Rakuten Launches Enterprise AI Assistant on Custom LLM

Rakuten, the Japanese e-commerce and fintech giant often called the 'Amazon of Japan,' has officially launched an enterprise-grade AI assistant built on its proprietary large language model optimized for Japanese-language business operations. The move positions Rakuten as a direct competitor to Western enterprise AI providers like Microsoft Copilot and Google Duet AI, while carving out a distinct niche in Asian-language enterprise intelligence.

The new product, integrated across Rakuten's sprawling ecosystem of over 70 services, represents one of the most ambitious deployments of a homegrown LLM by a non-US tech company. It signals a growing global trend: major regional tech players are rejecting dependence on OpenAI and Google models in favor of building sovereign AI capabilities.

Key Facts at a Glance

• What: Rakuten's enterprise AI assistant powered by its in-house large language model
• Target market: Japanese and Asia-Pacific enterprise customers across e-commerce, fintech, and logistics
• Model foundation: Proprietary LLM built on the RakutenAI model family, with 7B and 70B parameter variants
• Integration: Embedded across Rakuten's ecosystem including Rakuten Ichiba, Rakuten Mobile, and Rakuten Bank
• Competitive positioning: Aims to outperform GPT-4 and Claude on Japanese-language business tasks by up to 30%
• Availability: Rolling out to enterprise partners in Q3 2025, with broader availability planned for Q4

Rakuten Builds Its Own AI Stack From the Ground Up

Rakuten's AI ambitions are not new, but this launch marks a significant escalation. The company has been developing its own large language models since 2023, when it first released the open-source RakutenAI-7B model. That initial release demonstrated strong performance on Japanese-language benchmarks, often outperforming multilingual models like Meta's Llama 2 on tasks involving complex Japanese grammar and business terminology.

The enterprise assistant builds on a much larger model in the RakutenAI family, reportedly utilizing a 70B-parameter architecture that has been fine-tuned on proprietary business data from across Rakuten's ecosystem. This gives the model an unusual advantage: access to real-world enterprise data spanning e-commerce transactions, financial services, telecommunications, and logistics.

Unlike OpenAI or Anthropic, which build general-purpose models and then adapt them for enterprise use, Rakuten has taken the opposite approach — building a model that is enterprise-first and Japanese-language-native from the ground up.

Why Japanese Language Models Matter for Global AI

The significance of Rakuten's move extends far beyond Japan. Japanese is one of the most challenging languages for AI systems to process accurately, due to its three writing systems (hiragana, katakana, and kanji), complex honorific structures, and context-dependent meaning. Most Western LLMs treat Japanese as a secondary language, resulting in performance gaps that can be critical in business contexts.

Rakuten claims its proprietary model achieves substantially higher accuracy on Japanese business document processing compared to GPT-4o and Claude 3.5 Sonnet. Key areas of improvement include:

• Contract analysis: 30% fewer errors in interpreting Japanese legal terminology
• Customer service automation: More natural and culturally appropriate responses
• Financial reporting: Better handling of Japanese accounting standards and regulatory language
• Supply chain communication: Improved accuracy in parsing logistics documentation

This performance gap matters because Japan remains the world's 4th-largest economy, with enterprise IT spending exceeding $150 billion annually. Companies operating in Japan have long struggled with AI tools that were designed primarily for English-language use cases.

Enterprise Features Target Business Automation

The AI assistant itself offers a suite of enterprise features designed to compete directly with Microsoft 365 Copilot and Google Workspace AI. Rakuten has built the assistant to function as a cross-platform business tool that integrates with existing enterprise workflows.

Core capabilities include document summarization, meeting transcription with action item extraction, automated report generation, and intelligent search across company knowledge bases. The assistant also features a conversational interface that allows employees to query internal databases, generate analyses, and draft communications in both Japanese and English.

One particularly noteworthy feature is real-time cross-language business communication, which allows Japanese-speaking teams to collaborate seamlessly with English-speaking counterparts. The system translates not just words but business context, adjusting tone, formality levels, and cultural references automatically.

Rakuten has also emphasized data sovereignty as a key selling point. Unlike cloud-based AI services from US providers, the enterprise assistant can be deployed on-premises or within Rakuten's own cloud infrastructure, keeping sensitive business data within Japanese jurisdiction. This addresses a major concern for Japanese enterprises, many of which face strict regulatory requirements around data localization.

The Global Race for Sovereign AI Heats Up

Rakuten's launch fits into a broader global pattern of sovereign AI development. Countries and major corporations worldwide are increasingly investing in homegrown AI capabilities rather than relying exclusively on US-built models.

France's Mistral AI has raised over $600 million to build European-focused LLMs. The UAE's Technology Innovation Institute developed the Falcon model series. China's Baidu, Alibaba, and ByteDance have each launched competing LLMs. And now Rakuten joins a growing cohort of Asian tech companies building proprietary models optimized for regional languages and business practices.

This trend is driven by several factors:

• Regulatory pressure: Data protection laws increasingly require local processing
• Language performance: Native models consistently outperform multilingual alternatives
• Strategic independence: Companies want to avoid vendor lock-in with US AI providers
• Competitive differentiation: Proprietary AI becomes a moat against commoditization
• National security: Governments encourage domestic AI development for strategic reasons

For Western enterprises, this fragmentation of the AI landscape means that a one-size-fits-all approach to enterprise AI is becoming less viable. Companies operating across multiple regions may need to adopt different AI systems optimized for local languages and regulatory environments.

What This Means for Businesses and Developers

For enterprise buyers, Rakuten's launch introduces a credible alternative to Western AI assistants for Japanese-language business operations. Organizations with significant Japanese operations should evaluate whether a purpose-built Japanese LLM delivers better ROI than adapting a general-purpose model like GPT-4 or Claude.

For developers, the move highlights the growing market opportunity in building language-specific and region-specific AI applications. Rakuten has indicated it will offer API access to its enterprise model, potentially enabling third-party developers to build specialized applications on top of the RakutenAI platform.

The pricing strategy remains competitive. While Microsoft 365 Copilot costs $30 per user per month, Rakuten is reportedly positioning its enterprise assistant at a comparable price point but with significantly better performance on Japanese-language tasks. For large Japanese enterprises with thousands of employees, the cost differential could be substantial when factoring in reduced error rates and improved productivity.

Looking Ahead: Rakuten's Broader AI Ambitions

Rakuten has signaled that the enterprise assistant is just the beginning. The company's CEO, Hiroshi Mikitani, has spoken publicly about building a comprehensive AI ecosystem that spans consumer and enterprise applications.

Near-term plans reportedly include expanding the model's capabilities to cover Korean and Southeast Asian languages, targeting Rakuten's growing presence across the Asia-Pacific region. The company is also investing in multimodal capabilities, with plans to add vision and audio processing to the enterprise assistant by early 2026.

Longer-term, Rakuten appears to be positioning itself as the AI infrastructure provider of choice for Asian enterprises — a role analogous to what Microsoft and Google play in Western markets. With its existing ecosystem of over 1.6 billion member accounts globally and deep integration across e-commerce, fintech, mobile, and media, Rakuten has the distribution advantage to make this vision plausible.

The success or failure of Rakuten's enterprise AI strategy will be closely watched by the global tech industry. If a regional tech giant can build a competitive AI assistant that outperforms Western alternatives in its home market, it could accelerate the trend toward AI localization worldwide — and challenge the assumption that Silicon Valley will dominate the enterprise AI market indefinitely.