Tetsuya Yamada

Elyza developed Japanese large language models and enterprise AI applications, building on top of open-source models including Llama to create fine-tuned Japanese language variants with high Japanese language quality, and offering enterprise AI services using its Japanese LLM capabilities for customer service, document processing, and content generation applications. The Tokyo company focused on the specific characteristics of Japanese language including script mixing, honorific speech levels, and domain-specific vocabulary.\n\nThe company raised approximately $30 million in venture funding before being acquired by KDDI, the Japanese telecommunications operator, in 2024. The acquisition gave KDDI an in-house Japanese AI capability to deploy across its telecommunications services and enterprise AI offerings for KDDI business customers, as Japanese telecoms operators compete to add AI features to their enterprise service portfolios.\n\nElyza operated in the Japanese LLM market alongside CyberAgent AI Lab, NTT AI research, and the open-source Japanese AI community centred around the National Institute of Informatics. The acquisition by KDDI reflects the pattern of large Japanese corporations acquiring AI startups to accelerate internal AI capability development rather than building Japanese-language AI expertise from scratch. Japanese LLM development requires significant training data in Japanese and expertise in Japanese language evaluation that international foundation model providers building Japanese language versions of their models must develop separately.