NAVER eases burden of $9 billion GPU investment ... the pros and cons of its 18.7 trillion won guarantee [fn Market Watch]

[Financial News] NAVER has put in place a business structure that will allow it to secure up to $9 billion worth of artificial intelligence (AI) computing infrastructure without directly raising funds to purchase GPUs. Under the arrangement, global asset manager Brookfield bears the investment costs, which is expected to ease the financial burden of a large upfront investment. The 18.74 trillion won payment guarantee provided by NAVER is also a nominal total of supply payments, calculated on the assumption that the entire investment limit is deployed; there are no currently confirmed payment obligations. However, depreciation expenses expected to rise by around 2 trillion won annually and the ability to secure customers are key factors to watch.
Yoo Young-bin, a senior researcher at Korea Investors Service, said in a report on the 29th, “Because Brookfield bears the obligation to fund GPU purchases, which account for the largest share of the investment, the company does not face a large upfront investment burden.” 
NAVER’s global AI factory business is structured around NAVER, NVIDIA Corporation and Brookfield sharing responsibilities. NAVER is responsible for building and operating data centers, while NVIDIA Corporation supplies the GPUs. Brookfield provides the funds to purchase GPUs through an investment vehicle. NAVER plans to secure AI infrastructure with a capacity of 100 MW by the end of 2027 and a cumulative 200 MW by 2028, with plans to scale it up to gigawatt (GW) levels over the longer term.
A feature of the business structure is that Brookfield owns 100% of the investment vehicle, yet the company is included as a consolidated subsidiary of NAVER. This is because NAVER Cloud holds substantive decision-making authority over GPU operations and services. Accordingly, the GPUs will be recognized as property, plant and equipment on NAVER’s consolidated financial statements, while Brookfield’s contribution will be recorded as equity attributable to non-controlling interests.
However, the increase in GPU assets will inevitably bring a higher depreciation burden. Korea Investors Service estimates the GPUs’ useful life for depreciation purposes at five to six years and forecasts that NAVER’s annual depreciation expense will rise by about 2 trillion won after 2028, when the investment is fully deployed. This would amount to an additional expense roughly three times the 680 billion won in depreciation recorded on a consolidated basis in 2025.
The 18.74 trillion won payment guarantee provided by NAVER is also drawing attention. This is the nominal total of payments NAVER Cloud would owe under GPU supply and operations contracts, assuming the entire $9 billion computing infrastructure investment limit is deployed. The actual guarantee amount will vary depending on the scale and timing of GPU investment and the costs incurred. There are currently no confirmed supply payments or other payment obligations, and the actual guarantee period is six years from the date the obligation to pay for supplies begins.
Senior researcher Yoo explained, “NAVER’s payment guarantee covers NAVER Cloud’s payment obligations under its GPU services and operations contracts, while Brookfield’s contributed principal, dividends, return on investment and the GPUs’ residual value are excluded from the guarantee.” In light of this, NAVER is considered to bear limited actual risk in the project.
However, securing customers and maintaining stable contracts will be factors affecting NAVER’s creditworthiness going forward. If customers’ financial condition deteriorates after the GPUs are procured, or if service contracts are too short to allow the investment costs to be sufficiently recovered, the business could incur losses. Retained earnings of the investment vehicle that may accrue to Brookfield are also a factor in credit assessments.
Senior researcher Yoo said, “We will monitor cash flows related to the business as they take shape, focusing on factors including the scale and structure of contracts with external customers, their duration, and the scale of computing infrastructure procured.”


[email protected] Kim Hyun-jung Reporter