What Mastercard Users Can Expect of the Firm's New AI Engine
AI / Machine Learning | 5 min read
Mastercard has unveiled a new generative AI foundation model — powered by NVIDIA and Databricks technology — designed to serve as an insights engine across payments and commerce. The model, revealed at NVIDIA GTC 2026 in San Jose, is one of the first payments-specific AI systems built to understand the nuances of global commerce at scale, trained on hundreds of millions of transactions and already showing promising signs of outperforming advanced machine learning techniques across multiple operations.
A Large Tabular Model: Different by Design
Unlike popular large language models (LLMs) trained on unstructured data such as text, video, and images, Mastercard's new foundation model belongs to a fundamentally different class of architecture — the Large Tabular Model (LTM). It is trained on structured data: large-scale tables and datasets, specifically the rich transaction data that flows through Mastercard's global network every second.
"Our new foundation model is a different kind of deep learning neural network, called a large tabular model, or LTM, which is trained on structured data, such as large-scale tables or datasets. We are training the latest version of our LTM on billions of anonymised transactions. Our plan is to ramp up this work to include hundreds of billions of payments transactions, as well as additional types of datasets, including merchant location data, fraud data, authorisation data, chargeback data and loyalty programme data. As we train the model on more data and more kinds of data, it will be able to provide more insights and predict future transactions with greater accuracy."
— Steve Flinter, Distinguished Engineer, Mastercard
The model is being developed using NVIDIA NeMo AutoModel and NVIDIA accelerated computing together with the Databricks platform — and Mastercard is showcasing results at NVIDIA GTC 2026 as one of the first financial services firms to demonstrate how payments-specific AI can optimise global commerce and fight cybercrime at this scale.
From Fraud Detection to Loyalty: What the LTM Enables
Mastercard's immediate priority for the LTM is strengthening its cybersecurity infrastructure. Traditional fraud detection requires data scientists to manually engineer features from raw transaction data to identify anomalies — a time-consuming process that relies on human expertise. The LTM learns key characteristics and patterns with minimal human intervention, analysing data independently to uncover connections that human analysts might overlook. Early testing shows improved performance in reducing false positives — particularly in recognising legitimate but infrequent high-value purchases that older systems tend to flag as suspicious.
Beyond cybersecurity, the LTM is designed to serve as a unified platform across a wide range of applications:
- • Loyalty and rewards programmes — more accurately identifying relevant offers and incentives for individual cardholders based on their full transaction history
- • Personalisation models — enabling more accurate predictions of customer preferences, behaviour, and next-best actions
- • Portfolio optimisation — improving how Mastercard and its partners manage card programmes and product offerings across markets
- • Data analytics and business insights — delivering richer, more actionable intelligence across operations, markets, and customer segments
- • Model consolidation — potentially reducing the thousands of specialised AI models Mastercard currently maintains for different markets, use cases, and customers into a single, more flexible foundation
Part of a Broader AI Roadmap
"This reflects a broader shift we've been driving across Mastercard's AI roadmap — moving beyond point solutions to foundation-level capabilities that learn from the complexity of global commerce. In just the last month, we've advanced multiple efforts that show how AI can deliver smarter security, more relevant experiences, and stronger performance across the payments ecosystem."
— Greg Ulrich, Chief AI and Data Officer, Mastercard
The LTM builds on Mastercard's growing AI track record. In 2024, the company's Decision Intelligence Pro system was scanning one trillion data points to improve fraud detection rates by an average of 20% — and by as much as 300% in some cases — while reducing false positives by more than 85%. The new foundation model is designed to go further: a backend infrastructure tool rather than a narrow application, capable of serving as the unified analytical engine across Mastercard's entire product portfolio. Mastercard's Value-Added Services segment generated US$13.3 billion in net revenue in 2025 — up from US$10.8 billion in 2024 — signalling the commercial weight behind this AI-led strategy.
The payments sector is watching closely. Revolut built a comparable transaction foundation model using masked prediction — achieving a 20% increase in fraud detection precision, better credit risk predictions, and a 9.6% uplift in cross-sell accuracy using NVIDIA's full AI stack. Mastercard's LTM signals that foundation-model thinking is becoming the defining competitive architecture for global payments at scale.
Key Takeaways
- • Mastercard has unveiled a new generative AI foundation model — a Large Tabular Model (LTM) trained on structured payments data — developed with NVIDIA NeMo AutoModel and Databricks, unveiled at NVIDIA GTC 2026.
- • Unlike LLMs trained on text and images, the LTM is trained on billions of anonymised transactions — with plans to scale to hundreds of billions across fraud, merchant, authorisation, chargeback, and loyalty datasets.
- • The LTM detects fraud patterns with minimal manual feature engineering, reduces false positives, and is already outperforming advanced machine learning techniques in early testing.
- • Applications span cybersecurity, loyalty programmes, personalisation, portfolio optimisation, data analytics, and the consolidation of thousands of market-specific AI models into a single foundation.
- • Mastercard's Value-Added Services grew to US$13.3 billion in net revenue in 2025 — this LTM signals a strategic shift from point AI solutions to foundation-level capabilities that learn from the full complexity of global commerce.