Read the April 15 filing and the April 15 press coverage side by side and you are looking at two different transactions.
The coverage said Samsung SDS Co., Ltd. (KRX:018260) had raised ₩1.22tn from KKR to fund artificial-intelligence infrastructure and overseas acquisitions, with KKR taking an advisory role on M&A and capital allocation. The shares rose about 20% on the day. The filing that went to the Financial Services Commission and the exchange says something narrower. Its use-of-proceeds table has five boxes: facility investment, business acquisition, operating capital, debt repayment, and acquisition of securities issued by other corporations. Four are dashes. The entire ₩1,220,000,000,000 sits in operating capital, and the schedule underneath assigns all of it to calendar 2026, with 2027 and 2028-onward blank.
The filing also never says KKR. The subscriber is Startech AI L.P., whose general partner is Startech AI GP L.L.C. and whose sole hundred-percent investor is Startech Aggregator L.P. A footnote explains that financial details are omitted because Startech AI L.P. was formed on April 1, 2026 — fourteen days before the board resolution.
Payment settled on April 30, so the second quarter is the first full look at what happened to it. Non-current liabilities went from ₩851.3bn at the end of March to ₩1,937.1bn at the end of June, which is the debt half of the convertible arriving on the balance sheet. Total assets jumped ₩1,454.0bn in the quarter.
What didn't happen is spending. Purchases of property, plant and equipment in Q2 were ₩129.9bn, up from ₩68.8bn in Q1 but ordinary for this company, which averaged ₩378.4bn a year over the last five years. Cash and equivalents actually fell, from ₩2,190.8bn to ₩1,556.5bn, while investing activities consumed ₩1,927.4bn. Set the ₩129.9bn of capital spending against that ₩1.93tn and it's clear the bulk went somewhere other than construction. Whatever it bought, it wasn't a data centre.
That's not damning by itself. Two AI data centre projects have been described publicly, one in Haenam targeted for 2028 and one in Gumi a year later, and neither would absorb much cash in the first sixty days. But it does mean the case for this deal still rests entirely on a plan, and the filing is the one document where the company had to commit that plan to a category and chose the least specific one available.
The single most informative line in a Korean convertible filing is a box a US investor would never think to look for. Korean private-placement convertibles routinely carry a refixing clause: if the share price falls, the conversion price resets downward, handing the bondholder more shares and diluting everyone else. The mandatory field is 최저 조정가액, the minimum refixing price, and issuers must state it. Samsung SDS's filing shows a dash.
So the conversion price is ₩180,000 and it stays there. The filing spells out that it will not adjust for cash dividends, for declines in the share price or trading volume, or for employee option and stock-award plans. It adjusts only for below-market issuance, splits and bonus issues, and it adjusts upward if the company shrinks its share count. The holder took the downside.
The rest of the structure runs the same direction. The coupon is 2.5%, with the yield to maturity also 2.5%, meaning there's no back-end premium — pay the coupons and nothing extra is owed in 2032. There are no put, call, or early-redemption rights on either side outside an event of default. Conversion can't even be requested until April 30, 2027, a year after payment. And by agreement the holder cannot transfer the bonds, or any shares received on conversion, for six years from issue.
Six years of no exit, no reset, and a 2.5% carry. That's a shareholder who has agreed to behave like a founder. It's also why the advisory arrangement described in the press release is more credible than the average PE headline: this investor cannot sell into a good quarter.
At ₩180,000 the bonds convert into 6,777,777 shares. The filing computes that as 8.76% of the 77,377,800 shares outstanding, or 8.06% of the enlarged count. The reference price used to set ₩180,000 was about ₩152,200, since the filing states the conversion price is 118.265% of it.
The stock closed at ₩245,000 on September 1. That's 61% above the April reference and 36% above the conversion price, which puts the embedded gain around ₩440bn on paper. Nobody can act on it for another eight months, and can't sell for another five and a half years after that, but anyone valuing the equity today should be counting 84.2m shares rather than 77.4m. On the enlarged count the ₩19.0tn market cap becomes closer to ₩20.6tn of claim on the same business.
One Korean mechanic makes the private placement possible at all: because this was a 사모 issue, a private placement, no securities registration statement was required, on the condition that conversion and denomination-splitting are barred for a year from issue. That one-year bar is why the conversion window opens in April 2027 rather than immediately. The six-year transfer restriction is separate and contractual.
This is the part that should temper the enthusiasm. First-half revenue of ₩7,070.7bn was up 1.0% on ₩7,001.7bn. First-half operating income of ₩310.1bn was down 38% from ₩498.7bn. Q1 was the damage: ₩78.3bn of operating income against ₩268.5bn a year earlier, on gross margin of 11.8% versus 15.6%. Q2 recovered to ₩231.8bn, roughly flat year on year, so the annual shape depends on whether Q1 was a one-off.
Samsung SDS earns most of its systems-integration revenue from Samsung affiliates and runs a large logistics-technology business alongside it, and neither is a high-multiple business on its own. FY2025 operating income was ₩957.1bn on ₩13.93tn of revenue, a 6.9% margin, and that was the best year since 2019. Annualise first-half net income of ₩275.8bn and you get ₩551.7bn, which puts the shares at about 34 times, against 24 times trailing FY2025 earnings of ₩782.7bn. Book value per the Q2 balance sheet is ₩10,643.3bn, so 1.78 times.
The strongest counter is that the operating-capital classification is bureaucratic, not meaningful. Korean issuers get penalised for missing a stated use of proceeds, and a company that hasn't picked its acquisition targets has a real reason to file the flexible box rather than name a facility it hasn't broken ground on. If that's the case, the filing tells you about legal caution and nothing about intent.
The second counter is that a six-year lock-up from a large private-equity investor at a premium, with no downside protection, is a genuine vote of confidence that no amount of category-box reading offsets. That investor is exposed to exactly what minority holders are exposed to, for longer.
And the counter to the counter is that Samsung SDS held roughly ₩6.4tn of cash before the raise, by press accounts, against ₩1.56tn of cash and equivalents on the June balance sheet with the rest in other financial assets. A company that cash-rich taking dilutive money at 2.5% is either about to do something large or was buying an adviser. It's fair to want to see which.
The Q3 report in late October and the FY2026 annual report in March. Three specific lines. Purchases of property, plant and equipment, which need to move well above the ₩199bn spent in the first half if the data centre story is real. Cash from investing, to see whether the ₩1.93tn Q2 outflow reverses out of financial instruments into something operating. And any 타법인 증권 취득 filing — an acquisition of securities in another company — which is the disclosure Samsung SDS would have to make if the M&A half of the KKR story starts happening.
Then April 30, 2027, when conversion first becomes possible. Nothing forces it, and the transfer ban runs to 2032 regardless, but the share count on the FY2027 statements is where this deal stops being a balance sheet item and starts being dilution.
kstock reads DART every morning and writes up what moved — the contract, the buyback, the number that does not add up. The daily post and a Saturday roundup, by email.