NVIDIA and Japanese Financial Institutions Showcase How Generative AI Is Redefining Financial Services

Generative AI is rapidly transforming the financial services industry, accelerating innovation across fraud detection, risk management, investment research, customer engagement, and many other areas. On April 17, NVIDIA joined “NVIDIA Financial AI Meet-up with Macnica,” hosted by Macnica, where Ase Blanco, NVIDIA’s global banking strategy lead, introduced how AI is transforming financial businesses while sharing the latest global trends. Major Japanese financial institutions and partner companies also presented their latest initiatives for meeting domestic needs such as balancing data sovereignty with accuracy. Blanco opened by saying that AI adoption is not merely an evolution but a true revolution. He emphasized that NVIDIA is not simply a chipmaker, but a platform supporting the AI revolution. NVIDIA designs and provides an integrated stack spanning chips, infrastructure, models, and applications, which is why AI companies need NVIDIA. He also highlighted NVIDIA’s strong commitment to open source, including not only models but also training data and trained weights, enabling users to build new businesses. In finance, he said, it is essential to build intelligence that institutions can control themselves. Financial services generate about $1.2 trillion in annual profit, making it the world’s largest profit-generating industry. AI-driven process improvements alone are estimated to create an additional $200 billion to $340 billion in profit. Fraud causes more than $500 billion in annual losses and is growing by more than 10% each year, meaning improved AI fraud detection could create value on the scale of hundreds of billions of dollars. NVIDIA’s research also shows that 99% of financial institutions plan to continue or expand AI investment, 89% already see revenue gains or cost reductions from AI, and 84% recognize the importance of open source. According to Blanco, the bank of the future will be powered not by datasets but by intelligence. In 2025, several major banks introduced AI factories, a turning point comparable to banks adopting mainframes in the 1960s. AI-centric banks should combine proprietary internal data with open-source models to create their own unique intelligence. Future banks will operate multiple specialized models, including models for credit underwriting, product and pricing knowledge, and transaction models that better understand customer behavior. AI models can improve existing processes and raise employee productivity. For example, RBC Capital Markets reduced analyst research work from 40 hours to 15 minutes and expanded coverage from 1,500 stocks to 2,500. Blanco also shared cases where GPU acceleration reduced heavy risk analysis and pricing calculations from overnight workloads to under one hour, creating time for new analysis, discoveries, and opportunities. Blanco concluded that finance is a data industry, and AI converts data into intelligence. AI factories are the foundation for scaling this opportunity. The challenge is no longer technical, because the technology already exists and works; the challenge is on the business side, where institutions must decide how to invest in and use AI. Every bank must redefine what a bank is, and the next decade will be a period of continuous innovation. Eight leading Japanese organizations then shared their latest initiatives. Kyoto University Graduate School of Management’s Shotaro Minami discussed LLMs that mathematically guarantee information quality and the need for a new OS in the AI era. KDDI presented its collaboration with MUFG, focusing on finance-specific AI in environments equivalent to on-premises systems for use cases requiring confidentiality, specialization, and model stability. Daiwa Institute of Research introduced a validation project using NVIDIA NIM in an on-premises GPU environment to perform speech transcription and text summarization with local LLMs. Nomura Research Institute described building efficient and accurate industry- and task-specific LLMs for financial regulation use cases, achieving accuracy above GPT-5.2 on three practical tasks. First Accounting presented work on enterprise-document-specialized SLMs using NVIDIA Nemotron, NeMo Data Designer, and DGX B200, improving information extraction accuracy from 89% to 100%. Mizuho Financial Group described the phased development of “Mizuho LLM,” an AI foundation for financial operations, and its exploration of NeMo Curator and synthetic customer persona data to expand training datasets. Rakuten Group discussed the design philosophy, training process, benchmark experiments, and NVIDIA NeMo Curator usage behind fine-tuning a finance-focused vertical LLM. Ricoh showed business-led generative AI use cases on-premises, including Dify-based use of Nemotron-Nano-9B-v2-Japanese, integration with kintone and internal storage, image understanding for financial documents, business feasibility assessment, and loan approval draft generation. The venue also featured demos of technologies that accelerate AI application development for financial operations. Macnica showcased an on-premises LLM for financial institutions that balances persona-based market research simulation with data sovereignty, using NVIDIA Nemotron-Nano-9B-v2-Japanese, Nemotron-Personas-Japan, and DGX Spark to simulate surveys reflecting Japan’s real geographic and demographic distribution. Ippu Senkin, a member of NVIDIA Inception, exhibited Local AI Agent, an on-premises generative AI service designed for finance-specific use cases that autonomously handles local workflows ranging from document creation to coding.