Fujitsu Focuses on Private AI with Japanese LLM Utilizing Nutanix
Fujitsu is collaborating with Nutanix to bring its Japanese language-optimized large language model (LLM) called Takane to the Nutanix Enterprise AI (NAI) platform. This partnership signifies that Takane is now validated as a certified LLM suitable for deployment on NAI, which is capable of supporting generative AI models both on-premises and across public cloud environments, as well as on the Nutanix Cloud Platform (NCP) that operates in a hybrid multicloud setup.
Advantages of Takane for Japanese Language Tasks
Fujitsu asserts that Takane provides enhanced performance specifically for Japanese language tasks, outperforming general-purpose LLMs. This is particularly relevant given the complex nature of the Japanese language, which includes a mix of character sets such as hiragana, katakana, and kanji. Takane is designed to address unique linguistic challenges, such as:
- Mixed character sets
- Omitted subjects in sentences
- Nuanced honorific expressions commonly used in Japanese business communications
This tailored approach aims at enterprises, primarily in Japan, that wish to adopt generative AI technologies but may struggle with the effectiveness of global models. Additionally, companies may need to run AI workloads in private settings due to concerns about data sensitivity, regulatory requirements, or the need for lower latency.
Deployment and Management of Takane
Fujitsu will initially deploy Takane via its PrimeFlex integrated virtualization platform for Nutanix. This decision supports the establishment of a reliable AI infrastructure, especially for organizations that face limitations on public cloud usage. The collaboration takes hybrid deployment needs into account, where development may occur in the public cloud while production is handled on-premises or at the edge. Fujitsu highlights that compatibility with NAI and NCP simplifies the migration and integrated management of AI applications and data in diverse environments.
As part of its strategy, Fujitsu plans to offer Takane as a managed service beginning in July 2025. This will be in addition to its existing Fujitsu Cloud Managed Service, designed to enhance and manage hybrid cloud operations, which will also encompass environments utilizing Takane on NCP.
Fujitsu Uvance Initiative and Target Sectors
This development fits within the scope of the Fujitsu Uvance business initiative, which aims to tackle societal and business challenges. Particularly, Fujitsu sees Takane as having substantial potential in high-security sectors, such as finance, government, and research and development. In these areas, deploying Takane in private or hybrid cloud settings could provide significant advantages.
Japan’s Commitment to Domestic AI Development
The collaboration between Fujitsu and Nutanix coincides with Japan’s broader initiative to develop domestic LLM capabilities. This movement is fueled by various factors, including linguistic requirements, data sovereignty concerns, and a national economic strategy. Japanese corporations and research institutions are heavily investing in foundational models that are primarily trained on Japanese datasets.
Noteworthy projects include NTT’s ‘tsuzumi’ LLM and SoftBank’s investment in generative AI. SoftBank has also entered a joint venture with OpenAI to create AI agents that cater specifically to Japanese enterprises. Additionally, NEC recently launched its enhanced ‘cotomi’ model, which has gained accolades for its accuracy in Japanese language benchmarks, along with new technologies that potentially double the efficiency of graphics processing units (GPUs) without sacrificing generative AI performance.
The Japanese government has publicly backed the creation of a domestic AI ecosystem, viewing this initiative as vital for national competitiveness. By making Takane available on a hybrid cloud platform, Fujitsu is helping to realize these national goals and providing enterprises in Japan the secure options needed to leverage generative AI within their own infrastructure.
