Razorpay, the Mumbai‑based fintech unicorn, announced the launch of Vulcan, a transformer‑style AI foundation model designed to make digital payments faster, safer, and more reliable. The company says the system scores payment routes in real time and flags fraud that can be seen across merchants once the model is live.
In a market where the volume of online transactions is expanding by double digits every year, any improvement in payment success or fraud detection can translate into significant revenue gains for merchants and better customer experiences. Razorpay’s move comes as the company gears up for its public listing and as global rivals such as Stripe explore similar AI‑driven payment engines.
What You Need To Know
1. The launch of Vulcan and its technical backbone
Vulcan is built on a transformer architecture, a type of neural network that excels at processing sequences of data. Razorpay partnered with NVIDIA for accelerated computing and Amazon Web Services for scalable cloud infrastructure. This combination allows the model to ingest and analyze vast amounts of payment data at high speed.
Unlike conventional language models, Vulcan is engineered to understand the flow of money rather than text. It processes every transaction as a sequence of signals—such as card details, device fingerprint, and merchant information—to predict the most reliable route for settlement and flag anomalies.
2. Proprietary design and data ownership
The model is a ground‑up creation; Razorpay owns both the architecture and the training data. The company emphasizes that Vulcan is not a generic large‑language model but a specialized system trained exclusively on its own payment ecosystem.
By keeping the data in-house, Razorpay can fine‑tune the model to the nuances of the Indian market, where payment methods like UPI, debit cards, and international cards coexist. This ownership also positions the firm to protect sensitive financial data while still benefiting from AI insights.
3. Training scale and learning methodology
Razorpay claims Vulcan has been trained on roughly 3 trillion data points derived from 4 billion payments. Each transaction contributes about 3,000 distinct signals, ranging from merchant category codes to geolocation and device metrics.
Because the model learns from every transaction it processes, it continually refines its predictions. This continuous learning loop means that each successful payment teaches the system something new, gradually improving overall performance without the need for manual rule updates.
4. Early performance results
In pilot deployments, Razorpay reports several measurable gains. Payment success rates have risen by 8‑10% on average, while the detection of international card fraud has increased eightfold. The system also identifies five times as many disputed or potentially fraudulent transactions without inflating the number of alerts sent to merchants.
Additionally, 40% more shoppers are presented with their preferred UPI app during checkout, a change that is expected to add 100,000 to 200,000 extra purchases each month. These figures suggest that Vulcan not only improves security but also enhances the customer journey.
5. Strategic positioning within Razorpay’s AI ecosystem and IPO plans
Vulcan is part of a broader AI strategy that includes ChatGPT‑native storefronts and an Agent Studio platform. About 200 businesses already use these tools, indicating a growing ecosystem around Razorpay’s AI offerings.
The announcement arrives as Razorpay finalizes its confidential draft IPO filings. The company aims for a public offer between $600 million and $700 million, potentially valuing it at $5 billion to $6 billion—below its last private valuation of $7.5 billion. A robust AI platform like Vulcan could be a key differentiator for the firm in attracting investors and customers alike.
The Wider Picture
Razorpay is not the first fintech to explore AI in payments. Stripe, the U.S. payments giant, has already launched a payments foundation model trained on tens of billions of transactions. However, Razorpay’s claim to be the first Indian player highlights the growing appetite for AI‑driven solutions in emerging markets.
India’s digital payments landscape is expanding rapidly, with UPI transactions surpassing 100 billion in 2025 and the overall market expected to grow at a CAGR of 12% over the next five years. In such a competitive environment, even modest gains in payment success or fraud detection can have outsized financial impacts.
Beyond the numbers, the move underscores a broader shift in fintech: companies are increasingly treating AI as an internal engine rather than a marketing gimmick. By embedding AI directly into the payment flow, Razorpay aims to create a self‑optimizing ecosystem that continually improves as more transactions occur.
In Short
Vulcan is a transformer‑based AI model built with NVIDIA and AWS to improve payment success and fraud detection.
The model processes 3 trillion data points from 4 billion transactions, learning from 3,000 signals per payment.
Early pilots show 8‑10% higher success rates, 8× better international fraud detection, and 5× more disputed transaction identification.
Razorpay’s AI strategy also includes ChatGPT‑native storefronts and an Agent Studio platform, already used by 200 businesses.
Vulcan’s launch precedes Razorpay’s anticipated IPO, which could value the company at $5‑$6 billion.
This article is based on reporting published by Inc42.
