FSAS Technologies and NTT West announced that they have completed operational verification for the stable operation of the purely domestic LLM “tsuzumi 2” in an on-premises environment, and will begin offering a joint solution starting in September 2026. Behind this is the growing demand for “sovereign AI” in fields such as government agencies, finance, and manufacturing that handle highly confidential data, driven by the desire to securely manage data and AI models domestically without relying on external clouds.

Completion of Joint Operational Verification by Fujitsu Group and NTT West

FSAS Technologies and NTT West announced on August 26, 2026, that they have completed operational verification for stably running “tsuzumi 2,” a purely domestic large language model, in an on-premises (in-house) environment without the use of the cloud. This solution is scheduled to be launched in September 2026, with NTT West acting as the primary contractor. In recent years, the integration of generative AI into various business processes and business automation combined with existing core systems have been progressing at a rapid pace. However, under formats where sensitive data and prompts are sent to external AI services, it has been difficult to safely handle critical confidential information essential to corporate activities. To address this, thorough verification work was conducted to closely integrate the hardware infrastructure maintenance and operation technologies of both companies with the latest AI model to ensure practical-level operation. Through this collaboration, it has become possible to rapidly deploy a generative AI execution environment tailored for Japan that combines advanced security with stable processing performance.

The Shift Away from Cloud AI and Demand for Sovereign AI

Organizations required to implement strict security measures, comply with governance, and manage data residency—such as government agencies, research institutions, financial institutions, and manufacturing companies—have faced a serious barrier in adopting cloud-based generative AI services due to concerns over information leakage risks. Naturally, in order to maximize AI performance, it is essential to utilize detailed and highly critical internal data, such as research data, design information, customer personal information, and production data. However, with cloud AI services operating over the internet, data is sometimes sent to servers overseas, creating an “invisible risk” where tracking and the locus of administrative responsibility in the event of an incident remain ambiguous. Against the backdrop of such social demands for ensuring economic security and data sovereignty, demand for “sovereign AI”—which manages everything from AI models to accumulated data and the underlying infrastructure entirely within one’s own country or corporate environment—is expanding at an unprecedented and rapid pace.

Technical Features of the Domestic LLM “tsuzumi 2” and On-Premises Infrastructure

Lightweight design running on a single GPU and high Japanese language processing performance

tsuzumi 2, a home-grown LLM developed by NTT, features a lightweight design with approximately 30 billion (30B) parameters, yet boasts top-tier Japanese language processing performance compared to models in the same size class. Generally, operating ultra-large foreign models ranging from billions to trillions of parameters in-house requires multiple high-performance GPUs and complex cluster configurations, and it is not uncommon for initial hardware investments to reach tens of millions of yen. In contrast, tsuzumi 2 is designed to handle inference processing adequately with just a single NVIDIA A100 40GB GPU. According to estimates by industry media, implementation can begin with an initial investment of around 5 million yen, dramatically lowering the cost barrier for on-premise deployment in corporate environments. Moreover, it employs a proprietary tokenizer tailored to Japanese-specific vocabulary, frequent compound words, and honorific expressions, incorporating clever engineering that enables deep semantic comprehension with fewer tokens.

Building a Closed Environment That Keeps Confidential Data Internal

By combining the Private AI Platform on PRIMERGY, an on-premises generative AI infrastructure provided by Fsas Technologies, with tsuzumi 2, a purely domestic LLM, generative AI can be operated on a completely local closed network. Because processing is completed within an environment built exclusively for the customer, there is no risk of data leaking externally even when feeding internal technical manuals, contracts, or highly confidential blueprint data into the AI to perform document QA and summarization tasks using advanced RAG (Retrieval-Augmented Generation) technology. As shown in the system configuration diagram below, you can leverage cutting-edge generative AI in a network-isolated, closed environment while protecting your company’s valuable intellectual property.

Additionally, this solution offers a lineup of multiple server hardware models that can be flexibly scaled according to the customer’s usage scale and budget, ranging from a small start to enterprise-wide large-scale deployment.

Provision Structure and Future Outlook

A division of roles and support system that leverages the strengths of both companies

In providing this solution, NTT West will serve as the prime contractor, comprehensively handling overall proposals, contracts, technical support, and the provision of the tsuzumi 2 model as well as implementation support. Meanwhile, FSAS Technologies will be responsible for providing the hardware for the Private AI Platform using Fujitsu’s highly reliable “PRIMERGY” servers, as well as providing maintenance and operational support for building the generative AI infrastructure in an on-premises environment and ensuring its stable operation. Even for companies and local governments that do not have advanced AI engineers or an infrastructure operations system in-house, it is possible to receive one-stop services—from consideration of implementation to actual infrastructure construction and post-launch operation and maintenance—under a nationwide maintenance support system. This significantly reduces the administrative burden on management departments, such as system maintenance and library updates associated with the local operation of generative AI, allowing enterprises to focus on their primary objectives of data utilization and business transformation.

Deployment to Social Infrastructure and Medical Fields and Future Prospects

Both companies plan to expand the availability of this solution, for which operational verification has been completed, from initial target groups such as government agencies, research institutions, financial institutions, and the manufacturing industry into even broader domains. Specifically, they are eyeing deployment in social infrastructure and energy—where strict data protection and 24/7/365 stable operation are absolute requirements—as well as the medical and pharmaceutical fields, which handle confidential patient medical records and clinical trial data. They have already received numerous inquiries from local governments, medical institutions, and financial settings, and practical implementation is expected in areas requiring economic security and governance compliance. By generating a synergistic effect between NTT West’s community-rooted ICT solution delivery capabilities and Fujitsu FSAS’s cultivated stable server infrastructure operation technologies, the widespread adoption of highly reliable local generative AI environments across all industries in Japan is expected to further accelerate.

Reference Page

• [FSAS Technologies News Release] https://www.fsastech.com/ja-jp/resources/press-releases/2026/0826.html
• 【NTT West News Release】https://www.ntt-west.co.jp/news/2608/260826a.html
• [NTT R&D tsuzumi] https://www.rd.ntt/research/LLM_tsuzumi.html

【#GenerativeAI #DomesticLLM #tsuzumi #OnPremises #SovereignAI #NTTWest #FSASTechnologies #DataSovereignty】