Samsung SDS has said it will invest ₩10 trillion by 2031. Half of that, ₩5 trillion, is earmarked for AI infrastructure: a data centre in Gumi, a stake in Korea's National AI Computing Center, and further sites as demand appears. The National AI Computing Center is a state-backed facility meant to give Korean firms sovereign access to large-scale compute.
In all of 2025, Samsung SDS Co., Ltd. (KRX:018260) spent ₩315.3 billion on property, plant and equipment.
The company runs the Samsung group's enterprise IT systems and a large freight-forwarding arm, and it is the listed vehicle most Korean investors use to buy the group's AI spending. That role is why the plan matters and why the gap between the plan and the accounts is worth measuring.
Property, plant and equipment was ₩1,773.9 billion at December 2024. At June 30, 2026 it was ₩1,777.9 billion. That is a change of ₩4 billion over eighteen months.
Look back further and the picture is the same. PP&E was ₩1,654.4 billion at the end of 2023, so the asset base grew 7.5% across two and a half years. For a company positioning itself around data centres, that is not a build. It is maintenance.
Capex confirms it. The company spent ₩591.9 billion in 2022, ₩450.7 billion in 2023, ₩474.7 billion in 2024 and ₩315.3 billion in 2025. The trend is down 47% from the 2022 level. First-half 2026 spending was ₩198.7 billion, which annualises to roughly ₩400 billion. Better, and still nowhere near a ₩10 trillion programme.
One caveat matters here. Under IFRS, leased space appears in PP&E as a right-of-use asset, so leasing rather than owning would still show up in this line. What would not show up is buying capacity from a hyperscaler under a service contract. More on that below.
Spread ₩10 trillion evenly from 2026 to 2031 and you get about ₩1.7 trillion a year.
Samsung SDS generated ₩1,203.3 billion of operating cash flow in 2025. It paid ₩233.1 billion of dividends. The plan, at that pace, would consume everything the business produces and then some.
The company can carry it. Total equity was ₩10,643.3 billion at June 30 against ₩4,507.3 billion of liabilities, so leverage is low by any standard. Cash was ₩1,556.5 billion, and current assets of ₩11,025.4 billion sit against ₩2,570.2 billion of current liabilities. This is one of the most over-capitalised large companies in Korea.
It also raised ₩1.22 trillion in convertible bonds in April. That filing designated the proceeds as operating capital rather than facility investment, and the money had not left as capital spending through June. Investing outflows in the June quarter reached ₩1,927.4 billion, but capex was only ₩129.9 billion of it. The rest went into financial instruments. The money is parked, not poured.
First-half revenue was ₩7,070.7 billion against ₩7,001.7 billion a year earlier. That is growth of 1.0%.
Operating income over the same half was ₩310.1 billion against ₩498.7 billion. That is a fall of 37.8%.
Most of the gap is one item. The March quarter carried a ₩112 billion charge for retirement benefit provisions after the company changed how it calculates retirement allowances. Add it back and first-half operating income becomes about ₩422.1 billion, still down 15% year over year.
Zoom out and the flatness is structural rather than recent. Operating income was ₩990.1 billion in 2019 and ₩957.1 billion in 2025. Six years of no growth, on revenue that rose from ₩10,719.6 billion to ₩13,929.9 billion. Operating margin went from 9.2% to 6.9% over the same stretch.
The cloud business is real. First-quarter cloud revenue was ₩690.9 billion, up 5.8%, with the company's own cloud platform growing 12% on demand for GPU-as-a-service, meaning rented GPU capacity, largely from public-sector customers.
Now do the arithmetic that the press release does not. Cloud was about a fifth of the ₩3,352.9 billion the group booked that quarter. A fifth of the business growing 5.8% adds roughly one point of growth at group level. Group revenue fell 3.9% in the same quarter.
That is the whole problem in one comparison. The part of Samsung SDS that investors want is growing at a rate that a normal IT services company would call ordinary, and it is not yet large enough to lift the total. The logistics arm, which moves freight and carries thin margins, is the reason the revenue line is as big and as volatile as it is. Revenue peaked at ₩17,234.7 billion in 2022 when ocean freight rates were extreme, then fell to ₩13,276.8 billion the next year. Nothing about the IT business changed in between.
The honest counter is that 2031 is five years away and nobody spends a ₩10 trillion programme in its first six months. Data centre construction has long lead times, and the Gumi site and the National AI Computing Center both involve partners and government process. A flat capex line in mid-2026 tells you the programme has not started, not that it will not happen.
The second counter is more interesting. A cloud business does not need much of its own plant. Samsung SDS resells and manages hyperscaler capacity as well as running its own, and that model puts cost in the income statement rather than the balance sheet. Judging an IT services company by its fixed assets is a manufacturing habit that may not apply.
Third, the balance sheet argues for patience rather than against it. With equity at ₩10,643.3 billion and almost no net debt, the company can fund the plan without a rights issue, which is more than most of its Korean peers can say.
The valuation is where the patience gets expensive. The company's market capitalisation was ₩18.4 trillion at Friday's close. That is about 28 times the ₩664.9 billion earned over the last four quarters, and 1.73 times book. Twenty-eight times is an AI multiple on a business whose operating income is lower than it was in 2019. The shares have ranged from ₩148,300 to ₩362,000 over the past year, which is the market arguing with itself about exactly this.
The capex line in the third-quarter cash flow statement is the single number. It has run between ₩60 billion and ₩130 billion a quarter for two years. A quarter above ₩300 billion would be the first hard evidence that the ₩10 trillion plan has left the press release.
Two supporting markers. The Gumi data centre needs a construction start date and a capacity figure, neither of which has been published in the filings. And the cloud revenue growth rate, 5.8% in the first quarter, needs to reach double digits before the segment is big enough to change the group's trajectory. Both arrive with ordinary quarterly reporting, and neither requires believing anything in advance.
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.