Mastercard is rolling out a new generative AI foundation model built specifically for payments and commerce. The company appears to be taking a different route, building a model designed around structured transaction data instead of relying on a general-purpose chatbot approach.

The payment card services shared details in a March 17 post written by one of its engineers, timing the announcement alongside NVIDIA’s GTC 2026 event. NVIDIA, for its part, said Mastercard is using NVIDIA NeMo AutoModel, along with accelerated computing and Databricks, to power what it calls a proprietary transaction-focused foundation model.

What Mastercard says has changed with its AI transaction data model

According to Mastercard, this system is a “large tabular model,” meaning it learns from structured datasets rather than the text, images, or video typically used in large language models. The company says it is currently training the model on billions of anonymized transactions and plans to expand into other datasets, including merchant location, fraud activity, authorizations, chargebacks, and loyalty behavior.

That approach lines up with Mastercard’s scale. In its latest 10-K filing, the company says it handles more than 70% of Mastercard and Maestro transactions and nearly all cross-border activity. As a result, the volume makes a data-heavy model feasible. Still, some of the company’s broader claims around prediction accuracy and downstream impact come mostly from its own early testing and haven’t been independently verified.

Where the evidence is stronger

Fraud detection and cybersecurity stand out as the most concrete use cases so far. Mastercard has already talked about earlier generative AI efforts in this area, including a 2024 update that said it doubled the speed of identifying potentially compromised cards. This new model builds on that, with the company saying internal tests show better performance than traditional machine-learning systems and fewer false positives on unusual but legitimate purchases.

Those results sound promising, but they remain internal. The article doesn’t include third-party benchmarks or detailed metrics, which makes it harder to gauge how much of a leap this really represents.

What it could mean for smaller businesses

For small and midsize businesses, the impact is likely to be indirect for now. Merchants won’t be interacting with this model themselves. Instead, any benefits would show up through improved fraud scoring, fewer declined legitimate transactions, smoother review processes, or more refined data products tied to payments and loyalty programs.

Mastercard has been pushing more AI-branded tools for small businesses this month, which makes that connection believable. Even so, real-world gains will depend on how these systems perform outside controlled testing.

Featured image: Pexels