Naver and Kakao are focusing on developing artificial intelligence (AI) tailored to the domestic market. This is in stark contrast to big tech companies like Amazon Web Services (AWS), Google, and Microsoft (MS), which are pursuing strategies to develop multilingual AI aimed at the global market. There are concerns in the industry that due to the relatively small size of the domestic market, monetization through AI may be challenging, potentially leading to limitations in business expansion.
According to industry sources on the 17th, AWS has integrated a voice recognition model called ‘ASR-2.0’ that can speak more than nine languages into its service ‘Amazon Lex’, which enables the creation of conversational AI. It features European-based models that include Portuguese, Catalan, French, Italian, German, and Spanish, as well as languages like Chinese, Korean, and Japanese. AWS is learning the voice patterns of various languages to provide improved recognition accuracy and is enhancing the service to understand regional dialects.
Google launched its conversational AI ‘Gemini 1.5’ in February, capable of understanding 38 languages. It is characterized by its ability to analyze more than 700,000 words and one hour of video content based on various countries' languages. MS unveiled its ‘Phi 3.5’ series in August, which can understand over 20 languages. It can comprehend images, charts, and tables, as well as summarize video content.
Naver and Kakao chose to target the domestic market. Naver Cloud introduced the super-scale AI ‘Neurocloud for Hyperclova X’ for Korea Hydro and Nuclear Power Corporation this month. Naver Cloud plans to build a nuclear power-focused generative AI platform by March next year using data obtained through collaboration. It is known that after training the AI with information about nuclear power plants, Naver Cloud will begin building a Korean language model.
Kakao revealed the performance of its proprietary multimodal language model 'Kanana-v' this month, which specializes in the Korean language and culture. The Kanana model consists of three types of large language models (LLMs), three types of multimodal language models (MLLMs), two types of visual generation models, and two types of speech models. Among these, ‘Kanana-v’ is designed as an MLLM that can well understand unique characteristics of the Korean language and its cultural context.
Industry experts point out that the LLM strategies of Naver and Kakao, which focus on the domestic market, may not be effective for revenue growth. Professor Kim Doo-hyun from Korea University noted, “Naver and Kakao aim to penetrate the domestic market by focusing on the development of Korean-language models with specific niches, using small language models (sLM). However, many corporations frequently engage in global business, and a Korean-specific language model may not be very efficient.”
Concerns have also been raised about potentially missing the right timing for entering overseas markets. According to market research firm IDC, the size of the domestic AI market last year was estimated at 2.6123 trillion won, while the global AI market exceeded 640 billion dollars (about 92.096 trillion won), more than 40 times that of Korea.
Professor Choi Byung-ho of Korea University's Artificial Intelligence Research Institute said, “Currently, domestic AI startups are opting for strategies that involve simultaneously setting up headquarters in Korea, Japan, and the U.S. This strategy allows them to supply the developed AI models to both the domestic and overseas markets simultaneously, enabling rapid growth.” Professor Choi added, “If Naver and Kakao fail to secure profitability through their LLM businesses, they may take a passive approach to entering overseas markets and could miss the golden time for international expansion as competition in the AI market intensifies.”