Overview
- Razorpay publicly launched Vulcan on Tuesday, August 18, 2026, calling it a transformer-based foundation model built with NVIDIA GPUs and AWS infrastructure to optimize payments.
- The company says Vulcan was trained on about 3 trillion data points from roughly 4 billion payments and ingests around 3,000 signals per transaction to learn payment behaviour at scale.
- Razorpay reports early, company-measured gains from live deployments with customers such as Blinkit and RedBus: an 8–10% rise in payment success, 8x more international card frauds stopped, 5x more disputed or fraudulent transactions identified, and 40% more shoppers shown their preferred UPI app.
- Razorpay says it removes personally identifiable information before training, hosts Vulcan on private India-based infrastructure to meet data-localization rules, and intends to fund Vulcan internally while adding paid fraud, risk and credit products.
- The launch arrives as Indian regulators tighten fraud rules and banks adopt tools like MuleHunter, creating demand for real-time AI payment controls, though the model’s performance claims are currently company-reported and not independently verified.