Kawasaki, January 26, 2026
Fujitsu today announced the launch of a dedicated AI platform that allows enterprises to autonomously manage the entire generative AI lifecycle, including optimal model development, operation, and incremental learning as well as continuous improvement of models and agents. The platform will be rolled out sequentially in Japan and Europe, with preliminary trial registration slated to begin in February 2026.
Moving forward, Fujitsu will accelerate deployment of the platform for all AI scales from large enterprise AI to edge and physical AI. Fujitsu will continuously evolve generative AI to align with business changes, enabling its safe and reliable use across all industries and operations, ultimately fostering customer transformation and growth.
Overview of the new platform
Enterprises using generative AI need an environment that meets sovereign requirements, including the need for protection of confidential data and the ability to autonomously control AI models and agents optimized for their operations. However, deploying such an AI platform in a dedicated in-house setting faces challenges including AI engineer staffing, complex operation management, escalating computing resource requirements, and new security threat mitigation.
To address these challenges and realize highly sovereign AI, Fujitsu will offer a dedicated AI platform that can also be utilized on-premise. Fujitsu will provide the new platform on Private AI Platform on PRIMERGY [1] and Private GPT [2]. The platform comprises Fujitsu’s advanced AI technologies and platform products, including the Fujitsu Kozuchi AI platform. By delivering these as a single package, Fujitsu reduces initial deployment hurdles, enabling rapid adoption even without extensive specialized knowledge.
Platform features
- Infrastructure for dedicated environments● The platform ensures secure generative AI usage in a dedicated, closed environment, preventing external data exposure
● Customers can choose their preferred installation location, including their own data centers or Fujitsu’s data centers, based on their specific requirements
● Fujitsu provides extensive services, ranging from infrastructure deployment to advanced operationalization, to fully assist customers - Generative AI trust technologies● The platform utilizes a vulnerability scanner identifying over 7,700 vulnerabilities (including Fujitsu-defined types) and guardrail technologies to detect and suppress malicious attacks like prompt injections, inappropriate outputs, and unexpected behaviors pre-execution and at runtime
● Automated rule generation for detected vulnerabilities and their application to guardrails enable stable, safe, and reliable AI operation, even for non-specialists
● Fujitsu plans to further strengthen technologies that prevent hallucinations, thereby improving the reliability of generated information - High-precision generative AI models and customization/lightweighting technologies● The platform is built around Takane, [3] a large language model offering high-precision Japanese performance and image analysis, alongside an in-house fine-tuning feature for continuous improvement of business-specific models
● AI models are lightweighted to reduce memory consumption by up to 94%, [4] enhancing generative AI usability through more efficient computing resource utilization
● Quantization technology is integrated to substantially lower AI utilization costs while improving performance - Efficiency technologies for AI agent development● The AI agent framework offers low-code/no-code capabilities, accelerating AI agent construction and boosting development efficiency and speed for on-site teams
● The platform supports MCP (Model Context Protocol) and inter-agent communication, enabling seamless integration with existing systems and data, and cooperative operation among multiple AI agents for sophisticated applications
● Fujitsu is exploring plans to containerize and offer its proprietary technologies as AI agents on demand
Availability
Prior to the official launch, preliminary trials for select features such as in-house fine-tuning and model quantization will begin accepting registrations on February 2, 2026. Features will be progressively rolled out starting in February, with the official launch anticipated in July 2026.
