Samsung Electro-Mechanics Co., Ltd. (KRX:009150) does not usually file individual sales contracts. Multilayer ceramic capacitors move on rolling purchase orders, by the billion, and no single order normally clears the threshold that forces a Korean listed company to disclose one. This summer it filed two, twenty-four days apart.
The first, signed June 29, is worth $294,000,000. The second, signed July 22, is worth $200,000,000. Both are described only as an "MLCC supply contract." Both run from January 1 to December 31, 2027. Both name the counterparty as "a global large corporation" and then withhold the name, along with the key commercial terms, under a business-confidentiality reserve that runs to the end of 2027.
Two contracts for the same product, for the same year, from a customer nobody is allowed to name. That's the interesting part, not the money.
Korean rules require a listed company of this size to file a single sales contract once it passes a set share of the prior year's revenue. The June contract came in at 4.0% of FY2025's ₩11.31tn. The July one came in at 2.6%. Both just clear the bar, which is why they surfaced and why dozens of other MLCC orders this year did not.
That framing matters when you size them. Combined the two are ₩749.1bn, about 6.6% of last year's revenue, and they cover a full twelve months starting sixteen months after the first signature. This is not a step change in the top line. What it is, is evidence that at least one buyer wanted volume locked a long way out, in dollars, on paper, and did not want anyone to know who they were.
The reserve is the tell. A company that had signed a routine annual supply agreement with a long-standing handset customer would not need the counterparty's identity sealed for eighteen months. Somebody asked. The filings say the seal was requested by the counterparty for protection of its trade secrets, which is unusual language for a capacitor order and ordinary language for a customer whose build plans are themselves the secret.
I am not going to guess the name. The filings don't say, and the standard candidates run from a US hyperscaler to an automaker to a server OEM, with no way to distinguish between them from public documents.
Ignore the contracts for a moment and read the quarter. Q2 2026 revenue was ₩3,457.2bn, a record, up 24.2% from ₩2,784.6bn a year earlier. Gross profit was ₩844.5bn for a 24.4% margin, against 20.3% in Q2 2025. Operating income was ₩440.4bn, a 12.7% margin, against ₩213.0bn and 7.6%.
Doubling operating profit on 24% more revenue is a mix and pricing outcome, not a volume one. MLCCs sold into AI servers and into automotive carry specifications, tolerances and prices that phone-grade parts don't, and the same fab that makes the cheap ones makes the expensive ones. When the mix shifts, gross margin moves before revenue does.
For scale against the company's own history: FY2021 was the previous peak, ₩1,486.9bn of operating income on ₩9.68tn of revenue, a 15.4% margin. FY2018, the last great MLCC shortage, produced ₩1,149.9bn on ₩8.00tn. Then FY2023 collapsed to ₩660.5bn on ₩8.89tn as electronics inventories cleared. Annualise Q2 2026's ₩440.4bn and you get ₩1.76tn, ahead of the FY2021 record, on revenue that is 43% higher than 2021's. The margin has not fully recovered to 2021 levels. The absolute profit has more than made up the difference.
Something small in the filings is worth pulling out, because it affects everything a Korean exporter earns. The June 29 contract converted $294m at ₩1,544.20 to the dollar. The July 22 contract converted $200m at ₩1,475.60. In twenty-three days the won strengthened 4.4%.
Both contracts are denominated in dollars. Their won value falls as the won rises, and neither the filings nor the half-year report describe hedging on these specific agreements. A components exporter that books a dollar contract in 2026 and delivers it through 2027 is running an unhedged currency position unless it says otherwise, and at this size a 4% move is roughly ₩30bn on the combined value.
Anyone converting these numbers should use the rates the company used rather than a round assumption. I'm using ₩1,475.60 for anything dated after July 22, which puts the ₩103.5tn market cap at about $70bn.
Two lines deserve attention before anyone treats the margin as clean.
Trade receivables were ₩2,306.1bn at June 30, up 34.4% from ₩1,715.6bn a year earlier, while revenue grew 24.2%. Receivables outrunning sales by ten points over four quarters is not alarming on its own, and it's consistent with a shift toward larger customers on longer terms. It is worth tracking, because it's also what channel-stuffing looks like from the outside and the half-year report doesn't break out the ageing.
Capital spending is the louder signal. Purchases of property, plant and equipment ran ₩843.2bn in the first half, up 89.6% from ₩444.7bn in the same period of 2025, and equal to 70.7% of the ₩1,192.1bn spent in all of FY2025. Half a year has absorbed more than two-thirds of last year's full-year build.
Cash generation is keeping up so far. Operating cash flow was ₩1,060.2bn for the half against ₩674.1bn a year ago, up 57.3%, and the cash balance rose from ₩2,701.2bn at December 31 to ₩3,314.4bn at June 30. The company is also borrowing: total liabilities reached ₩6,036.8bn at June 30 against ₩4,185.1bn a year earlier, and financing activities brought in ₩425.5bn over the half rather than paying cash out. A company that has decided to build ahead of demand is a different risk than one harvesting a cycle.
The obvious counter is that two contracts totalling 6.6% of revenue prove nothing about 2027 as a whole, and that reading a sealed counterparty as significant is reading tea leaves. Fair. Korean confidentiality reserves are common enough, and the customer could simply be a large Chinese handset maker that doesn't want its component sourcing published.
The bigger risk is the one the sector always has. MLCC is a capacity business with a violent cycle, and Samsung Electro-Mechanics is one of four or five suppliers all expanding at once. FY2016 produced an operating margin of 0.4%. FY2023 produced ₩660.5bn of operating income on more revenue than FY2018 generated ₩1,149.9bn from. If AI server build rates flatten in 2027 while every supplier's new capacity lands, the pricing that produced this quarter's 24.4% gross margin goes first.
And the valuation leaves nothing for that. First-half net income was ₩581.0bn. Double it and the ₩103.5tn market cap is about 89 times earnings, against a book value of ₩10.53tn for roughly 9.8 times. A components maker at 89 times is priced as though the current margin is the floor.
The next single-contract filing. If a third MLCC agreement of comparable size lands before year-end, covering 2027 or extending into 2028, the pattern is a programme rather than two orders, and the customer is committing to a multi-year ramp. If nothing follows by the Q4 results in January, these two were opportunistic and the 2027 top line rests on ordinary demand.
The second marker is Q3, due in late October. The specific number is gross margin. Holding 24.4% through a quarter with no new capacity coming online would mean the mix shift is structural. Slipping back toward 21% would mean Q2 caught a pricing spike, and the contracts were signed at the top rather than at the start.
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.