Hello, I'm Jeffrey from Kakao's IR. We will now begin Kakao's 2026 Third Quarter Earnings Conference Call. Today, I am joined by Shina Chung, the CEO, and Jaden Shin, the CFO. Please be reminded that the earnings results are consolidated estimates under the K-IFRS basis and are subject to change upon the auditor's review. Also, forward-looking estimates are based on assumptions, so actual results may differ from figures included in today's presentation. Now, we will have a presentation from CEO Shina Chung.
Good morning. This is Shina, CEO of Kakao. In the global B2C AI market, the emergence of new technologies and services is bringing important changes to the competitive landscape for AI agents. In the past, the competitive factor was how accurately AI could answer users' questions and provide useful information. Recently, however, the execution capabilities of AI performing real tasks on behalf of users have been advancing rapidly.
As a result, it is becoming increasingly important how many different AI services can utilize and how much real value it can deliver in users' everyday lives. Over the past year, Kakao has introduced a range of AI features and services, experimenting with how KakaoTalk can evolve in the AI area and connect to new growth drivers. Through these experiences, we confirmed the potential to expand AI services centered on KakaoTalk. However, we concluded that to respond to the rapidly changing global competitive environment, we need to advance our existing strategy to the next level. Accordingly, Kakao will evolve its strategy toward connecting the user experiences of various AI services into one and rapidly expanding the execution capabilities of AI. At the center of this strategy is Kanana Space, which we unveiled at the recent if(kakao).
Kanana Space is a new AI layer that users can enter from anywhere in KakaoTalk with a light two-finger swipe. Previously, users had to find and navigate to the features or services they needed on their mobile screens to get things done. Going forward, we expect that users will simply make a request by voice or text in a single AI hub called Space, and the AI agent will carry it out on their behalf. At the simple feature level, users can change basic KakaoTalk settings, such as font size. Beyond that, AI agent can also execute frequently used services within KakaoTalk on users' behalf, such as Talk Gift and money transfers.
Also, even when users ask a simple question in Space, it identifies context and intent, and depending on the difficulty of the query, calls on ChatGPT for Kakao, Kanana in KakaoTalk, or even Kanvas to provide an answer or perform the requested action. The tasks that can be executed through Space will not be limited to KakaoTalk and are planned to be integrated with external services going forward. In this way, we expect Space to connect the fragmented AI features and services previously provided by individual services into a single touchpoint, delivering an integrated AI services to users. When the everyday conversational context of KakaoTalk is added through our own on-device model, Space can deliver an even more personalized experience. KakaoTalk holds a wide range of information closely tied to users' daily lives, such as appointments, schedules, and interests accumulated in conversations with family, friends, and colleagues.
Within the scope users have consented to, Space can understand this context, find the information they need, and connect it to the right next section. This reduces the need for users to explain their situation from scratch every time. As this experience accumulates, we expect a new usage habit to take hold where users naturally turn to AI agents within KakaoTalk. Kanana Space will launch a pilot service in December, starting with users who have pre-registered, and begin evaluation in a real usage environment. We will then improve the service quality based on user feedback and expand the range of tasks AI can perform while scaling up the user base in earnest next year. To build a mass-market AI service that 50 million users can trust and safely use in their daily lives, Kakao has prioritized A2A-based integrations with external partners.
However, during the actual integration process, we found that there were more variables than expected. These included not only partners' development resources and timelines, but also the business value of AI agents, which still requires validation and alignment of interests. Recently, global AI agent services have been demonstrating the ability to perform tasks on behalf of users across various services without separate partnerships through CUA approaches that control VMs or local browsers. In fact, we are also seeing active usage among some early adopters. The CUA approach clearly has the advantage of enabling broader and easier integration with long-tail services. However, the short-head players holding meaningful positions in each vertical have already begun restricting access by AI agents through browsers. In addition, as browser-based execution allows greater flexibility, token consumptions also rise significantly, and the success rate and the stability of actual tasks will still need further advancement.
Furthermore, to implement an end-to-end process through payment via CUA, a privacy protection framework tailored to Korea's local environment must be established. Taking into account the scope and speed of expansion of task AI agents can confer, the value of the agentic experience users feel as a result, and the sustainable cost structure for use by the entire population, the optimal approach for Kakao is to pursue direct integration and CUA in parallel. Using the CUA approach, we will rapidly expand the ecosystem across a variety of long-tail services and begin validating user demand and business value. At the same time, based on validated demand and business value, we will continue to pursue direct integrations with major services in each vertical through partnership-based A2A and MCP as before.
Through ongoing collaboration with key external partners, Kakao was able to identify in detail prerequisites for agentic AI integration and also confirm the limits of resources partners can commit. Accordingly, we have built technical foundations such as Agent Builder and PlayMCP to minimize the direct resources partners need to invest. In addition, we have also secured our own model specialized agentic actions and browser control capabilities on par with global competing services. On efforts over the past year, where meaningful process through which Kakao accumulate the technologies and experience needed to build an agentic AI ecosystem. At the same time, it was an opportunity to confirm the strong needs of partners who want to use KakaoTalk as a key user acquisition channel in the AI era.
Going forward, we will rapidly expand the execution scope of AI agents and the service ecosystem so the users in Korea can experience the new value of agentic AI in their daily lives. There are services where you can see how Kakao's browser control technology is actually implemented in Kanvas, which we unveiled at the recent if(kakao). Kanvas is Kakao's first AI service designed as a productivity tool. Unlike the VM approach, which operates a browser on a separate virtual computer, Kanvas uses a local browser control approach that runs and controls the browser within the user's PC environment. With Kanvas, user can search for information and organize materials and even automate the repetitive tasks.
For example, an office worker can entrust the agent with the service of task previously handled by switching across multiple programs, from gathering work files scattered across their PC to finding necessary information on the web, writing reports, and sharing with the colleagues. Users can also easily create specialized agents with simple natural language instructions, and we plan to introduce an agent marketplace where users can share the specialized agents they have created with anyone. As Kanvas is integrated with KakaoTalk, users can give tasks to agents in the familiar chat window and receive the results of completed work back there as well. Through this, we aim to show the potential for KakaoTalk to expand beyond the space for everyday communication into touchpoint for execution, where AI carries out users' various work and tasks on their behalf. Kanvas already began internal beta testing last week.
Kakao will launch the pilot service of Kanvas Space in December and is also preparing the official launch of AI for All and Kanvas within this year. In addition, active users of our already launched AI services are also trending upward. As we shared on last quarter's earning call, we are fully on track to achieve our goal of 10 million monthly active AI users by end of this year, and we are building the foundation to secure more than 20 million daily active AI users by 2030. As CEO, I humbly accept the painful assessment that the pace of AI service adoption this year has fallen short of the capital market's expectations. However, amid uncertainty created by the great wave of AI, Kakao has continued a variety of experiments and attempts to position itself as a winner to sustain its high growth.
During this period of governance transition, the gap between the market's current valuation of Kakao and the future growth potential we see internally, based on our fundamentals, has become wider than ever. Accordingly, as disclosed just a moment ago, we have decided to repurchase and cancel KRW 150 billion worth of treasury shares. Beyond the shareholder return, Kakao believes that the most important thing is to concentrate all of the resources on rapidly introducing AI services that exceed the market's expectations. We will expand our AI services with greater speed and live up to our shareholders' expectations through tangible business results and enhanced corporate value. Next, Jaden, our CFO, will present the 2026 third quarter financial results.
Hello, this is Jaden. I will now present Kakao's consolidated financial results for the third quarter. I will now present Kakao's consolidated financial results for the third quarter. Consolidated revenue for the third quarter was KRW 2,086,700,000,000 , up 6% YoY and down 1% QoQ. Platform revenue reached KRW 1 ,235,000,000,000 , growing 17% YoY and remaining broadly flat QoQ. First, revenue for Talk Biz, Kakao's core business, was KRW 632.9 billion, up 13% YoY and down 2% QoQ. Third quarter Talk Biz advertising and subscription revenue reached KRW 391.5 billion, up 11% YoY and down 2% QoQ. Starting with Business Message, revenue grew 7% YoY despite a high base from the government's consumer relief coupon program in the same quarter last year. Excluding this one-off base effect, advertiser demand for Business Message remains solid.
Business Message has firmly established itself as a core communication channel that meets advertisers' CRM needs. We are also strengthening its growth foundation by introducing various message products that improve users' ad experience and advertisers' campaign efficiency. With YoY revenue growth already accelerating again in October, we are confident that Business Message can return to double-digit growth in the fourth quarter. Moving on to Talk DA, revenue grew 35% YoY, supported by broader advertiser demand and a more diverse product lineup. Since the launch of feed ads, we have continued to attract small and mid-sized advertisers that had not previously advertised with Kakao. We have also introduced product catalog ads and engagement ads, making it easier for advertisers to run campaigns and addressing a wider range of marketing needs. These efforts continue to attract new advertisers and encourage existing advertisers to increase their marketing spend.
With this strong growth foundation, we expect Talk DA to comfortably maintain double-digit growth in the fourth quarter, despite a high base from the launch of feed-based services in the fourth quarter last year. Moving on to commerce, third quarter Talk Biz commerce GMV was KRW 2 ,300,000,000,000 , up 13% YoY and 7% QoQ. Starting with Talk Gift, performance benefited from Chuseok shifting from the fourth quarter last year to the third quarter this year. We strengthened our product lineup, particularly in fresh food and health food, and actively ran holiday promotions. This Chuseok, we also attracted consumers looking to purchase holiday gifts to deliver in person, further broadening our buyer base. As a result, Talk Gift recorded its highest-ever monthly GMV in September. Self-purchases on Talk Gift also continued to grow through regular promotions such as For Me Week.
In the third quarter, the number of self-purchasing customers grew 23% YoY, while self-purchase GMV increased 33%. The Chuseok peak season, together with expanding self-purchases, drove 13% YoY growth in total Talk GMV. For Talk Store, we expanded collaborations with major brands across various categories, including home appliances and food, and broadened purchase touchpoints through live commerce and influencer deals. As a result, third quarter Talk Store GMV grew 8% YoY. Beyond gifting demand, which has historically driven GMV growth, our commerce business has been actively addressing a wider range of purchasing needs since last year, broadening its role as a commerce platform. Encouragingly, self-purchase accounted for around 30% of total commerce GMV in the third quarter. We will continue expanding our products and services into everyday consumption, enabling customers to enjoy a wider range of shopping experiences within KakaoTalk.
For reference, commerce revenue was KRW 241.3 billion, up 16% YoY, supported by higher 1P sales during the Chuseok peak season, and remained broadly flat QoQ. Platform and other revenue reached KRW 602.1 billion, up 21% YoY and 3% QoQ. Starting with mobility, steady revenue from its existing businesses was complemented by strong performance in new businesses, supporting continued YoY revenue growth. For Kakao Pay, growth slowed QoQ amid stock market uncertainty. On a YoY basis, however, both revenue and operating profit grew significantly, supported by balanced growth in financial and payment services. Pay also recorded a double-digit operating margin for the third consecutive quarter. Moving on to the content segment. Third quarter content revenue was KRW 851.7 billion, down 6% YoY and 2% QoQ. First, story revenue was KRW 200.6 billion, down 17% YoY and 5% QoQ.
Piccoma and Kakao Entertainment's story business will maintain their focus on profitability while sourcing new IP and extending the life cycle of existing IP to build medium-to-long-term growth drivers. Next, music revenue was KRW 508.3 billion, down 10% YoY and 9% QoQ. Despite solid MD and licensing revenue, including aespa's IP collaborations, revenue declined due to fewer album releases by major artists and fewer large-scale concerts. Lastly, media revenue was KRW 142.8 billion, up 49% YoY and 44% QoQ, reflecting higher revenue recognition based on production progress for major titles. Next, regarding operating expenses, third quarter operating expenses were KRW 1,846,400,000,000 , up 6% YoY and 1% QoQ. Labor costs increased 4% YoY, reflecting one-off performance bonuses at certain subsidiaries. As we maintained a conservative hiring stance, they remained broadly flat QoQ at KRW 446.2 billion.
Marketing expenses were KRW 89.7 billion, down 2% YoY and 7% QoQ, as the content segment continued to focus on efficient marketing spend. For reference, marketing expenses accounted for 4% of consolidated revenue in the third quarter. Cost of revenue was KRW 790 billion, up 13% YoY and 6% QoQ, reflecting higher media revenue and expanded 1P sales in commerce. Outsourcing and infrastructure costs declined 2% YoY as fewer large-scale concerts in the music business reduced outsourcing costs. On a QoQ basis, however, costs increased 8% to KRW 240.8 billion, reflecting higher infrastructure demand for AI model training and key service operations. Depreciation and amortization expenses were KRW 188.8 billion, down 3% YoY and 1% QoQ. The decline reflected cost efficiencies from reduced reliance on external data centers and greater utilization of our own data centers, as well as lower amortization of music-related intangible assets.
Consequently, third quarter consolidated operating profit was KRW 240.3 billion, up 7% YoY and down 13% QoQ, with an operating margin of 12%. For reference, standalone operating profit for the third quarter was KRW 107.8 billion, up 17% YoY. The standalone operating margin improved by approximately 1% point up to 17%. Moving on to non-operating item and net income. Third quarter consolidated net income was KRW 177.7 billion, down KRW 15.2 billion YoY, but up KRW 159.7 billion QoQ. The YoY decline was due to a non-operating loss of KRW 16.7 billion compared with non-operating income in the same quarter last year. Last year's results included gains from the sale of stakes and associates. This quarter, lower exchange rates also led to higher foreign exchange losses on foreign currency deposits and trade receivables held by our subsidiaries.
On a QoQ basis, income tax expense and losses from discontinued operations were significantly lower than in the second quarter, when the items reflected the simplification of our governance structure. As a result, net income improved substantially QoQ despite the non-operating loss. Lastly, total CapEx for the third quarter was KRW 186.3 billion, remaining at a similar level YoY and QoQ. This consisted of KRW 140.4 billion in PP&E and KRW 46 billion in intangible assets.
[Non-English content]
Now Q&A session will begin. Please press star one, that is star and one, if you have any questions. Questions will be taken according to the order you have pressed the number star one. For cancellation, please press star two, that is star and two on your phone.
[Non-English content]
The first question will be provided by Junhyun Kim from HSBC. Please go ahead with your question.
[Non-English content]
[Non-English content]
Thank you for taking my question. I would like to ask you two questions this morning. Recently, we've seen global AI companies expand the scope of their agentic AI. So I would like to understand as to what competitive edge Kakao has vis-a-vis these other global players. I would like to first understand as to the key rationale behind you implementing a KakaoTalk-centric and AI-centric KakaoTalk strategies, and I would like to understand how your service offerings will be different from that of the global players. What is the differentiating factor that Kakao can offer to the market? And I would think that one of those competitive edge would be your partnership with the outside external partners. So if you could provide some insight on that would be helpful.
And also because Kakao is mostly considered as a messenger platform, would like to understand as to what your plans and strategies are in expanding your user base. Second question, would like to understand as to what your plans are relating to monetization of AI. I would think that because your AI business takes up a big portion of your key business domain, strengthening the competitiveness of ad business must be one pillar, and the other will be creating new opportunities. So would like to know as to how these different effort is going to contribute going forward. Therefore, what are some of the new monetization models that you are currently considering? Do you believe going forward that it will be from the advertising business or from the new monetization model that you think is going to make bigger contribution?
[Non-English content]
This is Shina. I will respond to the first part of your question relating to under the current global competitive environment, what is the rationale behind us having that focus on KakaoTalk? As you know, many AI agents are currently being launched across the world, and the first step to personalize is to have the agents first understand the users. Way of doing that is to integrate with external services such as emails, calendars, and messengers. So if you look at AI standalone apps, it must be connected to external platform, and the experience thus is going to be limited subject to the scope and information access that the platform allows.
[Non-English content]
But the starting point is different for KakaoTalk because context is already there, what users are interested in and the relationship that they have. So even with fast advancement of AI technology and execution capabilities, Kakao's unique moat is the relationship network built by 50 million users through their engagement over the long period of time, which cannot easily be mimicked by others.
[Non-English content]
When we bring memory and preference based on users' consent, even though AI service and agents may undergo changes, there's no need for users to explain themselves every single time because AI will inherently understand the user, helping the user with suggestions and leading to actions.
[Non-English content]
We believe KakaoTalk-based relationship network can go beyond communications between people and expand to communication and collaboration between agents. Conversations within KakaoTalk is a process of making time commitments to friends and families, align each other's schedule, and making decisions together. This mimics the way that AI agents can collaborate in checking the information and coordinate in different terms and conditions on behalf of the users. If AI can understand the context of conversation and dialogue that takes place and actually perform the task, KakaoTalk will evolve into a space where not only users and agents interact, but where agents will collaborate to perform work on behalf of users.
[Non-English content]
Relationship network can facilitate the spread of AI usage. If one shares the output of AI with friends and families, then others will start to use it, even people who have not actively used such AI services. With more people using the agent, more opportunities for agents to collaborate and communicate with each other will take place, and there will be more tasks that AI can handle, which will lead to higher frequency and scope of use. We expect to be able to establish a virtuous flywheel.
[Non-English content]
Responding to the next question about our monetization strategy based upon our AI-centric advertisement approach, as AI service in KakaoTalk adoption increases, we expect new opportunities in advertisement monetization as well. If interaction between the user and the AI grows, we can start to understand the intent and context much better, which was difficult to do in the past. Unlike the conventions of search ad, AI service will be able to recommend products and services at the right timing without user engaging in the act of searching. We believe that this would create new demand that Kakao previously did not have, but which is similar to search ad.
[Non-English content]
With more AI service and agents being embedded inside KakaoTalk, there will be much more touchpoint for search and recommendation, and we expect that this will create new advertisement inventory and revenue opportunity as well.
[Non-English content]
Also, we will use AI technology to strengthen advertisement business and enhance advertisers' performance using machine learning to scale up recommendation and targeting. We will also complete common memory infrastructure by the first quarter of next year, which will enable AI to understand and utilize users' behavior and experience-related information. Using this data, we will be able to offer more precise ad recommendation that considers even the preferences and purchase intent of the users, going beyond simple usage of track records on purchases and click-through rates.
[Non-English content]
In December, AI Crew will go through closed beta testing, which is an agent that supports advertisers to manage and execute their advertisement. It will be based on conversational interface, so when advertisers enter their business target, AI will suggest the most befitting product and services as well as most effective strategy, supporting the whole process of setting up the ad to performance analytics. With this, even smaller and mid-sized companies can run their advertisement more efficiently, and overall ad efficiency will be enhanced. Through this approach, we are going to expand on onboarding of new advertisers as well.
[Non-English content]
Based on this strategy, we will be able to create new monetization opportunities starting next year and drive revenue growth moving beyond the growth of simply expanding ad inventory.
[Non-English content]
Next question, please.
[Non-English content]
The following question will be presented by Dongwoo Kim from Kyobo Securities. Please go ahead with your question.
[Non-English content]
Thank you for taking my question. First question relates to your outlook for the performance for your ad and commerce business in the fourth quarter. I would like to know whether you believe that you will be able to achieve the annual guidance that you shared in terms of your revenue growth rate and OPM. Second question is that the global big tech companies are making significant investment into their AI model and agent development. I would like to know that with Kakao's own AI technological competitiveness, do you believe that you will be able to achieve ample amount of competitive moat? The second part of the question is that going forward, as there is going to be more AI service users, there is going to be higher cost pressures as well. What are your plans to control that investment-related pressure?
Do you believe that from mid to long term, you will be able to ensure AI business profitability?
[Non-English content]
This is Jaden responding to your question. For Q4 on the performance outlook, first, if you look at the advertisement business, it is correct that the high level of base effect from the previous year on the display ads will work as a bit of a pressure. However, we expect to be able to see a recovery in the top line growth of the Business Message, and we will be able to maintain a double-digit growth rate on a YoY basis. In terms of the commerce business, there is going to be a base effect because Chuseok last year, Chuseok, which is the peak season, was located in the fourth quarter. So we do expect there to be some of a base effect.
[Non-English content]
Now, broadly speaking, we see the growth from the platform segment quite solid, but it is true that on the content business side, there is a delay in the recovery trajectory. So right now for our content business, rather than focusing on growing the top line, we are more focused on defending the bottom line. So we expect the trend to be quite similar in the fourth quarter compared to the third quarter, which means that we are looking at a limited improvement in terms of profitability.
[Non-English content]
That being said, on a Kakao standalone basis, operating profit for three consecutive quarters this year have continued on with a double-digit YoY uptrend. Hence, going forward, we believe this trajectory to continue. Therefore, if you look at Kakao on a separate basis and platform business, growth is continuing. Therefore, we believe that it is amply achievable, meaning the guidance that we have shared at the beginning of the year, which is 10% of consolidated revenue and 10% guidance on the OPM. We expect that is amply achievable.
[Non-English content]
Regarding the next question, I will address the technical side regarding how to secure cost efficiency while expanding AI-related investment, and I will turn it over to Jaden, who will talk about the investment plan and the financial outlook. Kakao's AI model strategy is not to use high-performance model for all of the requests, but to select the right model depending on the type and the level of requests, and use only right-sized amount of computational resources. Now, this is to drive both quality and efficiency of our service offerings.
[Non-English content]
We expect most of the demand on the AI agent is going to be in relation to everyday tasks rather than highly difficult coding or any request that requires specialized level of analysis. So we are going to continue to evolve Kanana on-device model so that most of the simple and easy requests could be handled on-device.
[Non-English content]
For example, summarizing and organizing conversations and Voice Talk dialogue, which does not directly generate revenue, are still very convenient features for our users, which will be dealt with on-device so that AI experience can be used by as many people as possible without giving them the burden of having to pay for such services. Also, Kanana's on-device model has multimodal capabilities, capable of understanding the voice of the users, and so it will be able to support various modes of communication going forward.
[Non-English content]
On the other hand, complex requests will be routed to the server as it requires agentic-based action and not a simple AI function. But majority of such requests will also be handled by our own proprietary model, Kanana 2.6 flagship model. We recently unveiled this model and we were able to demonstrate that it reaches the level of global frontier models in terms of the agent capabilities, as it can choose the tool based on the user's request, call on the relevant APIs, and perform the follow-up task after receiving the output.
[Non-English content]
In the past, we had to design processes in advance in order to connect requests to actual performance of tasks, but now the model decides which function is required based on the objective that is given and performs the next step. We have the technological base to respond to various requests without depending on predefined scenarios.
[Non-English content]
We will also use high-performance third-party model if necessary on top of Kanana 2.6 model. Kakao's foundational technology, AI Conductor, orchestrates between proprietary model and third-party model for complex processes required to complete one request. This can help minimize cost and response delays without having to use massive amounts of computational resources.
[Non-English content]
As mentioned before, browser control can access variety of services without any separate partner integration, but it generates huge amounts of computational expense in the process of setting and maneuvering through what is on the browser screen. If you are using browser to, let's say, go up to the payment stage, cost per AI call or retrieval will be around KRW 800-KRW 900. But if direct integration through agent to agent, A2A, and MCP is enabled, it can go down, that cost level can go down to around KRW 5 level.
[Non-English content]
Ultimately, underpinned by technical competitiveness that can both drive AI service quality and efficiency, even if the user base expands to include the entire people of Korea, we will develop a foundation where cost efficiency is ensured.
[Non-English content]
Moving on to the answer on the AI investment update as well as our profit outlook going forward. In terms of the financials, we are rigorously controlling cost efficiency to make sure that AI investment actually leads to profit increases. In the very beginning of this AI spread of AI service adoption, in order to minimize the pressures of capital investment and also risk that may arise from operating the facilities, we are taking on an asset-light strategy using the GPU cloud rather than directly purchasing the GPU chips.
[Non-English content]
From the end of this year, we will see expanded user base for the AI service, and also we believe that we will be able to see and our AI services will be able to demonstrate its core service competitiveness with the increase in the usage, repeat usage. In the initial phase, there may be cases where cost may be incurred depending on the necessity, but from a mid to longer term perspective, we are going to monitor our cost efficiency very closely so that there is no linear and proportionate increase in the cost in step with the rise in the number of users.
[Non-English content]
Starting next year, we are going to be validating new and different business models and revenue models that incorporate the use of agentic AI, and we will very actively use AI in our ad and commerce business so that we can speed up our top-line revenue growth. Underpinned by these efforts, as mentioned in our previous earnings conference call, in 2028, we expect that new AI revenue will account for a double digit compared against the Talk Biz revenue as a whole, and we think that it will establish itself as a key lever behind new growth.
[Non-English content]
Driven by such revenue expansion and cost efficiency, we expect our AI business performance to improve to the level of break-even point.
[Non-English content]
This brings us to the end of the third quarter 2026 earnings conference call of Kakao. For any of the unanswered questions, we at the IR team would be happy to assist you. Thank you.