[Edaily Reporter Lee So-Hyun ] Kakao(035720)has set a goal to raise the profitability of its artificial intelligence (AI) business to the break-even point by 2028. The company plans to reduce the investment burden during the initial rollout of its services by adopting an “asset-light” strategy that leverages the cloud rather than making large-scale direct purchases of GPUs (graphics processing units), and to enhance cost efficiency by selectively utilizing both its own models and external models.

Kakao’s third-quarter operating loss in the AI services segment widened to 64 billion won from 57 billion won in the second quarter. As demand for infrastructure to train AI models and operate key services increased, outsourcing and infrastructure costs rose 8% quarter-over-quarter to 240.8 billion won.

Shin Jong-hwan, Kakao’s Chief Financial Officer (CFO), stated during the third-quarter earnings call on the 5th, “We are rigorously managing cost efficiency to ensure that AI investments lead to tangible profit growth,” adding, “In the early stages of service expansion, we are adopting an ‘asset-light’ strategy that leverages GPU cloud services rather than directly purchasing GPUs to minimize the burden of capital investment and the risks associated with facility operations.”

The company is also reducing costs in AI service operations. For simple tasks, it utilizes on-device models; for complex tasks, it deploys its proprietary “Kanana 2.6” model and uses high-performance external models only when necessary. Through “AI Conductor,” which coordinates proprietary and external models, the company aims to avoid allocating high-cost computational resources to every request.

CUA (Computer Using Agent), where AI directly manipulates the browser, also strikes a balance between scalability and cost. Kakao explained that while CUA allows for quick access to various services without the need for separate partner integrations, it incurs higher computational costs because it requires direct analysis and manipulation of the screen. Therefore, the company plans to first validate user demand through CUA and then transition key services to direct integration via A2A (Agent-to-Agent) and MCP (Model Context Protocol) to reduce costs.

Monetization will begin in earnest next year. Kakao plans to validate the potential of new revenue models based on agentic AI in 2027, while also applying AI to its existing advertising and commerce businesses to accelerate revenue growth.

CFO Shin stated, “By 2028, new AI revenue will establish itself as a new growth engine, accounting for a double-digit percentage of total TalkBiz revenue,” adding, “We expect the profitability of the AI business to improve to the break-even point by 2028 through revenue expansion and cost efficiency.”