Razorpay is set to launch ‘Vulcan’ on August 18, 2026, describing it as India’s first transformer-based AI foundation model built specifically for payments. The model has been developed to handle payment decisions across routing, fraud detection, risk assessment and checkout personalisation.
According to Razorpay, Vulcan has been trained on approximately 3 trillion data points covering 4 billion payments. The model analyses around 3,000 signals for every transaction to identify patterns that can influence payment success, fraud risk and customer payment preferences.
Razorpay Vulcan Targets Payment Success And Fraud Detection
Early components of Vulcan are already operating across Razorpay’s payments network, with Blinkit, Bachatt and redBus among the companies using the technology in live payment environments. Razorpay said the system has improved payment success rates by 8 to 10% during early deployment.
The company also reported improvements in fraud detection. Vulcan detected and stopped 8 times more international card fraud and identified 5 times more fraudulent or disputed transactions, without increasing the number of alerts generated for payment teams.
Vulcan is also being used to improve the checkout experience. Razorpay said the model helped 40% more shoppers see their preferred UPI application through Magic Checkout, contributing to an additional 1 to 2 lakh purchases every month.
How Vulcan Uses AI Across The Payments Network
Vulcan is designed as a single foundation model rather than separate machine learning models for individual payment functions, allowing it to support payment routing, fraud detection, risk assessment and checkout personalisation.
It can assess transactions in real time, identify the payment route most likely to succeed, detect fraud patterns across merchants, flag potentially risky Cash on Delivery orders and recommend the payment method most likely to work for each customer.
Razorpay said Vulcan has been developed as a proprietary model, with its architecture and training data owned by the company. NVIDIA GPUs were used for training and deployment, while AWS infrastructure, including Amazon SageMaker, supported model development, training and deployment.
Razorpay CEO and cofounder Harshil Mathur said the model was developed around the need to make digital payments more reliable for users across India. The company plans to extend its use across authentication, payment routing, fraud detection and lending as India’s digital commerce market moves towards a projected $350 billion by 2030.
