CONNECT WITH US

Sakura Internet expands GenAI business, estimates need for 10,000 Nvidia GPUs annually

Chiang, Jen-Chieh, Taipei; Jingyue Hsiao, DIGITIMES Asia 0

Credit: Sakura Internet

Sakura Internet is collaborating with NEC to develop a generative AI (GenAI) service platform, integrating NEC's "cotomi" AI technology into its offerings. To support this initiative, Sakura Internet plans to procure 10,000 Nvidia GPUs annually, highlighting the growing demand for GenAI solutions.

The data shared by NEC and Sakura Internet outlines the strategy to establish a GenAI platform reliant on GPU cloud infrastructure. Central to this platform will be NEC's "cotomi," an AI known for its superior Japanese language processing abilities and seamless integration into business systems.

NEC makes a pioneering move as the first supplier of large language models (LLM) on GenAI services for Sakura Internet's platform. Through a foundational cooperation agreement, the two firms aim to develop sophisticated AI capabilities that cater to various industry-specific requirements.

Sakura Internet's platform will provide access to AI applications, foundational LLMs, and an API named Koukaryoku or "High Thermal Power," a retrieval-augmented generation (RAG) vector database.

Beyond its partnership with NEC, Sakura Internet is actively seeking collaborations with both domestic and international companies to broaden the range of foundational models available on its platform. This expansion ensures that all data communication and storage processes are conducted within Japan, thereby bolstering security and reliability, particularly for Japanese manufacturers.

In AI hardware, Bloomberg reports that the company's CEO and founder, Kunihiro Tanaka, has highlighted the necessity of securing approximately 10,000 Nvidia GPUs each year to cope with escalating GenAI demand. Sakura Internet is expanding its data center capacity, with plans to deploy roughly 10,800 GPUs for high-performance servers by March 2028. However, Tanaka also noted that maintaining a capacity to invest in 10,000 GPUs annually will be crucial to meeting continuous demand effectively.

As global competition for Nvidia GPUs heightens, Sakura Internet has accelerated its GPU acquisitions, yet the supply remains insufficient. Current demand suggests a requirement for between 50,000 and 100,000 GPUs, making the slated introduction of 10,800 GPUs by March 2028 potentially inadequate in the face of sustained demand. Tanaka pointed out that a longstanding partnership with Nvidia's Japan branch, established about eight years ago, equips Sakura Internet to manage large-scale orders.

With major players like NTT Data Group, SoftBank, Daiwa House Industry, and Kyocera entering the data center arena, Japan's market is poised for substantial growth. Projections from the Ministry of Internal Affairs and Communications in their 2024 "Information and Communication White Paper" suggest that the Japanese data center market, valued at approximately JPY2.0938 trillion (approx. US$13.6 billion) in 2022, is expected to escalate to JPY4.1862 trillion by 2027.