Comparing Naver and Kakao's AI Strategies: Monetization and Investment Efficiency Matter More Than MAU

Reports that the Google app has overtaken the Naver app in monthly active users in South Korea are highly symbolic. However, app user numbers alone do not directly indicate search competitiveness, platform loyalty, advertising revenue, or the success of an AI business.

Naver and Kakao’s AI strategies are also overstated when reduced to a simple choice between “proprietary models versus external models.” Naver is expanding the scope of its control from models to cloud infrastructure and services, while Kakao is focusing on combining powerful external models with its own technology to enable users to complete real actions within KakaoTalk.

How to Interpret Reports of the Google App Taking the Lead in MAU

MAU is the number of unique users who use a service at least once during a one-month measurement period. It is useful for understanding reach, but it does not show how often a user visited, what the user did, or whether the user generated revenue.

App statistics also vary depending on the research firm and aggregation method. Factors to check include the coverage of Android and iOS, the treatment of preinstalled apps, criteria for excluding background activity, sample adjustment methods, and whether multiple apps from the same company are aggregated. The Google app can serve multiple entry purposes beyond search, including voice search, image search, recommended content, and connections to Gemini.

Accordingly, “the Google app ranks No. 1 in MAU” can be read as a sign that Google’s mobile touchpoints have expanded, but it is not evidence that Naver’s search, advertising, and commerce businesses have immediately fallen behind. Multiple quarters should be tracked using the same criteria, while DAU, usage time, search share, transaction conversion, and revenue should be compared together.

Why Search Intent Matters More Than Search Share

Not all searches have the same value. Fact-checking or document summarization can be completed quickly through general-purpose search and generative AI services such as Google, ChatGPT, and Gemini. By contrast, searches closer to action—such as comparing products immediately before a purchase, finding local stores, or checking reservation availability—favor platforms that connect reviews, prices, maps, and payment systems.

What is regarded as Naver’s strength is not merely the volume of Korean-language documents. The key is its structure connecting user discovery and transactions through services such as blogs, cafes, maps, Place, shopping, and reservations. Information about local businesses and user reviews accumulated over long periods may be difficult for general-purpose AI to replace immediately at the same level.

However, these assets should not be viewed as a permanent defensive barrier.

• If review credibility declines or promotional content increases, search satisfaction will fall even if the volume of data is large.
• If generative AI clearly cites sources while comparing information from multiple sites, the need to search within portals may decline.
• External models can also strengthen local data through partnerships, licensing, the public web, and information provided by businesses.
• If AI answers resolve questions directly on the search results page, traffic moving to content pages and advertisements may decrease.

What Naver must protect is not simply search volume, but reliable, up-to-date information and a complete user journey that leads through to reservations and purchases.

Why Naver’s Corporate Value Does Not Rise on AI Investment Alone

Stock-price movements cannot be explained by a single specific cause. Interest rates, market risk appetite, earnings forecasts, regulation, competition, and capital allocation are reflected simultaneously. However, the questions investors seek to answer when a mature platform begins large-scale AI investment are relatively clear.

1. Can it maintain the growth rates of its existing search advertising and commerce businesses?
2. How much will increases in GPU, data center, and R&D expenses reduce operating margins?
3. Can AI search generate new revenue without cannibalizing existing advertising products?
4. Will enterprise AI and cloud revenue grow enough to recoup investment costs?
5. Is there an appropriate balance between investment and shareholder returns such as share cancellations and dividends?

Naver can be compared to a large retailer or an aircraft carrier changing course, but an analogy does not prove causation. Naver can repeatedly sell intangible data and software, whereas retailers have a high proportion of inventory and store costs. The core of the comparison should be limited to their shared position in “a transition period requiring investment in new infrastructure while maintaining existing cash-generating businesses.”

In the market, visibility into the path to recouping investment is more important than the scale of the investment. Even if AI features increase user numbers, economic performance may remain weak if inference costs rise faster than revenue.

Comparison of Naver and Kakao’s AI Strategies

The two companies have made different choices because their platforms have different starting points and revenue structures.

Naver: A Full-Stack Strategy That Expands the Scope of Control

Naver connects proprietary models such as HyperCLOVA X with NAVER Cloud, data centers, and consumer services. This strategy is compelling to customers for whom data location, security, Korean-language and local context, and enterprise-specific customization are important. The sovereign AI market, in which countries and institutions place importance on their own language, regulations, and data control, is also a potential business area.

However, “full-stack” does not mean independently producing every semiconductor and piece of software. Collaboration with external suppliers is necessary for GPUs, key components, and parts of the software ecosystem. Success depends less on whether each layer is internally owned than on how efficiently those layers are integrated and whether paying customers are secured.

Kakao: A Strategy Leveraging Its Distribution Network and End-to-End Action Completion

Kakao announced a strategic partnership with OpenAI and has pursued a direction in which AI goes beyond answering conversations within KakaoTalk to assisting users with tasks. If it can reduce the steps from messages to schedules, searches, reservations, purchases, and ride requests, Kakao’s network of connected services could become an advantage.

However, it would be inaccurate to conclude that Kakao does not invest at all in its own LLM or AI infrastructure. Its approach is closer to a hybrid strategy that develops proprietary AI models and services while also using leading external models. Using external models can accelerate development, but Kakao must manage costs, personal data processing, incident response, and dependence on suppliers.

Metrics for Determining Which Strategy Has Won

A business winner cannot be determined solely by model evaluation scores or app download numbers. The following metrics must be checked repeatedly over the same periods.

In the short term, a strategy using external models may appear faster and less expensive. If usage increases significantly or supplier prices rise, the economics of proprietary infrastructure may improve. Conversely, if technology generations change rapidly and demand is uncertain, large-scale proprietary infrastructure becomes a burden. Neither approach is always superior; outcomes vary according to demand, utilization, and margins per service.

Why South Korean Software Competitiveness Cannot Be Explained Solely as a Stage Following Hardware Deployment

The explanation that GPU and data center investment expands first in the AI industry and application software grows afterward captures part of the trend. However, it is difficult to say that software monetization begins only after infrastructure has been built. Cloud, advertising technology, games, enterprise software, and mobile services are already competing simultaneously with hardware.

The following conditions should be distinguished when evaluating the software competitiveness of South Korean companies.

• Can services proven in the domestic market be expanded into other languages and regulatory environments?
• Can products be sold as repeatable subscription services rather than one-off deployments?
• Can they build APIs and ecosystems for use by overseas developers and enterprises?
• Do they have global sales, technical support, and partner systems?
• Have they incorporated privacy, copyright, and AI safety requirements into product design?

Naver’s overseas AI and cloud business may demonstrate potential, but individual partnership announcements must be distinguished from sustained revenue. Actual competitiveness can be evaluated when verifiable business metrics such as contract size, recurring revenue, customer retention, and profitability are disclosed.

How Should Participation in Public AI Projects Be Interpreted?

Public campaigns, free-use support programs, or consortium projects that use similar names such as “AI for All” may have different participation requirements depending on the announcement date and the administering institution. To assess a particular company’s participation or non-participation and its motives, official announcements, partner institutions, budgets, service periods, and procurement documents must first be reviewed.

Kakao’s participation may help expand user touchpoints, while Naver’s non-participation may indicate a focus on its independent business. However, unless a company has officially disclosed its motives, commercial benefits, technological direction, or promotional objectives cannot be asserted as facts. Participation in a public project is only one clue to strategy and does not prove model competitiveness or long-term profitability.

An Easily Overlooked Battleground: Control Over Agent Transactions

An issue easily missed in simple model comparisons is where an AI actually completes a transaction after understanding the user’s intent. If an AI agent goes beyond recommending a restaurant and handles the reservation and payment, the platform’s revenue structure changes.

Four types of control are important in this process.

1. Intent data: Can the platform determine what users are searching for and intend to purchase?
2. Transaction connections: Can it access sellers, stores, mobility, and payment systems in real time?
3. User consent: Does it transparently obtain and securely manage personal data and payment permissions?
4. Revenue distribution: How is value divided among model providers, platforms, content creators, and sellers?

Naver starts from the connection between search and commerce data, while Kakao starts from its conversational relationship network and access to daily-life services. The final contest is likely to be decided not by more fluent answers, but by the ability to complete trustworthy actions in fewer steps and generate sustainable profit in the process.

Three Possible Development Scenarios

When Naver’s Integrated Strategy Has the Advantage

This would occur if enterprises and public institutions prioritize data control and localization, Naver secures high utilization of its own infrastructure, and cloud and AI revenue grows faster than investment costs.

When Kakao’s Service Strategy Has the Advantage

This would occur if foundation models rapidly become commoditized and less expensive, while Kakao uses KakaoTalk’s distribution power to quickly spread action-oriented AI for reservations, purchases, ride requests, and similar tasks.

When the Two Strategies Converge

Naver may also combine external models where necessary, while Kakao may operate cost- and security-sensitive functions internally, causing both to converge on hybrid structures. In the actual market, various combinations between purely proprietary development and complete external dependence are likely to compete.

Conclusion

The rise in the Google app’s MAU is a warning sign for Naver, but not a declaration of defeat. Naver’s key challenges are the quality of its search and transaction data, recouping its AI infrastructure investment, and expanding in the enterprise market. For Kakao, it is important to use external models while creating experiences possible only within KakaoTalk and controlling service costs and supplier dependence.

The two companies’ strategies are not a simple contest between a heavy approach and a light one. One prioritizes control and the long-term economics of infrastructure, while the other prioritizes speed of distribution and service conversion. The winner will not be the company ranked No. 1 in MAU, but the company that turns AI into recurring revenue and cash flow.