Telecoms ramp up AI investment amid questions over returns

SK Telecom's booth at MWC 2025, held March 3 to March 6, 2025, in Spain / Courtesy of SK Telecom
Major telecom companies are increasing capital expenditures as they expand into artificial intelligence (AI) data center businesses, with questions remaining over their ability to recoup the growing investments as construction and operating costs mount.
According to the companies' regulatory filings on Friday, SK Telecom, KT and LG Uplus spent a total of 3.68 trillion won ($2.75 billion) on capital expenditure in the first nine months of this year, up 13.7 percent from a year earlier.
LG Uplus posted the biggest increase, with capital expenditure on a standalone basis reaching 569.8 billion won in the third quarter, up 30.1 percent year-on-year. The increase was largely driven by investment in its AI data center project in Paju, Gyeonggi Province.
The company is building a 200-megawatt-scale facility in the city, with four buildings scheduled to begin operations by 2028. Its first 50-megawatt building is expected to be completed by next June, while the company says it has secured customer demand for all four buildings.
KT's cumulative capital expenditure, including major affiliates, reached 1.599 trillion won, up 20.3 percent. The company plans to invest a total of 6 trillion won in AI transformation infrastructure over the next five years and expand its data center capacity by more than 1 gigawatt by 2031.
SK Telecom's cumulative capital expenditure, including its new AI data center unit SK Hyper, totaled 1.234 trillion won, up 3 percent. However, excluding SK Hyper, spending declined, reflecting the company's strategy of bringing in external investors to share the financial burden of large-scale data center development.
A 200-megawatt artificial intelligence data center is being constructed in Paju, Gyeonggi Province, June 5. Korea Times photo by Kim Jin-wook
The AI investment push comes as telecom companies seek new growth engines beyond their mobile businesses. Demand for computing power and data storage is increasing as companies adopt generative AI and expand their use of AI inference services.
However, industry officials said the telecom and AI data center businesses have different revenue structures.
"While telecom operators generate revenue from a broad subscriber base, returns on data center investments can vary significantly depending on each facility's ability to attract customers, its utilization rate and contract terms," an industry official said.
Data centers require substantial upfront spending on land, construction, power infrastructure and cooling systems, while depreciation, electricity and maintenance costs continue after facilities begin operating.
Of particular notice are depreciation costs for graphics processing units (GPUs). U.S. hyperscalers typically depreciate GPUs over five years in their financial statements, but questions remain over whether this is appropriate, because AI accelerator makers such as Nvidia release new products almost every year.
This means the period during which existing GPUs can handle cutting-edge AI workloads may be considerably shorter than their accounting lifespan, effectively shortening the time companies have to generate sufficient returns on their investments.
During its recent earnings call, SK Telecom said that "a certain level of increase in investment is inevitable," while emphasizing its plan to finance a substantial portion of AIDC development through external funding.
Securing customers before facilities become operational and maintaining high utilization rates will be critical to recovering investment costs. Long-term contracts can provide more predictable revenue, but profitability will also depend on electricity prices, operating expenses and the timing of equipment deployment.