Dedicated platform for companies in Japan and Europe

Going forward, Fujitsu will accelerate the adoption of the platform for all AI application areas - from enterprise-wide AI solutions to edge and physical AI. Fujitsu will continuously evolve generative AI to adapt to business changes, ensuring its safe and reliable use across all industries and operations. In this way, the platform will support customers' digital transformation and growth. For European companies, this means They can use AI confidently and in compliance with the law, even at the core of production or for sensitive data applications.

Overview of the platform

Companies that use generative AI need an environment that meets sovereign requirements - such as the protection of confidential data and the ability to independently control AI models and agents that are tailored to their processes. However, deploying such an AI platform in a dedicated internal environment is fraught with challenges: These include a shortage of skilled AI engineers, complex operational management, increasing demand for computing resources and minimizing new security risks.

To address these challenges and enable the deployment of highly sovereign AI, including in Europe, Fujitsu will offer a dedicated AI platform that can be used directly on site. The new platform will be deployed on the Private AI platform on Primergy and Private GPT. It combines Fujitsu's AI technologies and platform products, including the Fujitsu Kozuchi AI solution, into a complete package. Through this bundled deployment, Fujitsu significantly lowers the barrier to entry and enables rapid implementation, even without extensive expertise.

Functions of the platform

According to the provider, the platform ensures the secure use of generative AI in a dedicated, closed environment and thus prevents the disclosure of external data. Customers can choose their preferred installation location, including their own data centers or those of Fujitsu, depending on their specific requirements. Fujitsu also offers comprehensive services ranging from infrastructure provisioning to advanced operationalization to fully support customers.

The platform relies on a vulnerability scanner that identifies more than 7700 vulnerabilities, including Fujitsu-defined types, and Guardrail technologies that detect and prevent malicious attacks such as prompt injections, inappropriate output and unexpected behaviors both before execution and at runtime. The automated creation of rules for detected vulnerabilities and their application to the guardrails enable stable, secure and reliable AI operation, even for users without in-depth expertise. In addition, Fujitsu plans to further expand technologies to prevent model-related inaccuracies and thus continuously increase the reliability of the information generated.

The platform is based on Takane, a language model that enables high-precision Japanese language processing and image analysis. This is supplemented by an integrated fine-tuning function that allows business-specific models to be continuously developed. The AI models are deliberately designed to be lightweight, which, according to the provider, reduces memory requirements by up to 94 percent and makes generative AI more widely applicable through more efficient use of computing resources. By integrating modern quantization technologies, the costs for AI operation can be significantly reduced, while at the same time increasing the performance of the models.

The AI agent framework provides low-code/no-code capabilities that accelerate the building of these agents, significantly increasing the efficiency and speed of development for on-site teams. The platform supports MCP (Model Context Protocol) and communication between agents, enabling seamless integration with existing systems and data, as well as the cooperative operation of multiple AI agents for demanding applications. Fujitsu is currently exploring containerizing its proprietary technologies and making them available as on-demand AI agents.

Prior to the official market launch, pre-tests for selected functions, such as internal fine-tuning and model quantization, will be offered from 2 February 2026. The functions will be made available in stages from February, while the official market launch is planned for July 2026.