Razorpay has launched Vulcan, a payments-focused artificial intelligence foundation model designed to make digital transactions more reliable, secure and personalised. Built specifically for the payments industry rather than general-purpose text generation, the transformer-based model has been trained on nearly 3 trillion data points covering about 4 billion payments.
The launch marks a major expansion of Razorpay’s AI strategy as the fintech company looks to use machine intelligence across payment routing, fraud detection, risk assessment and checkout personalisation. Developed with NVIDIA and AWS, Vulcan is designed to operate as a unified intelligence layer across multiple payment functions instead of relying on separate machine-learning systems for each task.
Razorpay Launches Vulcan AI Foundation Model
Vulcan is a proprietary AI foundation model developed by Razorpay specifically for payments.
Unlike general-purpose large language models such as ChatGPT, Vulcan is not designed primarily to understand or generate human language. Instead, it is trained to recognise patterns in transaction data and understand the factors that influence whether a digital payment succeeds, fails or becomes potentially fraudulent.
Razorpay said the model has been trained on nearly 3 trillion data points generated across approximately 4 billion payments.
| Metric | Details |
|---|---|
| AI model | Vulcan |
| Developer | Razorpay |
| Model type | Payments-focused AI foundation model |
| Training data | Nearly 3 trillion data points |
| Payments covered | About 4 billion |
| Technology partners | NVIDIA and AWS |
| Key applications | Routing, fraud, risk and checkout |
| Signals analysed per transaction | About 3,000 |
The scale of the underlying transaction data is intended to allow Vulcan to identify patterns that individual machine-learning models may not be able to capture independently.
How Vulcan Is Different From an LLM
Vulcan is not another chatbot or general-purpose AI model.
Traditional language models are trained to process text, understand instructions and generate responses. Payments foundation models instead work with structured transaction information and behavioural signals.
Razorpay’s model is designed to understand the movement of money.
It can examine information associated with a transaction and determine which payment route is most likely to succeed, whether a transaction could be fraudulent and which payment method may be most suitable for an individual customer.
This makes Vulcan closer to an intelligence engine for financial infrastructure than a consumer-facing conversational AI system.
Vulcan Analyses Thousands of Transaction Signals
Razorpay said Vulcan can evaluate roughly 3,000 signals for each transaction.
These signals can help the model assess factors affecting payment outcomes and identify the route most likely to result in a successful transaction.
Instead of simply processing a payment through a predetermined path, the system can evaluate available routes in real time.
How AI-Powered Payment Routing Works
Customer initiates payment
↓
Vulcan analyses transaction signals
↓
Approximately 3,000 signals evaluated
↓
Available payment routes assessed
↓
Most suitable route identified
↓
Payment processed
↓
Transaction outcome becomes additional learning input
The objective is to reduce payment failures while making the overall checkout experience more reliable.
Payment Success Rates Improve by 8–10%
Razorpay said early components of Vulcan have produced an 8–10% improvement in payment success rates.
The company has already deployed parts of the technology in live payment environments.
The improvement is important because failed payments can result in abandoned purchases, additional support requests and lost revenue for merchants.
Even a relatively small improvement in payment success can therefore have a significant commercial impact when applied across billions of transactions.
Fraud Detection Has Also Improved
Fraud prevention is another major use case for Vulcan.
Razorpay said the model has detected eight times more international card fraud compared with previous systems.
It also identified five times more fraudulent or disputed transactions without increasing the number of alerts generated for customers or businesses.
This is important because simply increasing fraud alerts can create another problem: legitimate transactions may be incorrectly flagged.
The objective is therefore not just to identify more suspicious activity but to improve detection accuracy.
Vulcan Can Detect Fraud Across Merchants
One potential advantage of a payments-focused foundation model is its ability to identify patterns across a broader payments network.
A fraudulent behaviour pattern that appears at one merchant could potentially provide useful information when evaluating a transaction elsewhere.
This network-level intelligence can be difficult to replicate with isolated models trained for individual merchants or payment products.
Razorpay said Vulcan can detect fraud across its network and identify suspicious behaviour that becomes visible across multiple merchants.
The Model Can Flag Risky Cash-on-Delivery Orders
Vulcan’s applications also extend beyond digital payment authorization.
The model can identify potentially risky Cash on Delivery orders.
This can help merchants assess whether an order has a higher probability of becoming a return, failed delivery or other loss.
By applying AI to payment and commerce behaviour together, Razorpay is attempting to expand the model’s role beyond the moment when a customer clicks the payment button.
Checkout Personalisation Is Another Application
Vulcan is also being used to improve checkout personalisation.
Razorpay said 40% more shoppers on its Magic Checkout platform are now seeing their preferred UPI application.
That improvement has contributed to an additional 100,000 to 200,000 purchases every month, according to the company.
Showing the right payment option at the right time can reduce friction and potentially increase the likelihood that customers complete purchases.
Razorpay Wants One AI Layer Across Payments
Historically, payment companies have used separate machine-learning models for different functions.
One system might handle fraud detection, another could optimise payment routing and another could support risk assessment.
Razorpay’s approach with Vulcan is to bring these functions together under a common intelligence layer.
From Fragmented AI to Unified Intelligence
Separate fraud model
+
Separate routing model
+
Separate risk model
+
Separate checkout model
↓
Multiple systems
↓
Razorpay Vulcan
↓
Unified payments intelligence
↓
Routing
+
Fraud
+
Risk
+
Checkout
↓
Potentially faster learning and decision-making
The company believes this architecture can allow insights from one part of the payments ecosystem to improve other functions.
Live Customers Are Already Using the Technology
Razorpay said components of Vulcan are already being used in live payment environments by customers including Blinkit, Bachatt and redBus.
This indicates that the technology has moved beyond an experimental research project.
The real-world deployment also provides Razorpay with transaction feedback that can help improve the model.
The company expects the system to become increasingly useful as more payment patterns are incorporated into its intelligence layer.
NVIDIA and AWS Provide the Technology Infrastructure
Razorpay developed Vulcan in partnership with NVIDIA and AWS.
NVIDIA’s accelerated computing technology provides the GPU infrastructure required to train and operate the model.
AWS provides cloud infrastructure and services including Amazon SageMaker to support development, training and deployment.
The collaboration reflects the increasing importance of specialised computing infrastructure in financial AI.
Technology Stack
Razorpay
↓
Payments data + proprietary model
↓
NVIDIA
↓
GPU computing
↓
AWS
↓
Cloud infrastructure + Amazon SageMaker
↓
Vulcan
↓
Real-time payments intelligence
The combination gives Razorpay access to the computing resources required to train and operate a model at significant transaction scale.
Why Payments Need Specialised AI
Payments generate enormous amounts of structured information.
Every transaction can contain signals related to:
- Payment method
- Merchant
- Customer behaviour
- Transaction amount
- Time
- Location
- Device
- Network
- Authentication
- Previous transaction history
- Payment route
Analysing these signals in real time can help payment companies identify patterns that are difficult to capture through simple rules.
AI can potentially make these decisions more dynamic and adaptive.
AI Could Reduce Payment Friction
Payment failures remain a major problem for digital commerce.
A customer may have sufficient funds but still experience a failed transaction because of bank downtime, technical problems, authentication issues or payment-routing failures.
For merchants, every failed payment represents a potential lost sale.
AI-powered routing can potentially reduce these failures by selecting the payment route with the highest probability of success.
India’s Digital Payments Market Creates a Large Opportunity
India has one of the world’s largest digital payments ecosystems.
UPI has transformed everyday payments, while cards, wallets, net banking and other digital methods remain important for online commerce.
The scale of India’s transaction ecosystem provides a large environment in which payment-focused AI systems can be trained and deployed.
Razorpay believes the opportunity will expand further as digital commerce grows.
E-Commerce Could Reach $350 Billion by 2030
Razorpay has pointed to India’s digital e-commerce market as a major opportunity for payments intelligence.
The company expects India’s digital e-commerce market to reach approximately $350 billion by 2030.
A larger digital commerce ecosystem would mean more payments, more transactions and more data that can potentially be used to improve payment systems.
For Razorpay, this creates an opportunity to position Vulcan as infrastructure for the next phase of India’s digital economy.
Foundation Models Could Change Financial Infrastructure
Foundation models are increasingly moving beyond general-purpose AI applications.
Technology companies are developing specialised models for areas such as healthcare, finance, coding and scientific research.
Payments are another area where large-scale specialised models could become important.
The advantage is that a domain-specific model can be trained around the unique patterns and requirements of a particular industry.
The Model Could Eventually Support Lending
Razorpay said it eventually wants Vulcan to support payment decisions ranging from authentication and routing to fraud detection and lending.
Lending would significantly expand the model’s potential role.
Payment behaviour can provide valuable signals about business activity and transaction patterns.
However, using such information for lending decisions would require careful risk controls, regulatory compliance and appropriate safeguards.
AI Could Make Payments More Personalised
Payments are becoming increasingly personalised.
Customers may have preferred payment methods, banks, UPI applications or cards.
A system that understands individual preferences could potentially present the most relevant option automatically.
This could reduce the number of steps required at checkout.
For merchants, improved personalisation could translate into higher conversion rates.
Security Will Become Increasingly Important
The expansion of AI in financial services also increases the importance of security.
A payments foundation model must operate in an environment where mistakes can directly result in financial losses.
The model therefore needs to balance accuracy, speed and security.
Razorpay said the infrastructure developed with AWS is designed for enterprise-grade security, while the model’s fraud-detection capabilities are intended to strengthen protection across payment flows.
Data Quality Will Be Critical
Foundation models are only as effective as the data used to train them.
Payments data can be complex, noisy and highly sensitive.
Razorpay’s access to large volumes of transaction information provides an advantage in training a specialised model, but it also creates significant responsibility around data governance and security.
The company will need to ensure that the model continues to operate within applicable regulatory and privacy requirements.
Real-Time Decisions Create Technical Challenges
Payments happen in seconds.
Any AI system used during checkout therefore needs to produce decisions extremely quickly.
A model that provides highly accurate predictions but adds significant latency could damage the customer experience.
Razorpay’s use of accelerated computing and cloud infrastructure is intended to support real-time decision-making at large scale.
The Technology Could Benefit Smaller Merchants
One potential advantage of a unified payments intelligence system is that its benefits could extend across merchants of different sizes.
Large companies can afford to build sophisticated fraud and payment-routing systems internally.
Smaller businesses generally cannot make the same investment.
A payment infrastructure provider can potentially distribute advanced AI capabilities across its merchant network.
This could help smaller businesses benefit from technology that would otherwise be difficult to build themselves.
Razorpay’s AI Strategy Is Expanding
Vulcan is not Razorpay’s first AI initiative.
The company has been integrating AI into its payments infrastructure for several years and has launched tools aimed at helping businesses manage payments, payroll and other financial operations.
It has also been working on agentic payments, where AI systems can assist with transactions.
Vulcan represents a deeper move toward making AI part of the underlying payment infrastructure itself.
Agentic Payments Could Be the Next Step
Razorpay has already worked with NPCI and AI companies on agentic payments.
These systems are designed to allow AI assistants to help users discover products and complete transactions through conversational interfaces.
A payments foundation model could provide an important intelligence layer beneath such systems.
The combination of agentic AI and specialised payment intelligence could eventually allow AI agents to make more sophisticated transaction decisions while operating within user-authorised payment frameworks.
Competition in Fintech AI Is Increasing
Razorpay is not alone in exploring AI for financial services.
Banks, payment companies and fintech startups are investing heavily in machine learning for fraud prevention, risk assessment and customer personalisation.
The emergence of specialised foundation models could increase competition around the underlying intelligence powering financial infrastructure.
Razorpay’s advantage could come from its large transaction network and integration with merchant payment flows.
Scale Could Become a Competitive Advantage
The more transactions a payments model can observe, the more patterns it may be able to identify.
This creates a potential network effect.
More transactions
↓
More behavioural data
↓
More patterns identified
↓
Better predictions
↓
Higher payment success
↓
More merchants attracted to the platform
↓
More transactions
The ability to operate at large scale could therefore become an important competitive advantage in payments AI.
What Razorpay’s Launch Means for Merchants
For merchants, the most immediate potential benefits are improved payment success, stronger fraud detection and smoother checkout experiences.
Higher payment success can increase completed orders.
Better fraud detection can reduce losses.
More personalised checkout experiences can improve conversion.
If these benefits are sustained at scale, AI could become an increasingly important component of merchant payment infrastructure.
What It Means for Consumers
Consumers may not directly interact with Vulcan as they would with a chatbot.
Instead, they may experience its effects indirectly.
A payment may succeed more often.
The preferred UPI application may appear automatically.
A suspicious transaction may be blocked more effectively.
Checkout may require fewer decisions.
These small improvements could collectively make digital payments more reliable.
Key Risks
Despite its potential, payments AI also carries risks.
These include:
- False fraud detections
- Incorrect transaction routing
- Data-security risks
- Model bias
- System failures
- Cyberattacks
- Regulatory concerns
- Excessive dependence on automated decisions
- Rising computing costs
Razorpay will need to maintain strong monitoring and human oversight as the system expands.
What Investors and the Industry Will Watch
Several indicators will determine whether Vulcan becomes a major competitive advantage for Razorpay.
These include:
- Payment success-rate improvements
- Fraud-detection accuracy
- Merchant adoption
- Transaction volumes processed through the system
- Checkout conversion
- Infrastructure costs
- Model latency
- Expansion into lending and risk
- Regulatory compliance
- Enterprise adoption
The ability to demonstrate measurable improvements without increasing costs or risks will be critical.
Key Facts at a Glance
| Metric | Details |
|---|---|
| Model | Vulcan |
| Developer | Razorpay |
| Model category | Payments AI foundation model |
| Training data | Nearly 3 trillion data points |
| Payments covered | About 4 billion |
| Signals per transaction | Around 3,000 |
| Payment success improvement | 8–10% |
| International card fraud detection | 8x increase |
| Fraud/dispute identification | 5x increase |
| Magic Checkout preferred UPI visibility | 40% increase |
| Additional monthly purchases | 1–2 lakh |
| Technology partners | NVIDIA and AWS |
| Live customers cited | Blinkit, Bachatt and redBus |
| Potential future applications | Authentication, routing, fraud, risk and lending |
Infographic: How Razorpay Vulcan Works
RAZORPAY
↓
VULCAN
PAYMENTS AI FOUNDATION MODEL
↓
TRAINED ON
NEARLY 3 TRILLION DATA POINTS
↓
ACROSS
~4 BILLION PAYMENTS
↓
ANALYSES
~3,000 SIGNALS PER TRANSACTION
↓
REAL-TIME INTELLIGENCE
↓
PAYMENT ROUTING
+
FRAUD DETECTION
+
RISK ASSESSMENT
+
CHECKOUT PERSONALISATION
↓
EARLY RESULTS
8–10% HIGHER PAYMENT SUCCESS
+
8X MORE INTERNATIONAL CARD FRAUD DETECTED
+
5X MORE FRAUD/DISPUTES IDENTIFIED
↓
FUTURE
AUTHENTICATION
+
LENDING
+
AGENTIC PAYMENTS
The Bigger Picture
Razorpay’s launch of Vulcan marks a significant shift in how AI can be applied to India’s digital payments infrastructure. Rather than developing another general-purpose chatbot, the fintech company has built a foundation model specifically around transaction data, training it on nearly 3 trillion data points across about 4 billion payments. The model is designed to bring payment routing, fraud detection, risk assessment and checkout personalisation under a unified intelligence layer.
The early results indicate why specialised AI could become important in financial infrastructure. Razorpay says components of Vulcan have improved payment success rates by 8–10%, detected eight times more international card fraud and identified five times more fraudulent or disputed transactions without increasing alert volumes. With digital commerce expanding rapidly, the ability to make payments more reliable while reducing fraud could become a major competitive advantage for payment platforms.
Looking Ahead
Razorpay’s next challenge will be scaling Vulcan across more payment flows while maintaining speed, accuracy and security. The company plans to expand the model’s role beyond routing and fraud into authentication, risk and potentially lending. As more merchants use the system, its ability to learn from transaction patterns across the network could further strengthen its predictive capabilities.
Over the longer term, specialised payments AI could become an important layer of India’s digital financial infrastructure. The combination of foundation models, real-time payments and agentic commerce could allow transactions to become increasingly automated and personalised. However, the technology will also require strong safeguards around data, fraud, model errors and regulatory compliance. If Razorpay can demonstrate that its model consistently improves payment outcomes without adding new risks, Vulcan could become a significant building block in the next generation of digital payments.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
