Japan's leading enterprises and research institutions are building AI models tailored specifically to Japanese language, industries, and workforce needs, using open-source tools that give organizations control over their own AI infrastructure. Companies including SoftBank, Hitachi, NTT DATA, and ENEOS Holdings are customizing AI models to handle everything from remote-presence robotics to autonomous telecom networks, marking a significant shift toward locally developed AI ecosystems rather than relying solely on global models.

Why Are Japanese Companies Building Their Own AI Models?

Japan faces a unique demographic challenge: an aging population and workforce transition that demands AI solutions tailored to local industries and languages. Global AI models, while powerful, often struggle with Japanese language nuances and industry-specific terminology. By developing their own models, Japanese organizations can ensure AI systems understand local context, comply with data sovereignty requirements, and strengthen the country's technological independence.

"Every nation and every company should own and control its intelligence infrastructure. Open models make that possible. They give countries, enterprises and researchers the freedom to inspect, improve, adapt, secure and deploy AI for their own needs," said Jensen Huang, founder and CEO of NVIDIA.

Jensen Huang, Founder and CEO at NVIDIA

This approach reflects a broader global trend: countries and enterprises recognizing that relying entirely on external AI providers creates dependencies. By using open-source models and datasets, organizations can customize AI to their specific needs while maintaining control over sensitive data and business logic.

What Specific AI Projects Are Japanese Companies Launching?

The scope of Japanese AI development spans multiple industries and use cases. Here are the key initiatives underway:

• Institute of Science Tokyo: Developed the Swallow family of open foundation models that enhance Japanese language and reasoning performance while preserving English, math, and coding capabilities. Enterprises are already customizing Swallow for specialized applications like financial-document translation and asset-management report generation.
• SoftBank's Sarashina Models: SB Intuitions Corp., SoftBank's generative AI research subsidiary, trained the Sarashina series of models. Sarashina3 mini was selected by Japan's Digital Agency for specialized AI use cases, and SoftBank has deployed a Large Telecom Model to enable autonomous telecom network operations.
• Stockmark's Document Understanding: Released a specialized Japanese-language document-understanding model based on open-source technology, serving customers across manufacturing, energy, and chemical industries through Japan's Generative AI Accelerator Challenge national project.
• Avatarin's Enterprise Agents: An AI and robotics startup is developing Japanese-language speech and reasoning capabilities for enterprise AI agents, with digital avatar systems being deployed at airports and other locations across Japan.
• ENEOS Holdings' Materials Research: Using AI to accelerate materials exploration for applications such as immersion-cooling fluids and advanced catalysts, integrating technical document search, vision, and language understanding with simulation-backed molecular screening.
• NTT DATA's Multi-Agent Framework: Developing scalable systems that route tasks to the best-suited models, driving accurate and autonomous enterprise workflows across complex operations.
• Hitachi's Physical AI: Combining AI with proprietary IT and operational technology domain knowledge to connect and coordinate enterprise operations, helping transform business processes across complex workflows.

How Are These Models Being Built and Deployed?

Japanese developers are leveraging open-source tools and datasets to build and customize their models. The NVIDIA NeMo software stack, which includes libraries for continual pretraining and post-training, enables organizations to refine models for their specific use cases. This approach gives developers transparency into how models work and the ability to improve them over time.

Deployment flexibility is another key advantage. Organizations can run these models in environments that meet regulatory, sovereignty, and data localization requirements, whether on-premises, in private clouds, or through cloud service providers. This is particularly important for enterprises handling sensitive customer data or operating in regulated industries.

Sakana AI, a Japanese AI company, is taking model coordination a step further by integrating open-source models into its Fugu platform, which intelligently orchestrates multiple models to dynamically select the best one for each task. Early performance results on complex, real-world coding tasks demonstrate that thoughtful coordination can unlock capabilities beyond what a single model achieves alone.

What Does This Mean for the Broader AI Landscape?

Japan's approach signals an important shift in how countries and enterprises think about AI development. Rather than waiting for global AI companies to build solutions, organizations are taking control of their own AI infrastructure. This democratization of AI development means that specialized knowledge, local language expertise, and industry-specific requirements can be directly embedded into models from the start.

The availability of these models on platforms like Hugging Face, ModelScope, and OpenRouter, as well as through NVIDIA NIM microservices, makes it easier for other organizations to adopt and build upon them. This creates a virtuous cycle where improvements made by one organization can benefit others, accelerating innovation across the entire ecosystem.

For Japan specifically, this initiative addresses critical workforce challenges. As the population ages, AI systems that understand Japanese language, culture, and industry practices can help maintain productivity and support workers in roles ranging from customer service to research and development. The investments being made today will shape how AI serves Japanese society for years to come.