NPCI and NVIDIA Partner on Sovereign AI for Indian Payments
The National Payments Corporation of India (NPCI) is partnering with NVIDIA to develop a sovereign AI infrastructure for the country's payments ecosystem. The initiative will use NVIDIA's technology to build national-scale models for fraud detection, risk scoring, and anomaly detection across the UPI rails, enhancing security for one of the world's largest real-time payment systems.
- The initiative is a direct response to escalating fraud on the UPI network, which saw a 85% increase in FY 2023-24, amounting to ₹1,087 crore ($130M USD) in losses across 1.34 million reported incidents. - The collaboration will leverage NVIDIA's Nemotron, a family of open AI models, to build the foundational layer. NVIDIA's fraud detection blueprints often combine graph neural networks (GNNs) with other machine learning models to more accurately identify complex fraud patterns and reduce false positives. - This move marks a strategic shift from deploying AI for specific use cases to building a single, scalable foundation. NPCI previously piloted a federated AI model with banks for risk analysis and developed a smaller language model named FiMI for improving customer grievance resolution. - The "sovereign AI" aspect is critical, ensuring India's payments data remains within its borders and the models are developed independently of foreign technologies, aligning with national data sovereignty policies. - This infrastructure is not just for internal NPCI use; the plan is to make the platform accessible to the broader ecosystem of banks, fintechs, and other payment participants to foster innovation. - The underlying hardware for such large-scale AI training and real-time inference typically relies on purpose-built systems like NVIDIA's DGX AI supercomputers, designed to handle massive datasets and complex AI models. - NPCI's CEO, Dilip Asbe, has stated that embedding AI is the next critical step for the UPI platform's evolution, as traditional rule-based systems are insufficient to secure a network that processes over 18 billion transactions per month.
