
- HPSP sells one tool, and 89% of revenue is that tool. Its GENI-SYS machine anneals chip wafers in pressurised hydrogen β the filing’s spec range is 1β25 atm β to repair defects at the high-ΞΊ gate interface. Equipment was 89.2% of H1 2026 revenue, parts 6.4%, service 4.4% (Source: DART half-year report, filed Aug 14, 2026).
- The company says it has no competitor β and its own filing says its share can’t be measured. The August 2026 filing states “no competing product exists” and, separately, that “a segmented market-share calculation is impossible.” Those are different claims, and English coverage collapses them into “100%.”
- Revenue fell 11.5% and operating profit fell 6.1% in the half β but net income rose 36.4%. The gap is tax, not trading: the income-tax charge dropped from $8.19M to $0.76M.
- The controlling shareholder sold nearly half its stake in seven weeks. Crescendo’s vehicle went from 32,800,000 shares to 16,840,000 across two January and February 2026 block sales β filed reason: “recovery of investment funds.”
- There is no US listing. We found no ADR across OpenFIGI, SEC EDGAR (including Form F-6) and Yahoo Finance. Foreign buyers use the Korean line.
π’ What Is HPSP, and What Does It Actually Sell?
HPSP is a 136-person Korean company that makes a single category of semiconductor front-end tool β a furnace that bathes wafers in high-pressure hydrogen to heal defects in the transistor gate β and sells it to a handful of the world’s leading chipmakers.
The problem it solves is specific. Below the 28-nanometre node, chipmakers replaced silicon dioxide gate insulation with high-ΞΊ dielectric material (hafnium oxide) to stop electrons tunnelling through. It works β permittivity is more than five times higher than SiOβ β but it carries a penalty the company states plainly: interface defects rise 100-fold (Source: DART half-year report, rcpNo 20260814004015, filed Aug 14, 2026).
HPSP’s answer is pressure. Its GENI-SYS tool runs hydrogen or deuterium at up to 25 atmospheres β the filing gives the equipment range as “1 ~ 25 ATM” and describes operating the process at 20 atm β which lets the anneal happen at 250β450Β°C instead of the 600β1,100Β°C a conventional furnace needs β and above 600Β°C the metal gate and interconnect degrade, so the conventional tool cannot be used at all on advanced nodes. The filing’s comparison table is blunt: high-temperature annealing is “16γ μ΄ν 곡μ μ μ©λΆκ°λ₯”, not applicable below 16nm. HPSP labels that table’s source “λΉμ¬ λ΄λΆμλ£” β company internal data β which is worth remembering.
Founded on 14 March 2017, HPSP is not a startup that invented this from nothing. Five weeks after incorporation it bought the entire equipment division of Poongsan Corporation, the Korean copper and munitions group β board and shareholder resolution 5 April 2017, transfer completed 20 April 2017. It is headquartered in Hwaseong, Gyeonggi Province, employs 136 people as of 30 June 2026, and lists on KOSDAQ, Korea’s growth market. Its own English site claims something more modest than the coverage around it: that it is “leading the semiconductor ultra-fine process market by commercializing the high-pressure hydrogen annealing technology, which had existed only in theory.”
π Where Does HPSP’s Revenue Actually Come From?
Almost entirely from selling machines abroad: equipment was 89.2% of H1 2026 revenue and customers outside Korea were 84.8% of it. HPSP reports as a single operating segment β “νμ¬ μ 체λ₯Ό λ¨μΌ λ³΄κ³ λΆλ¬ΈμΌλ‘ κ²°μ νμμ΅λλ€” β so the product and geography splits below come from the revenue note, not from segment accounts.
| Revenue split (half-year, cumulative) | H1 2026 | % of total | H1 2025 | % of total |
|---|---|---|---|---|
| μ₯λΉλ§€μΆ β Equipment | $51.7M | 89.2% | $58.5M | 89.3% |
| λΆνλ§€μΆ β Parts | $3.7M | 6.4% | $5.1M | 7.7% |
| μ©μμμ΅ β Service | $2.6M | 4.4% | $2.0M | 3.0% |
| λ³Έμ¬ μμ¬μ§ κ΅κ° β Korea | $8.8M | 15.2% | $1.4M | 2.1% |
| μΈκ΅ β Outside Korea | $49.2M | 84.8% | $64.2M | 97.9% |
| Total | $58.0M | 100% | $65.6M | 100% |
Korean row labels reproduced from the filing. Revenue attributed by customer domicile, per the note. Source: DART half-year report rcpNo 20260814004015, note 26 (μμ λΆλ¬Έ μ 보), reporting period 1 Jan β 30 Jun 2026, separate (λ³λ) basis. FX: β©1,346/$1 as of Sep 4, 2026 β applied to every dollar figure in this article.
The domestic share went from 2.1% to 15.2% in a year. That is not a rounding change; Korean-domiciled revenue rose roughly sixfold in absolute terms while total revenue fell. Which leads to the more interesting table.
The customer table where the labels and the arithmetic disagree
HPSP discloses four major customers by letter. The disclosure attaches the identical label to all four columns in both periods β “νμ¬ λ§€μΆμ‘μ 10% μ΄μμ μ°¨μ§νλ μΈλΆ κ³ κ°” (external customers accounting for 10% or more of revenue). We have reproduced the label exactly as filed, and put the arithmetic beside it, because in the current half three of the four are nowhere near 10%.
| μ£Όμ κ³ κ° (major customers) | H1 2026 | % of revenue | H1 2025 | % of revenue |
|---|---|---|---|---|
| κ³ κ° A | $15.8M | 27.3% | $0 | 0.0% |
| κ³ κ° B | $4.2M | 7.2% | $24.4M | 37.2% |
| κ³ κ° C | $0.56M | 1.0% | $12.9M | 19.6% |
| κ³ κ° D | $0.20M | 0.4% | $6.6M | 10.0% |
| Four disclosed customers | $20.8M | 35.8% | $43.8M | 66.8% |
Percentages are ours, computed against total revenue of β©78,062,951k (H1 2026) and β©88,242,097k (H1 2025). Source: DART rcpNo 20260814004015, note 26, μ£Όμ κ³ κ°μ λν 곡μ.
Read the columns, not the label. A year ago three customers each cleared 10% and together were two-thirds of revenue. This half, one customer is 27.3% and the other three total 8.5%. The concentration did not intensify β it rotated, and disclosed named-customer revenue fell from 66.8% of the total to 35.8%.
One caution we will not paper over: the filing does not state that “κ³ κ° A” in 2026 is the same company as “κ³ κ° A” in 2025. Letters in these disclosures are not guaranteed to be stable across periods. So the honest reading is that the shape of the customer base changed sharply β one large customer where there were three β and that the jump in Korea-domiciled revenue is consistent with a Korean memory maker becoming the largest buyer. We cannot confirm identities from the filing, and we are not going to guess them.
ποΈ How Did HPSP Get Here?
HPSP is a private-equity carve-out: Crescendo Equity Partners bought Poongsan’s equipment division in 2017, listed it on KOSDAQ in 2022, and spent 2026 selling down.
| Date | Event |
|---|---|
| Mar 2017 | HPSP Co., Ltd. incorporated (14 March) |
| Apr 2017 | Acquires Poongsan Corporation’s entire equipment-business division |
| 2019β2021 | Production-tool deliveries to global system-semiconductor and multiple memory customers; named a MOTIE “World-Class Product” in 2017 and 2021 |
| Jul 2022 | KOSDAQ listing |
| Mar 2023 | 3-for-1 bonus issue (17 March): 60,869,250 new shares, taking the count from 20,289,750 to 81,159,000 |
| End-2023 | Share count 82,844,274 β the 1,685,274 above the post-bonus figure is stock-option exercises during the year |
| Mar 2024 | Moves to new Dongtan headquarters, Hwaseong |
| May 2025 | Crescendo’s Presto PEF No.6 contributes its 32,800,000 shares in kind to a new vehicle, ννΈ2025νλ©μ€ β registered largest shareholder changes, control does not |
| JanβFeb 2026 | Two block sales cut that vehicle’s holding from 32,800,000 to 16,840,000 shares |
| Mar 2026 | 1,699,120 treasury shares cancelled (13 March) |
| Jun 2026 | Extraordinary general meeting appoints a new inside director, named CEO the same day |
Sources: DART rcpNo 20260814004015 (μ°ν, μ΅λμ£Όμ£Όμ λ³λ, μλ³ΈκΈ λ³λμ¬ν); rcpNo 20260319000776 (FY2025 annual report).
π Is HPSP Really the Only Company That Makes This Tool?
HPSP’s own filing says no competing product exists β and a rival’s filing, submitted to the same regulator on the same day, lists a high-pressure annealing tool in its product table. Both statements are in the public record for 14 August 2026.
Start with what HPSP actually wrote, because the English-language version of this company has drifted a long way from it. Under the heading “(3) μμ₯μ μ μ¨ μΆμ΄” β market-share trend β the half-year report says, in full:
λ³΄κ³ μ μ μΆμΌ νμ¬ λ°λ체 κ³ μ μμ μ΄λλ§ μ₯λΉλΆλΆμ κ²½μ° λΉμ¬κ° λ 보μ μ§μλ₯Ό λλ¦¬κ³ μκΈ° λλ¬Έμ μ μ¬ μ μ’ νμ¬μ 곡μμλ£ λΆμ‘± λ±μΌλ‘ μΈλΆνλ μμ₯μ μ μ¨ μ°μΆμ λΆκ°λ₯ν©λλ€.
“As of the report submission date, because the company enjoys an unrivalled position in the semiconductor high-pressure hydrogen annealing equipment segment, a segmented market-share calculation is impossible owing to the lack of disclosure data from comparable companies.”
Elsewhere the same filing goes further: “λΉμ¬μ GENI-SYS μ νμ λ³΄κ³ μ μ μΆμΌ νμ¬ κ²½μ μ νμ΄ μ‘΄μ¬νμ§ μλν©λλ€” β as of the submission date, no competing product exists.
Now hold that against a second filing. YEST Co., Ltd. (KOSDAQ: 122640) filed its own half-year report on 14 August 2026 β the same day. Its product table lists “κ³ μ μ΄λλ§ μ₯λΉ”, application “λ°λ체 μ κ³΅μ ” (semiconductor front-end), purpose “κ³ μμμ(μ€μμ) μ΄μ²λ¦¬λ₯Ό ν΅ν΄ κ³λ©΄μμ κ²°ν¨μ μΉμ ” β healing interface defects by high-pressure hydrogen/deuterium heat treatment. That is a description of the same process, in a competitor’s own regulatory filing, dated the same day (Source: DART rcpNo 20260814000104).
The same YEST filing discloses pending patent-infringement litigation brought by HPSP in the Seoul Central District Court, at first instance, with a claim amount of β©100,000,000 (~$74,300). Companies do not usually sue products that do not exist.
We deliberately do not give a case count. The litigation appears in YEST’s filing four times β twice in the consolidated notes and twice in the separate notes, each pair covering the current half and the prior year-end β and the rows within each table are identical across all six fields, so the repetition is a presentation artefact of the two reporting bases rather than a count of cases. Reading it as “four suits” double-counts the same disclosure along two axes; even “two” cannot be established from the filing.
The claim amount is worth noting precisely because it is small. β©100,000,000 is roughly $74,300 β nominal for a company with $58.0M of half-year revenue, which suggests litigation aimed at establishing infringement rather than at recovering damages. And there is a striking asymmetry in how the two companies treat it: YEST’s own “significant litigation” section states “μ€μν μμ‘μ¬κ±΄ λ± κ³΅μλμ κΈ°κ° μ€ ν΄λΉμ¬ν μμ΅λλ€” β no significant litigation during the reporting period. HPSP is suing; YEST does not consider it significant. That temperature difference is itself a fact.
We want to be careful about what this does and does not establish, because the temptation to overclaim runs both ways:
- What is verified: a rival lists an HPA tool for the semiconductor front-end in its statutory filing, and HPSP has patent-infringement litigation pending against it at first instance.
- What the filing will not support: a number of cases. The same disclosure is repeated across consolidated and separate notes and across two period-ends, so any count read off it is an artefact of the presentation.
- What is not: we found no disclosure of HPA revenue at YEST β its filing does not break out sales for that line. Listing a product is not the same as shipping it in volume into a qualified production fab, which is the bar HPSP’s claim is really about.
- Reported, not verified by us: The Elec reported on 19 June 2026 that Korea’s Patent Court ruled YEST’s locking mechanism falls outside the scope of HPSP patent No. 1553027. That is a separate proceeding from the district-court litigation described above, which YEST’s August filing still shows at first instance.
- What we could not find: any credible report of a Japanese, US or Chinese equipment maker building or shipping a high-pressure hydrogen anneal tool. We searched for it and did not find it β which is a statement about our search, not about the world.
So the accurate position is narrower and more useful than “monopoly.” HPSP has a commanding technical lead, a fifteen-year head start, and β by its own account β no qualified production alternative. It also has a litigating competitor with a product on the books, and it has told its regulator that nobody, including HPSP, can put a number on its share. An investor should hold all three of those at once.
en.wikipedia.org/wiki/HPSP is a US military scholarship programme, and a Wikipedia search API query for the company returns zero hits.
π Is the Business Getting Better or Worse?
On operations, worse: half-year revenue fell 11.5% and operating profit fell 6.1%. Margins went the other way, and net income rose 36.4% β for a reason that has nothing to do with selling tools.
A note on how to read this table, because it is where these filings trip people up. Korean interim statements print a 3κ°μ (three-month) column beside a λμ (cumulative) column, and the two are easy to confuse. Every figure below is labelled. HPSP also files no consolidated statements at all β the report states “2. μ°κ²°μ¬λ¬΄μ ν: λΉμ¬λ ν΄λΉμ¬νμ΄ μμ΅λλ€”, and its investment-valuation line reads ν΄λΉμ¬ν μμ, meaning there are no subsidiaries to consolidate. Everything here is therefore separate (λ³λ, OFS) basis.
| Separate (λ³λ) basis | Q2 2026 3κ°μ |
Q2 2025 3κ°μ |
H1 2026 λμ |
H1 2025 λμ |
YoY (λμ ) |
|---|---|---|---|---|---|
| Revenue | $34.3M | $38.1M | $58.0M | $65.6M | β11.5% |
| Operating profit | $21.0M | $21.3M | $33.0M | $35.2M | β6.1% |
| Operating margin | 61.4% | 55.8% | 56.9% | 53.7% | +3.3pp |
| Pre-tax profit | $22.9M | $20.0M | $37.1M | $34.9M | +6.5% |
| Income tax charge | $5.0M | $4.6M | $0.76M | $8.19M | β90.7% |
| Net profit | $17.9M | $15.4M | $36.4M | $26.7M | +36.4% |
| Basic EPS | $0.22 | $0.19 | $0.45 | $0.33 | +35.3% |
| Basic EPS (as filed) | β©296 | β©257 | β©602 | β©445 | +35.3% |
Reporting periods: 3κ°μ = 1 Apr β 30 Jun; λμ = 1 Jan β 30 Jun. Source: DART rcpNo 20260814004015, 4-2. ν¬κ΄μμ΅κ³μ°μ, separate basis. Cross-checked against the DART structured accounts API (thstrm_amount vs thstrm_add_amount), which returns identical figures.
For longer context: FY2025 revenue was $128.5M (β4.7% on FY2024’s $134.8M), operating profit $66.8M (β4.3%), net profit $54.0M (β15.8%). So HPSP has now posted two consecutive down years on revenue, with the half-year continuing the trend. Operating margin, meanwhile, has barely moved at the full-year level β 52.0% in FY2025 against 51.8% in FY2024 β which makes the half-year jump to 56.9% the thing worth explaining.
Part of the explanation is that costs fell faster than sales: SG&A dropped 20.5% (from $13.0M to $10.4M) against an 11.5% revenue decline. But the statement of changes in equity shows why that is not a durable efficiency story. The share-based payment line swung from +β©3.06bn in H1 2025 to ββ©6.10bn in H1 2026 β a negative figure, meaning a reversal of previously booked stock-compensation cost. That is a β©9.16bn ($6.81M) swing, larger than the entire β©3.58bn fall in SG&A. R&D shows the same fingerprint: the “κΈ°ν” (other) line inside the R&D cost table is negative β©1,144,626k for the half, and R&D fell to 3.65% of revenue from 7.39% in FY2025.
We cannot split the equity-statement figure between cost of sales and SG&A from a half-year filing, so we will not claim a precise attribution. What we can say is that the margin improvement is not obviously operational, and a reader who annualises 56.9% is annualising a reversal.
π Why Did Net Income Jump 36% While Operating Profit Fell?
Because the income-tax charge collapsed by 90.7% β and HPSP’s own tax note gives a rate that does not reconcile with the face of its income statement.
The face of the statement is unambiguous and triple-confirmed. Cumulative pre-tax profit of β©49,998,279,020 carried a cumulative tax charge of β©1,029,339,795, leaving net profit of β©48,968,939,225. That arithmetic ties exactly to the statement of changes in equity and to reported EPS of β©602, and the DART structured accounts API returns the same numbers. The implied effective rate is 2.06%.
Note 11 of the same filing says something else:
λΉλ°κΈ°μ νκ· μ ν¨μΈμ¨(λ²μΈμΈλΉμ© Γ· λ²μΈμΈλΉμ©μ°¨κ°μ μμ΄μ΅)μ 21.8%μ λλ€.
“The average effective tax rate for the current half (income tax expense Γ· pre-tax profit) is 21.8%.”
The note defines the formula, and the formula applied to the filing’s own numbers gives 2.06%, not 21.8%. To get 21.8% you would need a tax charge of about β©10.9bn β roughly ten times what the income statement reports.
| Effective tax rate, cumulative half | H1 2026 | H1 2025 |
|---|---|---|
| Tax charge Γ· pre-tax profit, from the income statement | 2.06% | 23.50% |
| Rate stated in note 11 (νκ· μ ν¨μΈμ¨) | 21.80% | 23.50% |
| Reconciles? | No | Yes, exactly |
Source: DART rcpNo 20260814004015 β 4-2. ν¬κ΄μμ΅κ³μ°μ and μ£Όμ 11. λ²μΈμΈλΉμ©.
What makes this worth flagging rather than dismissing as a typo is the comparative column. For H1 2025, note 11 states 23.5% and the income statement gives 23.497% β an exact match. The same note, on the same page, prepared the same way, reconciles for the prior year and does not reconcile for the current one.
Two further facts sit alongside it. The quarterly split implies a tax benefit in Q1 2026: the cumulative charge (β©1.03bn) is smaller than the Q2 three-month charge (β©6.76bn), so Q1 must have carried roughly β©5.73bn of credit. And the cash-flow statement’s λ²μΈμΈ λ©λΆ line is a positive β©3,843,397,098 for the half, against β©(16,578,898,678) a year earlier β consistent with a refund received rather than tax paid.
There is a routine mechanism that would explain the gap, and it is worth stating so readers do not jump to something worse. Under IAS 34, interim tax is accrued using an estimated annual effective rate, and 21.8% is a plausible full-year estimate for a Korean corporate taxpayer. The charge actually booked, however, is that accrual adjusted for other items β and note 11 names them, referring to “μ κΈ° λ²μΈμΈμ κ΄λ ¨λμ΄ λΉλ°κΈ°μ μΈμν μ‘°μ μ¬ν” (adjustments relating to the prior year’s tax recognised in the current half). A large prior-year credit would push the booked charge far below the estimated-rate accrual, which is exactly the shape here, and the β©3.84bn tax refund in the cash-flow statement is consistent with it. On that reading the two numbers answer different questions: 21.8% is the forward estimate, 2.06% is what the period actually bore.
We are not alleging an error, and we are not asserting that explanation either β the filing does not spell out the reconciliation, so we could not confirm it. What we are reporting is that the filing contains two figures for the same stated ratio, that we could not reconcile them from the document, and that anyone valuing HPSP on trailing net income should resolve it before doing so. The mechanism itself is not exotic β we have written before about Korean earnings where the headline profit came from below the operating line. Strip the tax move out and HPSP’s pre-tax profit rose 6.5%, on revenue that fell 11.5%.
π° What Do You Actually Pay for HPSP Today?
About 57Γ trailing earnings and 13Γ book for a $3.11bn company β the most profitable business in its peer group, priced in the middle of it.
HPSP closed at β©50,800 (~$37.74) on 4 September 2026, giving a market capitalisation of β©4,180,840,000,000, or $3.11bn (Source: KRX settled close via our coverage dataset, as of 4 Sep 2026). That is 39.2% below its 52-week high of β©83,500, a level it closed at on 15 June 2026, and well above the 52-week low of β©26,900 β a range wide enough that any “trades at” statement needs a date attached to it. (52-week window: 5 Sep 2025 – 4 Sep 2026, closing basis, FinanceDataReader/KRX. The β©26,900 low was set on 5 Sep 2025, the opening day of that window; the 15 June high is confirmed in our own settled-price history, which begins 31 March 2026.)
The share count behind that market cap is 82,300,000, which reconciles exactly with the company’s own disclosure (Source: DART rcpNo 20260814004015, 4. μ£Όμμ μ΄μ λ±).
| Korean semiconductor equipment peers | Market cap | FY2025 revenue | Operating margin | P/E | Basis |
|---|---|---|---|---|---|
| HPSP (403870) | $3.11B | $129M | 52.0% | 57.5Γ | λ³λ OFS |
| Hanmi Semiconductor (042700) | $16.29B | $428M | 43.6% | 102.4Γ | μ°κ²° CFS |
| EO Technics (039030) | $4.05B | $283M | 21.2% | 94.8Γ | μ°κ²° CFS |
| PSK Inc. (319660) | $2.72B | $340M | 19.4% | 46.7Γ | μ°κ²° CFS |
| Eugene Technology (084370) | $2.32B | $260M | 14.7% | 70.3Γ | μ°κ²° CFS |
| Jusung Engineering (036930) | $6.16B | $231M | 10.1% | 232.2Γ | μ°κ²° CFS |
β οΈ Read the basis column. HPSP files no consolidated accounts, so its figures are separate (λ³λ); every peer is consolidated (μ°κ²°). These are not strictly like-for-like, and we have labelled them rather than blending them. P/E = market cap at the 4 Sep 2026 close Γ· FY2025 net profit. PSK’s net profit is μ§λ°°μ£Όμ£ΌλΉκΈ°μμ΄μ΅ (owners of the parent); the others are λΉκΈ°μμ΄μ΅. Sources: DART FY2025 annual reports via the structured accounts API; KRX settled closes, 4 Sep 2026.
The table says something the “monopoly” framing does not. HPSP earns a 52.0% operating margin β the highest in the group, and roughly five times Jusung’s β which is exactly what you would expect from a supplier with no qualified alternative. But it is also the smallest revenue base in the group, at $129M, and it trades at 57.5Γ against a peer set spanning 46.7Γ to 232.2Γ. The pricing is not obviously extreme; the profitability is.
Balance sheet, for completeness: total equity of $232.9M at 30 June 2026 puts the stock at about 13.3Γ book. Cash and cash equivalents were $50.3M with a further $110.9M in current financial assets, against total liabilities of $69.5M. Inventory jumped from $11.9M at end-2025 to $31.1M β a 163% build in six months, which on a tool maker usually signals shipments staged for the second half, and on a tool maker with falling revenue is worth watching both ways.
π·οΈ Which HPSP Ticker Are You Actually Buying?
There is only one β KOSDAQ 403870 β and we found no US-listed ADR for HPSP anywhere.
This is simpler than most Korean deep-dives: no holding company, no preferred line (the 205,000 preferred shares ever issued were all converted to common), no merged-away affiliate to confuse with the operating entity. The complications are elsewhere.
| Identifier | What it is |
|---|---|
| 403870 (KOSDAQ) | The operating company. 82,300,000 common shares in issue, 941,164 held in treasury, 81,358,836 outstanding. |
| 403870.KQ / KR7403870009 | Yahoo Finance symbol and ISIN for the same Korean line. |
| US ADR | None found. OpenFIGI returns no US-exchange instrument; SEC EDGAR full-text search for Form F-6 returns zero; Yahoo returns only the Korean line. HPSP appears in EDGAR only inside fund holdings (e.g. NPORT-P filings), which hold the Korean shares. |
| 122640 (KOSDAQ) | YEST β a different company, and HPSP’s litigation counterparty. Do not confuse the two when reading Korean HPA coverage. |
| ννΈ2025νλ©μ€ / Hit2025 / HEAT 2025 | The largest shareholder, romanised at least three different ways in English sources. Same entity. |
“None found” is a statement about three searches, not a proof of non-existence. Sources: DART rcpNo 20260814004015 (4. μ£Όμμ μ΄μ λ±); OpenFIGI search API; SEC EDGAR full-text search; Yahoo Finance symbol lookup β all checked 5 Sep 2026.
The practical consequence: unlike Samsung Electronics or SK hynix, HPSP cannot be bought through a US over-the-counter line. Foreign investors need direct KOSDAQ access β see our guides to buying Korean stocks as a foreign investor and to the Korean names that do have US-listed ADRs. HPSP is not on that second list.
π Is Foreign Money Buying or Selling HPSP?
Foreigners have bought HPSP across every window we can measure β but institutions and retail investors flip sign depending on which window you pick, and that flip is the story.
| Net buying in HPSP 403870 | 5 sessions Aug 31 β Sep 4 |
20 sessions Aug 7 β Sep 4 |
55 sessions Jun 18 β Sep 4 |
|---|---|---|---|
| HPSP 403870 β Foreign | +$4.3M | +$50.9M | +$141.5M |
| HPSP 403870 β Institutional | +$6.1M | +$83.9M | β$294.6M |
| HPSP 403870 β Retail | β$10.3M | β$133.6M | +$89.1M |
Method: net share counts multiplied by each day’s close, then summed β an approximation, not execution prices. 55 sessions is the full span of our HPSP flow history, not a chosen window. Price moves quoted for these windows are first close to last close within the window; measured from the close before each window opens, the 20-session move is +41.9% rather than +47.7%. Source: our investor_flows dataset, built from KIS Open API daily settled data, through 4 Sep 2026.
Set the flows against what the share price actually did over the same windows, because the two do not point the same way. Over the full 55 sessions HPSP (403870) fell β14.5% β and foreigners bought $141.5M into that decline. Over the most recent 20 sessions HPSP 403870 rose +47.7%, and foreigners kept buying. So foreign buying here is not a momentum signal: it was steady through a drawdown and through the rebound, which is closer to accumulation than to chasing. Institutions did the opposite β they were the ones selling into the fall, and only turned buyers after the stock had already re-rated.
Institutions and retail both reverse sign between the 20- and 55-session windows, and in opposite directions. Over the full span institutions were net sellers of $294.6M while retail absorbed $89.1M; over the last twenty sessions that is exactly inverted. Anyone quoting a single window here can produce “institutions are accumulating HPSP” or “institutions are dumping HPSP” from the same dataset, and both would be arithmetically true.
We are not going to tie the 55-session institutional selling to the controlling shareholder’s block sales, tempting as the shape is β those sales completed in January and February 2026, before this window opens, and our data cannot attribute a flow to a named holder. For the weekly version of this analysis across the whole market, see our Foreign Flow Watch series, and for why these categories move Korean prices at all, our explainer on foreign and institutional flows.
ποΈ Who Controls HPSP, and What Do Minority Holders Get?
A private-equity firm still controls HPSP with 20.46% β down from 39.28% as recently as November 2025 β and minority holders got a 55.66% payout ratio plus a cancellation that quietly raised the controller’s percentage without it buying a share.
The ownership path is fully documented in DART’s major-holding reports, and it is worth laying out precisely because English coverage gives three different romanisations of the same entity and at least two different percentages.
| Report date | Common shares held | Stake | Shares in issue | Filed reason |
|---|---|---|---|---|
| 7 Nov 2025 | 32,800,000 | 39.28% | 83,498,568 | β |
| 12 Jan 2026 | 24,440,000 | 29.23% | 83,610,568 | μκ°μΈλ§€λ β off-hours sale |
| 24 Feb 2026 | 16,840,000 | 20.05% | 83,999,120 | μκ°μΈλ§€λ, ν¬μμκΈ νμλͺ©μ |
| 14 Aug 2026 | 16,840,000 | 20.46% | 82,300,000 | No trade β denominator shrank |
Stake percentages are the μ£ΌκΆ (common-share) column as filed. Sources: DART rcpNo 20260112000224, rcpNo 20260226001249 (μ£Όμλ±μ λλ보μ μν©λ³΄κ³ μ), and rcpNo 20260814004015 (μ΅λμ£Όμ£Όμ λ³λ).
Two things fall out of that table. First, the holder sold 15,960,000 shares β 48.7% of its position β in seven weeks, with the filed reason stated as recovery of investment funds. Second, look at the last row: the stake rises from 20.05% to 20.46% with no purchase at all, because HPSP cancelled 1,699,120 shares on 13 March 2026 and shrank the denominator. 16,840,000 Γ· 82,300,000 = 20.46%, exactly. This is a standard and under-appreciated side effect of Korean cancellations, and one we have seen before in Value-Up era buybacks.
The vehicle itself, ννΈ2025νλ©μ€ μ ννμ¬, is not a new owner. Crescendo’s Presto PEF No.6 fund contributed its HPSP shares in kind to the SPC in May 2025; Crescendo controls the vehicle. In June 2026, HPSP’s extraordinary general meeting appointed a new inside director who was named CEO by the board the same day β consistent with reporting that Crescendo abandoned an outright sale and moved to direct management.
What minority holders actually received
Better than the Korean market average, and disclosed clearly:
- Dividends. FY2025 paid β©500 per share, $30.0M in total, a 55.66% payout ratio. FY2024 paid β©600 per share, $35.8M β so the total dividend fell 16.1% year on year, as HPSP’s own Value-Up filing states. At the 4 September close, β©500 is a 0.98% yield.
- Cancellation. 1,699,120 shares retired on 13 March 2026 β 2.02% of the pre-cancellation count. The board’s filing put the value at β©73,401,984,000 ($54.5M), explicitly computed at the β©43,200 close of 11 February 2026. The retained-earnings charge in the equity statement is a different number, β©53,337,547,800 ($39.6M), being the acquisition cost of the shares. Both are correct; they measure different things.
- Value-Up. HPSP filed its first corporate value-up plan on 30 April 2026 and qualifies as a κ³ λ°°λΉκΈ°μ (high-dividend company) under Article 104-27 of the Restriction of Special Taxation Act (Source: DART rcpNo 20260430901259).
- What the remaining treasury is for. The 941,164 shares still held are earmarked as the funding pool for employee RSUs and options, not for cancellation. The company states that any residue not used for stock compensation by the September 2027 statutory deadline will be cancelled or otherwise disposed of by board resolution.
One structural point in HPSP’s favour, relative to the usual Korea Discount complaints: there is no chaebol pyramid here, no cross-shareholding, and no controlling family. The governance risk is a different one β a financial owner whose job is to exit.
β οΈ The Bear Case β what would make this thesis wrong
- The moat is being defended in court, which is not where an uncontested moat lives. HPSP has patent-infringement litigation pending against YEST at first instance, and per The Elec the Patent Court has already ruled once that YEST’s locking mechanism falls outside the scope of HPSP patent No. 1553027. A position defended by a single patent claim is only as durable as that claim β and the technical distinction at issue (a rotating fastening ring versus rotating the outer door) is the kind of design-around that engineering teams are paid to find.
- The “unrivalled position” is self-reported and unmeasurable by the company’s own admission. The share claim comes from HPSP’s filing and its internal-data comparison table. There is no independent share figure β the filing says one cannot be produced β so an investor is trusting a self-assessment, not a market statistic.
- Revenue has now fallen two years running. FY2024 $134.8M β FY2025 $128.5M β H1 2026 down 11.5% year on year. For a company priced at 57.5Γ earnings, the growth has to come from somewhere, and it has not come from the last eight quarters.
- Earnings quality is doing real work in the headline. Net profit rose 36.4% while operating profit fell 6.1%; the difference is a tax charge that dropped 90.7% and a note that does not reconcile with the income statement. Margin expansion is flattered by a reversal of stock-compensation cost worth more than the entire SG&A decline.
- Customer concentration rotated rather than diversified. One customer is now 27.3% of revenue where three cleared 10% a year ago. A single tool order slipping a quarter moves this company’s reported half-year materially β which is arguably what already happened.
- The controlling shareholder is a seller with 20.46% left. It has already sold 48.7% of its position in two block trades. Whatever the stated strategy, the remaining stake is inventory, and Korean block deals price at a discount β the February sale was reported at β©41,600β42,800 against a β©45,000 prior close.
π Lingo Check
| Term | What it means | νκ΅μ΄ |
|---|---|---|
| High-pressure hydrogen annealing (HPA) | Heat-treating a wafer in pressurised hydrogen or deuterium so hydrogen atoms bond to defects at the gate interface and switch them off electrically. Pressure lets it work at 250β450Β°C instead of 600Β°C+, which is what makes it usable on advanced nodes. | κ³ μ μμ μ΄λλ§ |
| High-ΞΊ dielectric | Gate insulation (hafnium oxide) with far higher permittivity than silicon dioxide, adopted below 28nm to stop leakage β at the cost of ~100Γ more interface defects, which is the problem HPSP’s tool exists to fix. | κ³ μ μ μ¨ μ μ°λ§ |
| Front-end (μ 곡μ ) | The wafer-fabrication half of chipmaking β deposition, lithography, etch, anneal β as opposed to back-end packaging and test. | μ 곡μ |
| 3κ°μ vs λμ | The two columns in a Korean interim income statement: the three-month quarter and the year-to-date cumulative. Quoting one as the other is the single commonest error in reading these filings. | 3κ°μ / λμ |
| Separate vs consolidated accounts | λ³λ (OFS) covers the parent company alone; μ°κ²° (CFS) includes subsidiaries. HPSP files no consolidated statements at all because it has no subsidiaries β so every HPSP figure is separate, and is not directly comparable to a consolidated peer. | λ³λ / μ°κ²° |
| Effective tax rate | Income tax charge Γ· pre-tax profit. A sharp one-year fall can inflate reported net income with no operating improvement β here it turned a 6.1% operating-profit decline into a 36.4% net-profit rise. | νκ· μ ν¨μΈμ¨ |
| λλ보μ μν©λ³΄κ³ μ | The “5% rule” filing a Korean holder must make when crossing or moving 1%+ through a 5% stake. It states shares, percentage, method and reason β which is how a block sale becomes publicly traceable rather than inferred. | μ£Όμλ±μ λλ보μ μν©λ³΄κ³ μ |
| Treasury-share cancellation | Permanently retiring repurchased shares so the count falls. A side effect: every remaining holder’s percentage rises, including the controlling shareholder’s, without anyone buying. | μκΈ°μ£Όμ μκ° |
π― Why It Matters for K-Export Stars
Every English description of this company traces, ultimately, to one press release HPSP wrote in January 2024. It was syndicated three ways, absorbed into screener boilerplate, repeated by AI summary pages that added a “90%+ market share” figure nobody sourced, and mirrored by a wiki. None of it is dishonest. All of it is downstream of a single company sentence β and two and a half years stale on the developments that matter: a competitor with a product and a patent-court win, and a controlling shareholder that halved its stake.
The Korean record had all of it, on time, in public. None of it required access, sources, or a Bloomberg terminal β only opening the filing in the language it was written in. The wires will always beat us on what happened; they will not read two filings of 100,000-plus characters in Korean and notice that they contradict each other. HPSP also fills a real gap in our coverage: we have written extensively about Korea’s chipmakers β SK hynix, Samsung, and suppliers like Samsung Electro-Mechanics β but never about the companies that sell them the tools.
Conclusion
HPSP is a genuinely unusual asset: a 136-person company with a 52% operating margin, no debt to speak of, no subsidiaries, no holding-company structure, and a tool that advanced logic and memory lines appear to need. That is worth understanding on its own terms.
But it is not the frictionless monopoly the English record describes, and the company has never quite said it was. It told its regulator that its share cannot be calculated. A competitor filed a product description of the same technology on the same day. Revenue has fallen two years running, the headline profit growth is a tax artefact, and the private-equity owner that built the company has sold half its stake and holds 20.46% of what remains.
None of that makes HPSP a bad business. It makes the confident English version of HPSP an unreliable basis for a decision. Read the filing.
This company is one of ten in our K-Semiconductor Filing Map — a supply-chain map where every cell is marked by what the filing itself supports, and each one links to the receipt number it came from.
Disclaimer: This article is for informational and educational purposes only and does not constitute investment advice, a recommendation, or an offer to buy or sell any security. All figures are sourced from public regulatory filings (DART), exchange data (KRX) and our own datasets, with as-of dates stated inline; prices and valuations change and may be stale by the time you read this. K-Export Stars holds no position in HPSP Co., Ltd. Do your own research and consult a licensed financial adviser before investing. Found an error? See our corrections policy.
The rest of this supply chain, read the same way
Five Korean semiconductor companies, each decoded from its own Korean-language filing on the same day. The sector overview sits at Korea’s Chip Supercycle.
- Hanmi Semiconductor β the TC bonder that stacks HBM, and a quarter that reads two ways
- Isu Petasys β the AI accelerator boards, billed 97% abroad
- Dongjin Semichem β photoresist, and the anonymous customer note that has names attached
- ISC β the test sockets, and the treasury stock that was never really treasury
