Razorpay is taking a deeper bet on AI with the launch of Vulcan, its own foundation model built specifically for digital payments. Unlike a typical language model, Vulcan is designed to understand how money moves through the payments ecosystem, helping Razorpay decide the best payment routes in real time and spot suspicious transactions more effectively.
Built using NVIDIA and AWS technology, the model has been trained on roughly 3 trillion data points across 4 billion payments. Razorpay says Vulcan studies around 3,000 signals for every transaction and continues learning as more payments move through the system. The idea is to use that broader view of payment behaviour rather than solving individual problems separately.
The early results are already showing up in live transactions, according to the company. Razorpay says parts of Vulcan have improved payment success rates by 8–10%, helped detect eight times more international card fraud, and identified five times more fraudulent or disputed transactions without creating additional alerts. It has also helped more shoppers see their preferred UPI app at checkout.
For Razorpay, Vulcan fits into a much wider push to bring AI deeper into its payments stack. The fintech company has already introduced tools such as AI-powered merchant experiences and Agent Studio, and Vulcan now gives it a model trained specifically on financial transaction patterns rather than general-purpose data.
The launch also comes as Razorpay moves closer to a public listing in India. The company confidentially filed its draft IPO papers in June, with its proposed issue reportedly expected to raise around $600–700 million. As Razorpay prepares for that next phase, Vulcan shows where it sees another part of its growth story coming from: making the massive flow of digital payments smarter every time a transaction happens.
Filed by
Startup Unplugged
