NAVER Corporation (KRX:035420) had a good first half by the measure most people quote. Revenue of ₩6.63tn against ₩5.70tn a year earlier, up 16.3%, at a company that has held a search monopoly in its home market for two decades and is not supposed to grow like that any more.
Operating income was ₩1.06tn against ₩1.03tn. Up 3.4%.
In the second quarter alone the operating line went backwards: ₩520.3bn against ₩521.6bn a year earlier. Revenue in that quarter grew 16.3%.
Take the operating margin quarter by quarter and the shape is clear. Q1 2025: 18.13%. Q2 2025: 17.89%. Q3 2025: 18.18%. Q4 2025: 19.11%. Q1 2026: 16.72%. Q2 2026: 15.35%.
The peak was the fourth quarter of last year. Since then the margin has fallen 376 basis points across two quarters, on revenue that grew in both of them.
For historical perspective, NAVER earned operating margins between 25% and 27% from FY2015 through FY2017, collapsed to 10.77% in FY2019 as the Japanese messenger business consumed money, and recovered to 18.43% in FY2024 and 18.35% in FY2025. The current 15.35% is below that recovered range and heading down.
Something is absorbing the revenue growth before it reaches the operating line, and it is not hard to find.
Purchases of property, plant and equipment ran at ₩753.9bn in FY2021, ₩700.7bn in FY2022, ₩640.6bn in FY2023 and ₩554.0bn in FY2024. Four years of steady decline, which is what a mature internet business looks like.
Then ₩1,234.7bn in FY2025, up 123%. Then ₩809.8bn in the first half of 2026 alone, against ₩350.9bn in the same period of 2025, up 131%. Annualise that and FY2026 lands somewhere above ₩1.6tn, roughly three times the FY2024 figure.
The balance sheet registers it. Property, plant and equipment went from ₩2.84tn at the end of the third quarter of 2024 to ₩4.12tn at June 30, 2026, up 41.7% in seven quarters.
Buying that much equipment produces depreciation, and depreciation lands on operating income immediately whether or not the assets are generating revenue yet. That is the most likely single explanation for a margin falling while sales grow, and the shape of the spending, concentrated in servers and data centre capacity rather than offices, points at the same conclusion.
Cash generation tells the story more bluntly than the income statement does.
Cash from operating activities in the first half of 2026 was ₩1,186.6bn, against ₩1,124.9bn a year earlier, up 5.5%. Subtract the capital spending in each period and what is left over went from roughly ₩774bn to roughly ₩377bn.
Free cash flow cut in half, in a year when revenue grew 16%.
NAVER can afford it comfortably. Cash was ₩6.38tn at June 30 and total equity ₩31.03tn. But the company has also been borrowing: non-current liabilities went from ₩4.11tn at December 31 to ₩6.01tn at June 30, up ₩1.90tn in six months, and intangible assets rose from ₩3.42tn to ₩4.66tn over the same period, an increase of ₩1.24tn that the summary financials do not explain and that the notes would.
One more warning for anyone modelling this company from the bottom up. NAVER's reported net income is not a useful quarterly signal, because a large equity-accounted holding sits below the operating line and swings it.
In the first quarter of 2026 pretax income was ₩393.5bn against operating income of ₩541.8bn, so pretax came in ₩148bn below operating profit despite ₩52.6bn of net finance income. In the second quarter pretax income was ₩902.0bn against ₩520.3bn of operating profit, ₩382bn higher, with net finance income of only ₩96.4bn.
A swing of several hundred billion won a quarter, in both directions, from a line most investors never look at. Net income was ₩163.0bn in the fourth quarter of 2025 and ₩734.7bn in the third. Neither number tells you anything about how the search and commerce business performed.
Operating income is the line that does.
The uncomfortable part is that everything above describes NAVER before its major AI commitment begins.
The company's amended disclosure of July 27 sets out a data centre build with Nvidia: 55MW in the first half of 2027, 100MW cumulative by the end of that year, 200MW in 2028, and gigawatt scale eventually. Roughly $9bn of computing infrastructure is proposed to be arranged by Brookfield with NAVER contracting to use it, and Nvidia is subscribing to about $1bn of new NAVER shares.
If that structure holds, the GPU capital cost sits with Brookfield rather than on NAVER's balance sheet, which is the point of it. But NAVER still has to build and power the sites, and site and power spending is exactly the kind of capital expenditure that has already doubled twice.
So the margin compression visible today is the prelude.
The bull argument is straightforward and not weak. A company with a protected cash-generating monopoly, trading at about 1.06 times book value at the August 27 close of ₩216,500, is spending its cash flow to build a position in the one market that might matter more than search over the next decade. Margin falling from 19% to 15% while revenue compounds at 16% is a trade most shareholders should accept.
The bear argument is that NAVER has never operated infrastructure at scale for third parties, that the returns on AI compute capacity are unproven for anyone who is not a hyperscaler, and that a company can spend its way from an 18% margin to a 12% one considerably faster than it can build a new business.
Both are true statements about risk. Which one wins gets settled by revenue in the cloud and AI segment, not by announcements.
Operating margin in the third quarter, reported late October. Two consecutive declines can be a step change in depreciation working through. A third would establish a trend, and 15.35% falling toward 14% would mean the spending is outrunning the revenue it is meant to produce.
The second marker is the FY2026 capital expenditure figure in the annual report. If it lands near ₩1.6tn as the first half implies, and 2027 guidance goes higher again because of the Nvidia build, then investors should stop valuing NAVER as an asset-light internet company. The capital intensity of this business has changed, and the multiple people are willing to pay for it usually changes with it.
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.