ISC (095340): Korea’s AI Test-Socket Maker, and the Treasury Stock That Was Never Really Treasury (2026)

Every AI accelerator that ships has been pressed, at least once, into a small rubber pad full of conductive particles. The pad is called a test socket, and a 412-person company in Seongnam, Korea makes more of one particular kind than anyone else. In April 2026, that company handed 678,259 shares to a group of bondholders — and its share count did not move by a single unit.

🏢 What Is ISC, and What Does It Actually Sell?

ISC sells test sockets — the replaceable interface that holds a finished semiconductor against a test machine — and it is the company that first commercialised the rubber-based version of that part. Founded on February 22, 2001 and listed on KOSDAQ on October 1, 2007, ISC Co., Ltd. (아이에스시) is headquartered in Seongnam, Gyeonggi Province, just south of Seoul. It employed 412 people at the parent company as of June 30, 2026, with average tenure of 9.0 years (Source: DART semi-annual report 2026.06, rcpNo 20260814000114).

The product is easy to underrate. Before a chip is sold, it is tested — often several times, at different temperatures, sometimes for hours. It cannot be soldered down for that; it has to be pressed into a socket, tested, and pulled out again, thousands of times. The socket wears out. It is a consumable, it is specific to each package design, and it has to be delivered fast.

There are two ways to build one. The conventional way uses spring-loaded metal pins — pogo pins. ISC’s signature product uses a rubber pad instead: a silicone elastomer sheet with columns of conductive particles suspended in it. The company’s filings call this the elastomer socket (silicone rubber socket), and state that ISC was the first in the world to mass-produce it. ISC also makes pogo-pin sockets (entered 2014, shipping since 2018) and burn-in sockets (world-first silicone burn-in socket in 2019).

Since Q2 2025 it has added a second leg: test equipment and materials, through the acquisitions of iSemi and Tech Dream — module testers, high-speed burn-in testers, EFEM automation modules, and cleaning chemicals used after DRAM and HBM etch steps.

Three consolidated subsidiaries remain as of June 30, 2026: ISC International Inc. (Santa Clara, California; 80%-owned), ISC VINA Manufacturing (Vinh Phuc, Vietnam; 100%), and Tech Dream Co., Ltd. (Hong Kong; 100%).

📊 Where Does ISC’s Revenue Actually Come From?

Two-thirds of first-half 2026 revenue came from outside Korea, and the single largest destination was the United States at 30.7% — narrowly ahead of ISC’s home market.

Consolidated revenue by destination, six months ended June 30. Source: DART semi-annual report 2026.06, segment note (rcpNo 20260814000114). FX: ₩1,346/$1 as of Sep 4, 2026, applied to every dollar figure in this article.

What the destination table does not show is the mix change underneath it. The same filing gives quarterly breakdowns of the socket business that have no English equivalent anywhere:

Tables 1–4 of the business section, DART semi-annual report 2026.06. Two cautions, both worth stating rather than smoothing over. (1) The filing’s narrative text, a few paragraphs above these tables, says non-memory was 83% and memory 17% — but labels that figure “as of Q4 2025.” It is a carry-over from the March 2026 annual report, not a first-half number; the tables above are the current data. (2) The filing’s own memory/non-memory column for Q2 2025 sums to 96%, not 100%. We therefore quote only the memory row, which is unambiguous, and do not infer the missing four points.

Two things jump out. Memory doubled from 11% to 22% of socket sales — and management explicitly attributes that to legacy memory orders, which it describes as one-off and dilutive to profitability. And “AI memory” (HBM and SOCAMM test sockets) went from nothing a year ago to 11% of AI revenue in Q1 2026 — then back down to 9% in Q2. One quarter of decline in a line item this new proves very little, but we are not going to present it as a range when it is a direction, and the direction over two quarters is down. The AI-compute share fell too, 87% → 83% → 81%. If you want the broader memory backdrop those sockets are riding, our guide to Korea’s chip supercycle covers the demand side.

🗓️ How Did ISC Get Here?

ISC spent the 2010s as a founder-run KOSDAQ component maker, was bought in 2023 by a private-equity fund partnered with SKC, and spent 2025–2026 absorbing acquisitions while that private-equity partner exited. The history section is only in the annual report — semi-annual filings omit it by rule.

Sources: DART annual report 2025.12 (rcpNo 20260318000018), company history and major-shareholder sections; DART large-holding report (rcpNo 20260305001334); DART merger completion report (rcpNo 20251031000618).

🌍 Is ISC Actually a Global Leader, or Just a Korean One?

ISC’s filings claim world market-share leadership in test sockets and a dominant position in elastomer sockets specifically — but both claims are the company’s own, and one of them is written in a hedged passive voice with no source attached.

Here is exactly what the semi-annual report says, in two separate places, about two different markets. They are not the same claim and should not be merged:

• On elastomer sockets: they account for 80% of ISC’s revenue, and ISC’s share of the global elastomer socket market “is assessed at around 90%.” The Korean construction — 평가받고 있습니다 — is passive and unsourced. It says the company is regarded as holding that share; it does not say who regards it so. Treat it as a company characterisation.
• On test sockets overall: ISC states it has held the world’s number-one market share since 2015. Again company-stated, and a different, much wider market than the first claim.

What can be checked is narrower, and it comes from the same filing. ISC’s intellectual-property table reports 451 registered patents and 72 applications, plus 11 registered utility models, as of the semi-annual reporting date. (A narrative paragraph elsewhere in the same document gives 434 registered and 85 applications “as of the end of Q1 2026,” and a third passage says “about 430.” We use the structured IP table, which carries the report’s own reporting date; the discrepancy between the three is not explained.) ISC also says it serves roughly 300 customers including three global memory IDMs, North American CPU/GPU and AI fabless firms, Taiwanese and Chinese mobile fabless firms, OSATs, and automotive chip customers. The filing names Samsung Electronics and SK hynix as domestic buyers, which is the one place its customer list is explicit. It names no competitors anywhere.

More useful still: the same filing gives the market sizes, which put a ceiling on both claims. ISC estimates the global test-socket market at $1.4bn in 2024, growing 5–6% a year, with the separate burn-in socket market at $540M, and expects the test-socket market to grow 10% in 2026. Hold those against ISC’s own numbers. Its entire FY2025 revenue was $163.6M — about 12% of that $1.4bn test-socket market, which is a demanding reading of “world number one” unless the market is unusually fragmented. And if elastomer sockets are 80% of ISC’s revenue, roughly $131M, then a 90% share implies the whole global elastomer socket market is only about $145M — a tenth of the socket market. That is internally consistent: elastomer is a genuine niche, and ISC can plausibly dominate it while being a modest share of sockets overall. Neither figure is independently audited, but at least they now constrain each other.

📈 Is the Business Getting Better or Worse?

Much better over three years, and still improving in the first half of 2026 — but the second quarter was measurably worse than the first on margin, and the company says so itself.

Sources: DART annual report 2025.12 (rcpNo 20260318000018) and semi-annual report 2026.06 (rcpNo 20260814000114). All figures consolidated (K-IFRS). Three-month and cumulative columns are kept separate deliberately — they are different periods and must not be read across.

Three observations the headline numbers hide.

First, the profit recovery is bigger than the net-income line suggests. Operating profit rose 4.2× in FY2024 and another 34.1% in FY2025 — yet net income grew just 2.6% in FY2025. That is because FY2024’s net income ($40.8M) actually exceeded its operating profit ($33.3M), on non-operating items, while FY2025’s net income sat below operating profit. A screener reading “net income +2.6%” understates what happened operationally; anyone extrapolating FY2024’s 31.5% net margin overstates it.

Second, Q2 2026 was a step down from Q1 on margin. Backing Q1 out of the cumulative figures gives Q1 revenue of $50.7M with a 34.6% operating margin and a 51.8% gross margin. Q2 delivered more revenue ($54.2M) at a 29.2% operating margin and 47.0% gross margin. Management’s stated explanation is that legacy memory orders came in, which it characterises as one-off revenue that weighed on profitability. The income statement is consistent with a mix effect — the compression shows up at the gross-margin line, not in overheads. But “one-off” is the company’s characterisation, not a verified fact; the next two quarters will settle it.

Third, the product mix inside the revenue line has flipped. Using the sales-performance table (which is stated before intercompany elimination, so it does not tie to consolidated revenue), ISC’s own manufactured products were 38.0% export in FY2024, 78.5% in FY2025 and 87.8% in H1 2026. Over the same span, lower-value merchandise trading shrank from the largest line to a small one. That shift toward in-house manufacturing for overseas customers is the mechanical reason margins improved.

🔍 Why Did ISC’s Free Float Rise 3.3% Without a Single New Share Being Issued?

Because 678,259 shares that ISC reported as treasury stock had been pledged since 2022 as the settlement asset for two private exchangeable bonds — and on April 1, 2026 the bondholders claimed them.

This is the part that does not survive translation, so here is the arithmetic in full.

At December 31, 2025, ISC’s annual report showed 21,197,058 shares issued and 732,866 held in treasury — 3.5% of the company — leaving 20,464,192 shares in circulation. Directly beneath that table sits a single sentence of footnote: the treasury figure above includes 678,259 shares that are the subject of the first and second exchangeable bonds. In other words, 92.5% of what an English-language data feed would record as “treasury stock” was already spoken for.

Those bonds were issued on February 9, 2022, before SKC took control, and were disclosed with their terms in the bond note:

Face values and exchange prices are as filed — ₩6,523,050,500 at ₩26,762 and ₩13,476,949,500 at ₩31,016 — in the bond note of the DART semi-annual report 2026.06 (rcpNo 20260814000114), which also reports the 678,259 shares delivered and, in the consolidated cash-flow statement, ₩1,578 paid out as cash for fractional shares.

The share counts in the right-hand column are ours, not the filing’s — and they reconcile to the single share and the single won. ISC does not publish a per-tranche share count, so we divided each face value by its filed exchange price. The first tranche gives 243,743 shares with ₩334 left over; the second gives 434,516 shares with ₩1,244 left over. That totals 678,259 shares — exactly what ISC delivered — and the two remainders sum to ₩1,578, exactly the fractional-share cash the cash-flow statement records. Two independently filed figures, the delivered share count and a payment of ₩1,578, both fall out of the same arithmetic.

Now follow the shares. Treasury of 732,866 minus the 678,259 delivered leaves 54,607 — precisely the treasury figure in the June 30, 2026 report, where the treasury ratio has fallen from 3.5% to 0.3%. Shares in circulation went from 20,464,192 to 21,142,451, an increase of exactly 678,259, or +3.31%. Shares issued: 21,197,058, unchanged in FY2023, FY2024, FY2025 and H1 2026. The bridge closes to the single share.

Why it matters to a foreign investor: every one of those effects is invisible in the field most screeners track. “Shares outstanding: 21,197,058, unchanged year on year” is literally true and completely misses a 3.3% expansion of tradeable stock.

One important qualification, because it changes who is at fault. None of this was hidden from the Korean market — it was hidden by the language barrier and by the choice of metric. ISC has been publishing a diluted earnings figure on the fully-exchanged share count for years: its FY2025 annual report shows diluted EPS of ₩2,691 computed over an adjusted weighted average of 21,142,451 shares, which is 21,197,058 less 732,866 of treasury plus the 678,259 bond shares — the post-April share count, disclosed months before the bonds were exercised. The FY2025 treasury table likewise carries a footnote putting the retained balance at 54,607 “excluding the 678,259 shares.” The dilution was in the filings, in Korean, all along. What was invisible was the English-language summary of them.

And the same event reveals something about shareholder returns. In Korea, a buyback only shrinks the share count when the shares are formally cancelled (소각) — a distinction we have written about before, in the context of Celltrion cancelling 4% of itself while its share count went up. ISC’s own share-total table answers the cancellation question directly: cumulative shares retired through profit cancellation, zero. Its treasury stock was never a cancellation buffer; it was collateral, and it has now been spent.

Worth noting on the other side: the bondholders got a remarkable trade, and ISC’s existing shareholders funded it. Stock exchanged at an average of ₩29,487 was worth ₩264,000 at the April 1, 2026 close — the 678,259 shares had a market value of about $133M against $14.9M of bond principal. But the terms were struck in February 2022, before the AI socket ramp and before the change of control, and the dilution was fully disclosed in Korean at the time. This is a story about disclosure asymmetry between languages, not about anything hidden from the Korean market.

🏛️ Who Controls ISC, and What Happened to the Private-Equity Fund?

SKC controls ISC with 48.49%, and it reached that level in March 2026 by exercising a call option that bought out the private-equity fund that had co-controlled the company since 2023.

The 2023 change of control was not a plain corporate acquisition. A large-holding report filed on March 5, 2026 shows the structure and its unwinding in one table. Until that date, Helios Private Equity Fund No. 8 — a Korean institutional PE vehicle whose largest limited partner is the Korea Teachers’ Credit Union at 58.62% — was the reporting holder of a joint block of 11,013,626 shares and share-equivalents, or 51.96%, held for the express purpose of management participation. SKC was listed as a special-relationship party inside that block.

On March 5, 2026 the structure came apart in a single filing:

The filing states the method as “call option exercise, dissolution of special relationship,” and the reasons as disposal on call-option exercise, termination of the joint-holding arrangement under the shareholders’ agreement, waiver of the right of first refusal, and release of the share and bond pledges. Transaction price: ₩68,612 per share. Source: DART large-holding report, rcpNo 20260305001334.

The two discoveries in this article are therefore the same event seen from two sides. The PE fund left ISC’s register through two instruments negotiated in advance — a call option over its shares and the exchangeable bonds over ISC’s treasury stock — and both settled within four weeks of each other in March and April 2026.

Two governance observations follow, and they cut in opposite directions.

The strike price deserves attention. ISC closed at ₩202,000 on March 5, 2026, so SKC acquired 734,490 shares — about $37.4M worth at the strike — at roughly 66% below the market price that day. That is not improper; call-option strikes in a 2023 shareholders’ agreement are set years in advance and are the price of the PE fund’s original capital. But it is a right that existed only for the controlling shareholder. Minority holders had no equivalent instrument, and there was no tender offer at that price.

SKC also tightened board control alongside the stake. At the March 26, 2026 AGM, two SKC executives joined ISC’s six-person board as non-executive directors: Kim Jong-woo, currently CEO of SKC and SK Nexilis and himself a former CEO of ISC, and Kim Sun-hyuk, currently head of SKC’s strategy division and concurrently head of SK Inc.’s Value TF. Three internal directors resigned at the same meeting, leaving one executive director — Kim Jung-ryul, an ISC executive since November 2003, now sole CEO — plus three independent directors. In mitigation, the board is chaired by an independent director, Cho Dong-il, a position the filing describes as coordinating the board rather than running the company. Meanwhile the National Pension Service, which held 6.2% at the end of 2025, does not appear among holders of 5% or more at June 30, 2026.

The second-largest customer is the controlling shareholder’s sister company

ISC’s unnamed “Customer B” is SK hynix — and the two filings say so, if you read the customer note and the related-party note against each other. DART requires disclosure of any customer above 10% of revenue but not its identity, so the customer note calls them only “Company A” and “Company B.” The related-party note, a few pages later, breaks out sales to each SK Group affiliate by name. The figures are identical to the thousand won across every period both notes cover:

Figures shown in thousands of won as filed, because the point is the exact match rather than the magnitude. FY figures from the annual report (major-customer note and related-party note 33); half-year figures from the semi-annual report. In each period the related-party table’s column total reconciles to the sum of its named affiliates, which is how the SK hynix column is identified. This is a cross-reference between two notes, not an inference from context. Customer A remains unidentified and we make no guess about it.

That reframes the concentration risk entirely. ISC’s second-largest customer is not merely a large buyer — it is a fellow member of the SK conglomerate group, related to ISC through SKC’s 48.49% stake, and the filing classifies it as a related party under the Securities and Futures Commission’s rule on large business groups. Sales to that customer went from ₩2.6bn to ₩32.9bn — 12.6× in a single year — in the two years after SKC took control in October 2023. Total related-party revenue was ₩33.8bn in FY2025, 15.4% of consolidated sales, of which SK hynix is 97%.

This also identifies the “captive customer” that the filing credits for the equipment-and-materials business without naming: the same document says that unit supplies HBM and DRAM cleaning chemicals, EFEM modules and memory-module PCBs, and ties its second-half ramp to the Yongin cluster. Given the related-party reconciliation above, the captive customer is the same affiliate.

Two honest caveats. Nothing here implies the sales are not genuine — SK hynix is one of three global memory IDMs and an obvious buyer of memory test sockets, and ISC’s non-memory business, which is the larger part, is unrelated. But related-party revenue growing eightfold in the two years after a change of control is a fact a foreign investor should weigh themselves, and receivables give a second reason to look: the related-party note shows ₩14,381,412k ($10.7M) of receivables and similar balances due from SK hynix at June 30, 2026 against ₩15,621,921k of half-year sales to it — roughly 167 days outstanding from a single counterparty (the line is captioned “receivables etc.,” so it is not purely trade debt).

On returns to minorities: ISC has now paid a year-end dividend for 17 consecutive years. The annual report’s dividend-history table shows 16, but its own footnote says that count runs “through the 24th fiscal year (FY2024)”; the FY2025 dividend approved at the March 26, 2026 AGM makes 17 (Source: annual report 2025.12, dividend history). It declared ₩850 per share for FY2025, up from ₩810 and ₩200 in the two prior years, for a 31.0% consolidated payout ratio. The total paid in H1 2026 was $12.9M — which, divided by ₩850, gives 20,464,192 shares, exactly the pre-April float. There is no quarterly or interim dividend, and, as noted, no share has ever been cancelled. For the wider policy backdrop, see our guide to Korea’s Value-Up program and to the structural Korea Discount.

💰 What Do You Actually Pay for ISC Today?

About 68 times last year’s earnings — twice the multiple of Leeno Industrial, which earns a margin nearly twice as high.

Price and share count from KRX settled closes, cross-checked against three independent stores. Use the diluted line. This article has just argued that the basic share count was the wrong denominator — ISC’s diluted EPS is struck over 21,142,451 shares, the post-April float, and the difference between the two bases is the same 3.3% the article is about. Reconciling the two P/E figures you will see here: 66.4× and 67.6× divide the price by per-share earnings attributable to owners of the parent; the 68.5× in the peer table below divides market capitalisation by total consolidated net income including non-controlling interests. Both are correct on their own terms; the peer table uses the market-cap method because it is the only one we can apply identically across three issuers. Separately, the TTM figure is bridged as FY2025 ₩2,743 − H1 2025 ₩896 + H1 2026 ₩1,911, and the weighted-share bases of those three periods differ, so treat that one line as approximate. Equity and net cash are consolidated and include non-controlling interests; net cash is cash plus short-term financial instruments less lease liabilities, with the exchangeable bonds now extinguished. Sources: KRX for price and shares; DART filings above for earnings and balance sheet.

Peers, on a single consistent basis — market capitalisation divided by FY2025 net income as each issuer reported it on DART, all at the September 4, 2026 close:

Peer set selected by us — ISC’s filings name no competitors. Leeno is Korea’s pogo-pin socket and probe specialist; TSE makes probe cards and test interfaces. All figures on the same date and the same FX (₩1,346/$1). ⚠️ Basis caveat: ISC’s and TSE’s figures are consolidated; Leeno reports separate-company accounts only, so its row is not a consolidated number and cannot be made into one. Sources: FY2025 annual reports filed on DART by each issuer (ISC’s is rcpNo 20260318000018); prices and share counts from the KRX market data portal (daily settled quotations).

The comparison is uncomfortable for ISC on a static reading: it is the most expensive of the three on trailing earnings while sitting between the two on margin and growth. The natural counter-argument is that ISC’s margin is catching up — so we checked it on matched periods rather than comparing ISC’s half-year to Leeno’s full year, which would have flattered ISC by about two points:

Leeno H1 2026: revenue ₩242,958,602,798, operating profit ₩120,814,329,893, per its half-year report (rcpNo 20260814000445). TSE H1 2026 consolidated: revenue ₩339,448,089,387, operating profit ₩94,494,936,461, per its half-year report (rcpNo 20260814002616). ISC’s figures are from its own half-year report cited throughout. ⚠️ Basis differs and cannot be made identical: ISC and TSE are consolidated; Leeno files separate-company statements only — it reports no consolidated accounts.

So the gap did narrow — but by 2.3 points, not the 4.5 that the mismatched comparison implies, and it narrowed while Leeno’s own margin rose from 47.5% to 49.7%. ISC is improving into a target that is also moving. The sharper development is at the other end of the table: TSE’s operating margin more than doubled to 27.8%, erasing most of ISC’s advantage over it while trading at a lower multiple. ISC’s TTM P/E of 48.4× is the number that reflects its own improvement; whether that is enough is the entire investment question, and it is not one we are going to answer for you.

🏷️ Which ISC Are You Actually Buying?

There is one listed ISC in Korea — KOSDAQ 095340 — it has no ADR and no US listing, and the most common mix-up is with its listed parent, SKC.

A practical warning for anyone searching: “ISC” is a heavily reused acronym globally, and there is no English Wikipedia article for this company to disambiguate it. Search by the Korean ticker, 095340. Because there is no ADR, buying requires direct KOSDAQ access — see our walkthroughs on how to buy Korean stocks as a foreign investor, the complete list of Korean ADRs (ISC is not on it), and reading Korean share prices in USD.

🌏 Is Foreign Money Buying or Selling ISC?

It depends entirely on the window you choose — foreigners sold ISC over the last five and twenty sessions, and bought nearly 9% of the entire company over the last fifty-five.

Foreign investor net buying, settled sessions on the Korea Exchange. Values are derived: each day’s net share count multiplied by that day’s close, then summed — an approximation, not execution prices. The 55-session window is the full extent of our dataset for this stock, not a chosen period. Foreign ownership stood at 29.7% on Sep 4, 2026. Source: our own flow database, built from Korea Exchange and Korea Investment & Securities data. Verification is uneven and we should say so: the 5- and 20-session share counts match Naver Finance’s independently published figures exactly, but the 55-session total has no second source — it rests on our database alone.

The signs disagree, and that disagreement is the story. Reading only the 5-day column produces “foreigners are selling ISC.” Reading only the 55-session column produces “foreigners bought 8.9% of the company in under three months.” Both are arithmetically true. The reconciliation is that the recent selling is a small give-back on top of a very large accumulation — about 9.5% of what was bought over the longer window has been sold back over the last twenty sessions. That is trimming after a run, not distribution, and it is the kind of framing that a single-window headline systematically destroys. We publish both windows on every flow piece for exactly this reason; the ongoing series lives at our Foreign Flow Watch hub.

📚 Lingo Check

🎯 Why It Matters for K-Export Stars

ISC is a clean demonstration of the gap this site exists to close. Everything a foreign investor would most want to know about it — that AI is 81% of socket revenue, that memory share doubled on orders management itself calls one-off, that a private-equity fund co-controlled the company until March 2026, that 92.5% of its reported treasury stock was pledged collateral, that the anonymous “Company B” behind 15% of last year’s sales is the controlling shareholder’s own sister company — sits in Korean-language regulatory filings with no English equivalent. That last one took nothing more exotic than reading two notes in the same document and noticing the numbers were identical. The company has no English Wikipedia entry, no ADR, and no analyst coverage in English that we could find.

It also sits in a part of the AI supply chain that the wires rarely reach. Coverage of Korean semiconductors concentrates on the two giants — see our deep dives on what SK hynix actually makes and who buys it and on Samsung Electronics’ memory cycle, both of which are named domestic customers of ISC. The component layer beneath them, where Samsung Electro-Mechanics and ISC operate, is where a lot of the cycle’s operating leverage actually shows up — ISC’s operating profit rose 4.2× in a single year.

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 drawn from official regulatory filings and exchange data as cited, and are accurate as of the dates stated; prices, valuations and shareholdings change. Do your own research and consult a licensed financial adviser before making any investment decision.