With Naver presenting a "20% or higher" mature-stage margin for its artificial intelligence (AI) factory business, attention is focusing on whether AI factories (AI data centers) can take root as a new revenue source. Because AI data centers require massive capital for graphics processing units (GPUs), power, and data center construction, and carry a heavy depreciation burden, it is not easy to secure actual revenue even if sales grow quickly. In this context, Naver's target of a margin above 20% is seen as showing considerable confidence in the profitability of its AI infrastructure business.
Naver expects the AI factory business it is pursuing with Nvidia to begin generating meaningful sales in the first half of next year. Last month, Naver said it would conduct a $1 billion (about 1.4809 trillion won) paid-in capital increase for Nvidia. The funds will go into building an AI factory targeting 1 gigawatt (GW). Through this, Nvidia will serve as a strategic investor, and global alternative asset manager Brookfield will invest $9 billion (about 12.9 trillion won) to become the largest shareholder of a special purpose vehicle (SPV) that owns infrastructure assets such as GPUs and data centers. Naver will be responsible for operating the AI factory and acquiring customers.
The profitability target for the business structure is also specific. Kim Hee-cheul, Naver CFO (chief financial officer), said on an earnings conference call on the 7th of this month, "All sales, expense, and resulting margin from the computing business are captured by the AI factory operator (Naver)," adding, "To run the computing business, funding is required for massive computing assets, but rather than Naver directly raising that funding, Brookfield raises it through an SPV and supplies Naver." He continued, "The operator's (Naver's) expected return will vary depending on maturity," and "Margins may be low in the early stages, but as the business matures, we basically expect double digits and further, margins above 20%."
Whether Naver's target is excessively high can be compared with global AI cloud companies. AI-focused cloud firm Nebius is targeting a 20%–30% margin on an adjusted EBIT (earnings before interest and taxes) basis that reflects GPU depreciation. In its shareholder letter for the first quarter of 2025, Nebius said, "Assuming a conservative four-year depreciation schedule in the medium term, in our base scenario we expect to achieve multi-billion dollars in sales with adjusted EBIT margins in the 20%–30% range." In materials for the first quarter of 2026, it said the adjusted EBITDA (earnings before interest, taxes, depreciation and amortization) margin had risen to 45%, calling this "on track toward a 20%–30% EBIT margin."
EBIT indicates operating profitability that reflects depreciation for GPUs, servers, and data centers, while EBITDA shows operating profitability excluding depreciation. Because the AI infrastructure business is capital intensive, the gap between the two metrics can be large. That is because the AI infrastructure business requires massive upfront investment in GPUs, servers, and data centers, generating substantial related depreciation. For this reason, it is unreasonable to evaluate the profitability of the AI infrastructure business solely on an EBITDA basis.
AI-focused cloud provider CoreWeave also reported adjusted EBITDA of $1.51 billion (about 2.1375 trillion won) in the second quarter of 2026, but adjusted operating profit was only $128 million (181.2 billion won). Based on sales of $2.575 billion, the adjusted EBITDA margin was about 59%, but the adjusted operating margin, reflecting depreciation of GPUs and data centers, was about 5%. In the end, even if AI computing demand surges and sales and EBITDA rise rapidly, it shows that actual operating profitability, after reflecting depreciation of investment assets, does not improve at the same pace.
If Naver's cited 20% is on an EBITDA basis, it would not be an especially high target compared with global AI infrastructure players, but if it refers to an operating margin that reflects depreciation of GPUs and data centers, it could be seen as quite aggressive.
Of course, there are limits to simply comparing Naver's margin target with global companies. That is because Naver has not specified whether the 20% or higher margin it presented is based on EBITDA, adjusted EBIT, operating margin, or another metric. The business structures also differ. Nebius and CoreWeave are AI infrastructure specialists that provide AI cloud services based on GPUs and data centers. Naver, in addition to its own AI models and cloud, also operates existing platform businesses such as search, advertising, and commerce. There is also a difference in the revenue model in that computing capacity secured at the AI factory can be used for its own services and for enterprise AI business.
Brokerages are focusing on the possibility that the AI factory will become Naver's new growth engine. Ahn Jae-min, an analyst at NH Investment & Securities, said, "As the AI data center business gets underway and related sales begin to occur from 2027, the new business is expected to gain recognition for its value," adding, "We estimate Naver's AI factory sales will grow from 560 billion won in 2027 to 6.4 trillion won in 2030, with operating profit rising from 31.7 billion won in 2027 to 720.5 billion won in 2030." Calculated as an operating margin, NH Investment & Securities' projections come out to the low 10% range.
Ultimately, the actual profitability of the AI factory will likely depend on how efficiently utilization, GPU costs, electricity, depreciation, and customer acquisition costs are managed. For Naver to actually achieve its margin target of over 20%, the key variables will be how quickly it absorbs the initial investment burden and how much it can raise the utilization of GPUs and data centers.
Lee Kyung-jeon, a professor in the Department of Big Data Applications at Kyunghee University, said, "Naver will operate the AI factory competitively and well, but how effectively it secures customers (demand) will determine future AI factory revenue."