Razorpay launches Vulcan, an AI foundation model for payments, trained on 3 trillion data points
Razorpay has introduced Vulcan, an AI foundation model trained on nearly 3 trillion data points from 4 billion transactions. The model aims to improve payment success rates, fraud detection, and risk management in India’s digital payments ecosystem, where complexity and fraud risks are rising alongside rapid growth.
Editor, Lazyfounder

30 SEC SUMMARY
- Razorpay has launched Vulcan, an AI foundation model for payments, trained on 3 trillion data points from 4 billion transactions.
- Vulcan aims to improve payment success rates, fraud detection, and risk management in India’s complex digital payments ecosystem.
- The model was developed with NVIDIA and AWS, using 3,000 signals per transaction to predict optimal payment routing.
- Early results show an 8-10% improvement in payment success rates and significantly better fraud detection without increasing alerts.
TABLE OF CONTENTS
KEY HIGHLIGHTS
- Vulcan is described as India’s first AI foundation model built specifically for payments.
- Trained on nearly 3 trillion data points from 4 billion payments, covering UPI, cards, net banking, wallets, and cash on delivery.
- Developed in collaboration with NVIDIA (accelerated computing) and AWS (cloud infrastructure).
- Early results indicate an 8-10% improvement in payment success rates and 8x more international card fraud detection.
- The model processes 3,000 signals per transaction to predict the most successful payment route.
- UPI accounted for 81% of India’s retail digital payments in FY2024-25, with over 24,000 crore transactions in FY2025-26.
What Vulcan aims to solve
Razorpay has introduced Vulcan, an AI foundation model designed to address challenges in India’s digital payments ecosystem. According to YourStory, the model targets three core areas: improving payment success rates, enhancing fraud detection, and optimising risk management. India’s payments landscape is characterised by multiple payment methods—UPI, cards, net banking, wallets, and cash on delivery—all interacting with hundreds of banks and gateways. This complexity often leads to failed transactions and fraud risks.
How Vulcan works
Vulcan is trained on nearly 3 trillion data points from 4 billion payments, making it one of the largest datasets applied to payment-specific AI. Unlike traditional machine learning models that focus on a single task, Vulcan is a foundation model designed to identify broader patterns across multiple use cases. Razorpay reports that it processes around 3,000 signals associated with every transaction to predict the most successful payment route.
The model was developed in collaboration with NVIDIA, which provided accelerated computing for training, and AWS, which supplied cloud infrastructure for development and deployment. Razorpay emphasises that Vulcan is not a large language model; instead, it is trained to recognise patterns in financial transactions rather than text.
Early performance metrics
Razorpay has shared early results from Vulcan’s deployment. According to the company, the model has led to an 8 to 10% improvement in payment success rates. It has also detected eight times more international card fraud and identified five times more fraudulent or disputed transactions without increasing the volume of alerts. These metrics suggest that Vulcan can enhance reliability and security without adding friction for users.
Why this matters for India’s payments ecosystem
India’s digital payments ecosystem is growing rapidly, with UPI accounting for 81% of all retail digital payments in FY2024-25. Government data cited by YourStory shows that UPI processed over 24,000 crore transactions in FY2025-26, reflecting both high adoption and increasing transaction values. This growth has also introduced complexities, such as fragmented payment methods and rising fraud risks.
Razorpay co-founder Harshil Mathur highlighted that while digital payments are expanding, reliability remains a concern for many users. Vulcan’s focus on improving success rates and fraud detection directly addresses this issue, potentially making digital transactions more accessible and trustworthy for a broader segment of the population.
Broader industry trends
The launch of Vulcan aligns with a global shift toward using AI in payments. According to research by Mastercard, organisations globally lost an average of $60 million annually to payment fraud, and AI is increasingly being used to reduce false positives and improve customer retention. However, the same research notes that fraudsters are also leveraging AI to create more sophisticated scams, raising the stakes for payment providers.
Visa has also ramped up its AI initiatives, introducing AI-driven fraud tools and frameworks for what it calls "agentic commerce," where AI systems can eventually handle tasks like product searches and purchases. Visa has described AI as a defining force in the future of payments, alongside real-time transactions and digital identity.
What this means
Lazyfounder analysis — our interpretation, not reported fact.
Vulcan’s launch signals a shift in how AI can be applied to payments—not just as a tool for incremental improvements but as a foundational system capable of addressing multiple challenges simultaneously. For founders and operators in fintech, this underscores the importance of investing in large-scale datasets and specialised infrastructure to build AI models that can adapt to complex, real-world environments. It also highlights the growing competition in AI-driven payments, where access to data and computing power will likely be key differentiators.
Key takeaways
- Founders in fintech and payments can use Vulcan’s approach as a benchmark for leveraging AI to improve transaction reliability and fraud detection.
- The model’s focus on a single AI system for multiple use cases may reduce complexity for operators managing fragmented payment ecosystems.
- Access to large datasets and specialised infrastructure (like NVIDIA and AWS) is becoming critical for building effective AI systems in payments.
Sources
This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.
About the author
Editor, Lazyfounder
Tarun Mottlia edits LazyFounders, covering Indian startups, funding rounds, AI and product launches. Every story on the site is AI-assisted and checked against its cited sources before publication.
More stories by Tarun MottliaGet the LazyFounder Brief
Startup, funding and AI news in a five-minute read. Join the early-access list.


