Details
- Razorpay announced Vulcan, which it calls India’s first transformer-based AI foundation model built specifically for payments, on August 18, 2026.
- The model is designed to improve payment success, fraud detection, routing, checkout personalization, and risk decisions across Razorpay’s merchant network.
- Razorpay says Vulcan was built with NVIDIA and AWS, using NVIDIA’s accelerated computing and AWS infrastructure, including Amazon SageMaker, for training and deployment.
- The company says the system was trained on about 3 trillion data points across 4 billion payments, replacing separate ML models for routing, fraud, and checkout with one shared intelligence layer.
- Razorpay claims early live use has delivered 8-10% higher payment success rates, 8x more international card fraud stopped, and 5x more fraudulent or disputed transactions identified without more alerts.
Impact
The launch reinforces a broader shift toward domain-specific foundation models, especially in financial infrastructure where multiple signals must be acted on in real time. If Razorpay’s results hold at scale, the model could reduce payment friction and fraud while pushing India’s payments stack toward more adaptive, AI-native routing and risk systems over the next 12–24 months.
