Razorpay has launched Vulcan, an artificial intelligence (AI) foundation model designed specifically for payments specifics of the Indian market. The company describes Vulcan as India’s first transformer-based AI foundation model for payments, beyond using separate AI tools for individual tasks such as fraud detection or payment routing.
A foundation model is a large AI system trained on substantial amounts of data that can then support several different applications. Instead of building a separate model for every payment problem, Vulcan is designed by Razorpay to act as a common intelligence layer across different parts of the payment process.
According to Razorpay, Vulcan has been trained on around 3 trillion data points from more than 4 billion payments, allowing it to assess thousands of signals associated with a transaction. The company says the model is being used to improve payment success rates, identify fraudulent or disputed transactions and personalise checkout experiences.
Razorpay reports an 8–10% improvement in payment success rates from applications of the model, while international card fraud detection has increased eightfold. The company also says Vulcan has identified five times more disputed transactions and helped more customers see their preferred payment applications during checkout. These figures are Razorpay’s reported results rather than independently audited performance metrics.
The technology is being developed with infrastructure support from NVIDIA and Amazon Web Services. Its focus is significant because payment transactions generate highly structured data, including information about payment methods, merchants, banks, transaction timing and risk signals. That makes payments a potentially strong environment for specialised AI models that can learn from patterns across very large transaction volumes.
There’s currently a broader shift in financial services toward foundation models. In 2025, Stripe announced what it called the world’s first AI foundation model for payments, trained on tens of billions of transactions and intended to support fraud prevention, payment performance and other parts of its payments suite. The development suggests that major payment companies increasingly see proprietary transaction data as an important source of AI capabilities.
For India, the development fits into a wider effort to build AI systems around local data and use cases. The Indian government’s IndiaAI Mission is supporting foundation models designed around Indian languages, sectors and local requirements, intending to reduce dependence on technology developed for other markets.
Vulcan is therefore notable not simply because another fintech has launched an AI model, but because it shows that advanced AI development is increasingly happening outside Silicon Valley. Razorpay calls Vulcan the first transformer-based AI foundation model for payments in India, although India’s broader payments ecosystem is already pursuing other foundation-model initiatives, including work by NPCI. Besides, the country aims to eliminate possible bias that can be present in existing AI models predominantly developed by U.S. companies by incorporating diverse local languages and cultures into a future AI tool. Per rough estimations, the country is home to about 6,700 AI startups, so there is a wide selection of indigenous teams who might participate in local artificial intelligence development.
The trend is not limited to India. China’s DeepSeek-R1 became a prominent alternative to leading U.S. AI models after its 2025 release, demonstrating that competitive AI systems can emerge from outside the traditional Silicon Valley ecosystem. Its most recent DeepSeek-V4-Pro version has a Codex integration and enhanced agentic capabilities as well.
