Overview
- Razorpay announced the launch Tuesday of Vulcan, a transformer-based payments foundation model built with NVIDIA GPUs and hosted on AWS infrastructure in India.
- The company says Vulcan was trained on roughly 4 billion payments and about 3 trillion data points using around 3,000 signals per transaction to score routes, flag fraud and personalise checkouts.
- Razorpay published early deployment figures from pilots with customers such as Blinkit, redBus and Bachatt that it says show an 8–10% lift in payment success, an 8x rise in stopped international card fraud and a 5x increase in identified disputed transactions.
- Independent outlets and analysts have raised unanswered questions about which raw data were used, whether any identifiable consumer data were included, the legal basis under India’s DPDP Act for using merchant transaction data, and whether merchants can opt out.
- Razorpay says Vulcan is hosted in India, was funded internally, and will be extended into paid products such as advanced fraud prevention and real-time underwriting, a move that could change merchant costs, consumer checkout experiences and the company’s valuation ahead of an IPO.