AI infrastructure businesses require upfront capital. Expensive GPUs and servers must be purchased, and data centers must be built. Recovering the investment takes time. Naver is currently in a phase of bearing costs for future profits rather than generating significant revenue from AI at this moment.
Naver's (NAVER) investment pace is steep. Last January, it built an AI computing cluster with 4,000 units of NVIDIA B200 GPUs. It has agreed to expand the AI factory at the Sejong data center in partnership with NVIDIA and Brookfield from an initial 55MW to 200MW by 2028. This scale can accommodate approximately 100,000 NVIDIA GPUs.
Last year, Naver used about 88% of its 1.3171 trillion won in capital expenditures for data center servers and equipment. Free cash flow dropped from 320 billion won in the first quarter this year to 46 billion won in the second quarter. Even while waiting for AI investment results, operating profit must be evaluated by the market each quarter.
On the 7th, Naver announced that it extended the depreciation period for computer equipment such as GPUs and CPUs from five years to six years upon releasing its financial results. Calculating that equipment will be used for a longer period reduces the amount recognized as an expense annually. This change delays the recognition of approximately 100 billion won in costs per year.
The actual cash outflow does not decrease. The price of already purchased GPUs also remains unchanged. By spreading the equipment cost over six years instead of five, the expenses recorded in current results are reduced, and operating profit has increased. It is akin to receiving the AI investment bill over a longer period. Naver's second-quarter operating profit was 520.3 billion won, a 0.2% decrease from last year. Without the change in depreciation period, the decline could have been larger.
This stands in stark contrast to Kakao's approach. Kakao CEO Jeong Ji-na stated at the second-quarter earnings conference call on the 6th that it would not actively enter the AI infrastructure business, which requires massive capital. The strategy is to focus on models and services while leveraging partnerships with external providers for large-scale infrastructure.
This does not mean Kakao has stopped using GPUs. It is carrying out government GPU acquisition projects and operates its own data centers. However, considering the speed of GPU generation turnover and investment burdens, it intends to be cautious about directly owning large-scale equipment.
GPUs see significant improvements in performance and power efficiency whenever new models are released. Even after introducing the latest GPUs, faster and more energy-efficient products emerge quickly. This is why Kakao hesitates to make large-scale investments.
Amazon reduced the usage period for some of its servers and network equipment from six years to five years, citing the rapid pace of AI technology development. Naver, conversely, extended the accounting lifespan of its equipment. While the technology replacement cycle shortens, the recorded usage period on the books has lengthened. Naver explained that this reflects the actual equipment usage period.
Naver, which chose aggressive investment, instead calculated it would use equipment for one additional year. Since older GPUs can continue to be used in search and AI inference services, it cannot be definitively concluded that the judgment of six years is wrong. Whether Naver's GPUs can truly maintain economic viability for six years, or if this is merely deferring the AI investment bill, will likely be confirmed by future financial results.
in TECH
