The payments-focused AI model is designed to improve payment success, transaction routing, fraud detection, and risk assessment.
Indian fintech company Razorpay has launched Vulcan, a transformer-based AI foundation model designed specifically for digital payments. The model is aimed at improving payment success rates, transaction routing, fraud detection, risk assessment, and checkout personalisation.
Developed using technology from Nvidia and Amazon Web Services, Vulcan was trained on nearly three trillion data points from four billion payments. Razorpay said the model analyses around 3,000 signals for each transaction to identify patterns across merchants, payment methods, issuers and payment gateways.
How Vulcan works
Unlike generative AI models built primarily to understand and produce language, Vulcan is designed to analyse payment behaviour and predict which transaction route is most likely to succeed.
The model can assess transactions in real time and recommend payment routes, while also identifying potential fraud patterns. Razorpay said it can be used to assess cash-on-delivery risks and recommend suitable payment methods during checkout.
The company has already deployed parts of Vulcan across live transactions, with businesses including Blinkit, Bachatt and redBus using some of its capabilities.
According to Razorpay, early deployments have increased payment success rates by 8% to 10% and helped detect eight times more international card fraud. The company also reported identifying five times more fraudulent or disputed transactions without increasing the number of alerts.
On its Magic Checkout platform, Razorpay said Vulcan helped 40% more shoppers see their preferred UPI app, contributing to an additional 100,000 to 200,000 purchases each month.
The performance figures are based on Razorpay’s own deployment results.
Built as a common payments intelligence layer
Razorpay intends Vulcan to serve as a common intelligence layer across multiple payment functions rather than using separate machine-learning models for individual tasks.
The company plans to expand its applications into areas such as authentication and lending.
Razorpay describes Vulcan as India’s first transformer-based AI foundation model built specifically for payments. The launch also places the company among a growing group of payment companies using large-scale transaction data to improve payment infrastructure.
US payments company Stripe launched its own Payments Foundation Model in May 2025, trained on tens of billions of transactions and hundreds of payment-related signals.
Part of Razorpay’s wider AI and IPO push
Vulcan is the latest addition to Razorpay’s broader AI strategy. In March 2026, the company launched Agent Studio, an AI platform designed to help businesses manage payments, recover revenue and handle financial operations.
The launch also comes as Razorpay prepares to enter India’s public markets. The company confidentially filed draft papers for a $600 million IPO in June.
Founded in 2014 by Harshil Mathur and Shashank Kumar, Razorpay provides payment infrastructure for businesses across cards, net banking, UPI and digital wallets. The company has since expanded into financial services, including lending.
With Vulcan, Razorpay is positioning AI not simply as an additional product feature, but as infrastructure operating behind the payment itself, deciding how transactions are routed, assessed and protected in real time.
