Razorpay launched Vulcan, India’s first AI payments foundation model built with NVIDIA and AWS — trained on 3 trillion data points to cut failed transactions and fraud.
Razorpay’s New AI Model Just Boosted Payment Success Rates by Up to 10%
For millions of first-time or small-town Indian shoppers, a single failed digital payment — a card declined for no clear reason, an OTP that arrives too late, a subscription that silently lapses — is often enough to send them back to cash. Razorpay, India’s omnichannel payments platform for businesses, says it built its newly launched AI model specifically to close that gap.
On August 18, 2026, Razorpay announced Vulcan, described as India’s first transformer-based AI foundation model built specifically for payments, developed using NVIDIA’s accelerated computing and AWS’s cloud infrastructure, including Amazon SageMaker. The company frames it as foundational groundwork for India’s e-commerce market, which is projected to reach $350 billion by 2030.
Trained on 3 Trillion Data Points From 4 Billion Real Payments
The scale behind Vulcan is substantial: the model has been trained on approximately 3 trillion data points drawn from 4 billion payments processed across Razorpay’s network, learning from roughly 3,000 signals per transaction. Razorpay describes it as a proprietary, ground-up model — meaning both the underlying architecture and the training data belong to the company, rather than being built on top of an existing general-purpose AI system.
That distinction matters, according to Razorpay: this isn’t a large language model repurposed for finance. “LLMs understand text; this model understands the language of the movement of money,” the company said in its announcement.
Why One Failed ₹2,400 Payment Explains India’s Whole Payments Problem
Razorpay illustrates the underlying issue with a simple example: a shopper named Meera tries to buy running shoes for ₹2,400 at 9 p.m., taps “Pay,” and sees “Payment unsuccessful. Please try again.” Her card is fine, her bank account is fine, her money is fine — the payment simply had several possible routes through India’s fragmented payments infrastructure, and the wrong one was briefly selected at that moment.
That fragmentation is a uniquely Indian challenge: a single purchase can be processed through UPI, cards, net banking, wallets, or cash on delivery, routed across hundreds of banks and payment gateways. An internal Razorpay study spanning 1.5 million shoppers and more than 51,000 businesses found the same friction — failed transactions, drop-offs, delays — showing up identically whether the shopper was in a metro city or a small town. That consistency is what convinced Razorpay to build one shared model rather than keep refining separate systems for routing, fraud, and checkout individually, which the company compares to “several doctors examining a patient, each reading only their own test results.”
8x More International Card Fraud Caught — Without More False Alarms
Early components of the model have already been running live, with customers including Blinkit, Bachatt, and redBus seeing measurable results ahead of today’s full launch:
- 8-10% improvement in payment success rates
- 8x more international card fraud detected and stopped
- 5x more fraudulent or disputed transactions identified, without increasing the number of alerts sent to businesses
- 40% more shoppers now see their preferred UPI app on Razorpay’s Magic Checkout, which Razorpay says is helping complete 1-2 lakh additional purchases every month
What the Model Actually Does
Razorpay outlines four core capabilities built into the foundation model: hyper-precision routing, which sends each payment down the path most likely to succeed in real time; network-level fraud detection, which can flag a stolen card the moment it’s used across unrelated sellers — a pattern only visible when looking across merchants rather than within just one; RTO (return-to-origin) risk intelligence, which flags risky cash-on-delivery orders before checkout; and predictive checkout personalization, which recommends the payment method most likely to succeed for each individual customer.
“India’s appetite for digital payments is real, but it isn’t universal yet — for a large part of the country, going digital still comes down to one thing: does it work, every single time?” said Harshil Mathur, CEO and co-founder of Razorpay. “That’s the customer we built this for: the one still deciding whether to trust a screen over cash in hand. An AI-led payments foundation model doesn’t just solve today’s problem and stop there. Every payment teaches the system something that makes the next payment better.”
Built for a $350 Billion E-Commerce Future by 2030
The infrastructure behind Vulcan came from a three-way partnership. “NVIDIA’s work with Razorpay in partnership with AWS on AI payments foundation models has opened up a new frontier, turning complex payments data into real-time contextual intelligence,” said Pahal Patangia, Head of Global Industry Business Development and Payments at NVIDIA, whose GPUs powered the model’s training and live operation at scale.
“Razorpay is reimagining payments intelligence at India scale with an AI Foundation Model — built on Amazon SageMaker — that consolidates billions of transaction insights into a single, continuously learning intelligence layer, replacing fragmented ML models with unified AI,” said Kiran Jagannath, Head of FSI and Conglomerates at AWS India and South Asia.
Razorpay says this launch is a starting point rather than a finished product, with the eventual goal of routing every payment-related decision — authentication, routing, fraud, and even lending — through one continuously learning model as India’s digital economy scales toward that projected $350 billion e-commerce market by 2030.