Korean Cloud Firms Turn AI Demand Into Revenue as Data Center Expansion Tests Power and Financing

12 Min Read
Korean Cloud Firms Turn AI Demand Into Revenue as Data Center Expansion Tests Power and Financing

A 2 trillion won contract brings the expansion into focus

Naver has signed a computing services contract worth about 2 trillion won with a U.S. AI inference company, while NHN Cloud has sold all capacity reserved for outside customers at its FactoryX Seoul facility for the next five years. KT, meanwhile, plans to spend about 5 trillion won over five years on an expansion aimed at adding 1 gigawatt of AI data center capacity by 2031.

Contents
  1. A 2 trillion won contract brings the expansion into focus
  2. Revenue is rising across different business models
  3. Naver is selling capacity before installing equipment
  4. The financing plan has conditions attached
  5. NHN builds around capacity already sold
  6. Profitability depends on which costs are counted
  7. KT's expansion relies mainly on leases
  8. The opening schedule stretches to 2031
  9. National ambitions meet power and customer constraints
  10. Key Points

The announcements accompanying third quarter 2026 results show how South Korea's cloud companies are beginning to turn demand for artificial intelligence into contracted revenue. They also expose the next challenge: arranging enough electricity, cooling equipment, financing and paying customers to make much larger facilities profitable.

Naver's first AI Factory contract covers 20 megawatts of computing infrastructure and is expected to generate about 2 trillion won over five years. Chief Executive Choi Soo-yeon said on November 12 that commercial services and revenue would begin in the first half of 2027.

NHN Cloud recorded revenue growth of 75.5% from a year earlier, while Naver's enterprise division grew 33% and KT Cloud grew 7.7%. Each company is expanding from a different base: Naver is pursuing international AI customers, KT is combining leased facilities with corporate services, and NHN is building on government GPU projects and private customers such as Krafton.

GPUs, or graphics processing units, are chips that perform many calculations at once. Cloud providers rent access to them so customers can build AI models or run existing ones. Inference, the business of Naver's new customer, means using a trained model to produce answers or other results.

Advertisement

Revenue is rising across different business models

Naver's enterprise division recorded third quarter revenue of 193.3 billion won, up 33% from a year earlier and 25.8% from the previous quarter. Corporate projects, including its Saudi Arabian digital twin business, contributed to growth. A digital twin is a computer representation of a real place or object that can help users plan and test changes.

KT Cloud posted revenue of 268.2 billion won, up 7.7%. Higher occupancy at its Gasan data center increased colocation revenue, which comes from providing space, electricity and related services for customers' servers. Public agencies moving services to the cloud and support for high performance computing also contributed.

NHN Cloud's revenue rose 75.5% from a year earlier and 10.3% from the previous quarter, with the company maintaining profitability. Parent NHN's technology division, which also includes its Japanese AWS resale business, generated 171.7 billion won, up 53.6%.

The business categories cover different activities. Naver's enterprise revenue includes digital twins, while NHN's technology division includes both cloud operations and other technology services. The figures show growth within each business rather than a ranking of revenue from identical services.

NHN's consolidated revenue reached a quarterly record of 758.3 billion won, up 21.2%, and operating profit reached 58.8 billion won, more than double the previous year's level. Games and payments also grew. KT's group revenue, by contrast, fell 1.8% to about 7 trillion won, and operating profit declined 6.1% to 505.3 billion won, with customer compensation among the costs weighing on results.

Naver's first AI Factory customer has contracted for 20MW, equivalent to 10% of its planned 200MW supply by 2028. The agreement includes NVIDIA B300 and Vera Rubin technology. Naver says it is negotiating with more than 10 companies over the remaining 180MW, including large AI model developers.

Choi, Naver's chief executive, described the commercial importance of the agreement during the November 12 earnings call.

This contract is a meaningful deal that confirms actual market demand and early business viability for the AI Factory business.

Naver plans to make equipment investments using Brookfield funding after signing customer supply contracts. The approach ties major spending to booked demand. The company expects early margins to be lower because installing equipment and starting services can leave a gap of about two months, followed by stronger profitability as operations expand in 2028.

Advertisement

Naver also says capacity secured through data center contracts already exceeds its previously disclosed target for 2028. It says it has secured most facilities with liquid cooling available for use in South Korea and is discussing what it describes as Asia's largest server order with NVIDIA. Liquid cooling moves heat away from equipment using circulating liquid, supporting chips packed more densely than conventional air cooling can comfortably handle.

The financing plan has conditions attached

NVIDIA's July expansion announcement, dated July 24, 2026, places Naver's proposed 200MW AI factory at GAK Sejong. It expands an initial 55MW buildout and sets out Naver's longer ambition to deploy 1GW of NVIDIA infrastructure.

The announcement says NVIDIA plans to invest $1 billion in Naver. Brookfield has signed a nonbinding term sheet to provide up to $9 billion, with Naver funding the remaining project costs. NVIDIA's proposed investment is subject to customary closing conditions and Naver securing at least $9 billion in committed project financing, separate from NVIDIA's investment.

A nonbinding term sheet records proposed terms while leaving the parties to complete firm agreements. The financing conditions matter because the 200MW buildout requires funding commitments alongside customer contracts.

Brookfield brings a substantial infrastructure portfolio: the announcement puts its assets under management across AI infrastructure at about $100 billion and its assets managed in Korea at about $12 billion. The planned facility would use Blackwell and Vera Rubin platforms and serve customers in both Korea and the United States.

Naver's plans also connect computing capacity with its own AI products. The July announcement says HyperCLOVA X models will be developed using NVIDIA Nemotron 3 Ultra open models, and it schedules a Korean AI agent platform for the second half of 2026.

NHN builds around capacity already sold

NHN Cloud's FactoryX Seoul facility in Yangpyeong has 7,656 NVIDIA B200 GPUs installed under a government computing resources program. The company says all GPU capacity earmarked for external customers has been sold for five years, with other capacity reserved for internal use.

Private business is also contributing. Revenue from building and operating Krafton's GPU cluster began contributing in earnest during the third quarter. NHN added B300 GPUs at the Gwangju National AI Data Center in August, expanding its newer equipment base.

Advertisement

NHN aims to operate about 100MW of AI data center capacity by the end of 2027, including existing facilities. Its regional expansion includes a commitment to lease 20MW in Pohang, a planned 10MW facility of its own in the Gwangju Research and Development Special Zone, and a separate planned lease of 26MW at Jangseong Pine AI Data Center.

Pohang is scheduled for completion in the fourth quarter of 2027, while the Gwangju facility is targeted for completion by the end of that year. In Seoul's Gasan district and Incheon's Gajwa area, NHN plans to install more than 1,700 B300 GPUs using approximately 6MW of incoming power capacity. Services are scheduled to begin in stages in January 2027, with fuller service expected late that month.

NHN says newly secured facilities will support liquid cooling for Vera Rubin. The company is also preparing to compete for a government GPU project expected in 2027. The proposed government budget includes 3.85 trillion won for about 10,000 GPUs in the Vera Rubin performance class; NHN Cloud Chief Executive Kim Dong-hoon described the anticipated program as approximately 4 trillion won.

Profitability depends on which costs are counted

NHN Chief Financial Officer Ahn Hyun-sik said GPU services were generating profit close to 30% when measured against direct costs. This measure counts costs directly associated with providing the service; operating profit also reflects broader business expenses.

Equipment costs are already entering NHN's accounts. Third quarter depreciation reached 37.8 billion won, up 2% from the previous quarter, partly reflecting B300 equipment installed in Gwangju. Depreciation spreads the cost of equipment over its expected useful life.

NHN has not given a specific revenue forecast for the completed 100MW network, citing the number of variables involved. Its strategy combines government contracts, outside GPU customers and private infrastructure projects to support utilization as facilities open.

NHN is also expanding beyond shared GPU services. Innogrid, which became an NHN Cloud subsidiary in July and entered consolidated results in the third quarter, will support efforts to serve public agencies, defense and finance customers requiring private cloud systems. Such systems give an organization a dedicated computing environment with greater control over access and security.

Advertisement

KT's expansion relies mainly on leases

KT currently operates 152MW and plans to add 1GW by 2031 based on customer demand. Kim Bong-kyun, who heads KT's enterprise division and serves as KT Cloud chief executive, said 80% of the additional capacity would be leased and 20% owned. That translates to roughly 800MW leased and 200MW owned.

KT has identified a main project pipeline of 911MW across 20 sites and another 500MW across 10 reserve sites. Investment decisions have been completed for 200MW, with another 60MW expected to be approved before the end of 2026. KT Cloud also plans design, construction and operation work for 60MW in Gunsan and 50MW in Cheongna, Incheon.

Leasing brings substantial future payment obligations. KT Cloud's ratio of liabilities to equity rose from 35.73% at the end of 2024 to 70.47% at the end of 2025. Total liabilities reached about 1.1 trillion won, with an increase of 497.9 billion won in lease liabilities accounting for most of the annual rise.

Three disclosed leases for facilities not yet occupied carry combined payments with a present value of 1.3393 trillion won. Present value expresses future payments in today's money. Scheduled occupancy dates are October 2027, February 2028 and May 2028, when the associated lease liabilities are expected to enter the accounts.

KT says it is considering financing methods that take debt levels into account. A possible merger involving KT Cloud remains undecided, with another disclosure due by April 22, 2027.

The opening schedule stretches to 2031

KT plans six new data centers through 2029. It is pairing the expansion with more than 90 terabits per second of additional submarine cable capacity and plans to spread cable landing sites beyond their present concentration in Busan.

Corporate demand is growing alongside those plans. KT's cumulative business and AI transformation orders through the third quarter increased more than 20% from a year earlier. Combined quarterly AI transformation revenue at KT and KT Cloud reached 393.3 billion won, up 9.8%.

The main milestones extend across several years:

  • January 2027: NHN begins B300 services in Gasan and Gajwa in stages.
  • First half of 2027: Naver expects commercial AI Factory services and revenue to begin.
  • March 10, 2027: South Korea's special law promoting the AI data center industry is scheduled to take effect.
  • During 2027: KT plans to open Bucheon No. 1 and No. 2; NHN targets about 100MW in operation by year end.
  • 2028: Naver targets 200MW; KT plans openings in Gaebong, Gunsan and Yongin.
  • 2029: KT plans to open Ansan.
  • 2031: KT targets an additional 1GW of capacity.
Advertisement

National ambitions meet power and customer constraints

NVIDIA's October infrastructure announcement, dated October 30, 2025, outlined plans for more than 260,000 GPUs across Korean sovereign clouds and industrial AI factories. Sovereign AI means developing computing capacity and models under domestic control, using local data and serving national requirements.

The announcement described a government deployment growing from an initial 13,000 Blackwell and other GPUs through Naver Cloud, NHN Cloud and Kakao toward up to 50,000 GPUs. Naver separately planned more than 60,000 GPUs, Samsung and SK Group more than 50,000 each, and Hyundai Motor Group a factory with 50,000 Blackwell GPUs alongside approximately $3 billion of investment with NVIDIA.

NVIDIA Chief Executive Jensen Huang described the industrial ambition in that announcement.

Just as Korea's physical factories have inspired the world with sophisticated ships, cars, chips and electronics, the nation can now produce intelligence as a new export that will drive global transformation.

Power availability will shape where that capacity can be built. Korea Electric Power Corporation figures provided to lawmaker Kim Sung-won show that, by the end of September 2026, 313 of 648 completed technical reviews for metropolitan grid impact applications received decisions that power could not be supplied, a rate of 48.3%. Outside the metropolitan area, the figure was 33 of 327, or 10.1%.

The special law enacted in June 2026 provides for coordinated permitting, designated AI data center zones and support for transmission lines, substations and water infrastructure. The government is supporting plans for 8.4GW by 2029, associated with a stated 550 trillion won in private investment. SK Group and AWS are building approximately 103MW in Ulsan, while LG Uplus is developing a facility of about 200MW in Paju.

Customer commitments remain as important as electricity. CBRE Korea put the advance leasing rate for metropolitan capacity due to open in 2027 at 16%. Overseas, CoreWeave reported 1.7GW of active capacity, 4.6GW of contracted power and a customer contract backlog valued at 177 trillion won. For scale, 1GW equals 1,000MW, making CoreWeave's reported active capacity more than 11 times KT's current 152MW.

Naver's signed first customer, NHN's reserved GPU capacity and KT's growing orders give each company a commercial base for expansion. Their next scheduled openings will turn those commitments into a practical test of whether equipment delivery, power connections and customer demand arrive together.

Key Points

  • Naver's first AI Factory contract covers 20MW and approximately 2 trillion won over five years, with revenue expected from the first half of 2027.
  • Naver targets 200MW by 2028; its announced international financing includes conditions and a nonbinding Brookfield term sheet.
  • NHN Cloud aims to operate about 100MW by the end of 2027 and has sold five years of FactoryX capacity reserved for outside customers.
  • KT plans an additional 1GW by 2031, with 80% leased, increasing the importance of managing future rental obligations.
  • Power access is tighter in metropolitan areas, where 48.3% of completed technical reviews received decisions that electricity could not be supplied.
Share This Article

You May also Like