NeoSpace: Tabular Foundation Models for Banks With NVIDIA

June 16, 2026 - 21:26

A new collaboration between NeoSpace, NVIDIA, and Oracle Cloud is reshaping how financial institutions in Latin America assess credit risk and generate lending offers. The partnership combines NeoSpace's tabular foundation models with NVIDIA's GB200 NVL72 hardware and Oracle Cloud infrastructure to deliver measurable gains in accuracy and loan origination for banks serving over 60 million customers.

According to early deployment results, the integrated system improves predictive accuracy by 30 to 50 percent compared to traditional credit models. It also increases the number of installment credit offers by roughly 10 percent, allowing banks to extend more loans to qualified borrowers while maintaining risk controls.

Tabular foundation models are designed to handle structured data, such as transaction histories, income records, and payment behaviors, which are common in banking. NeoSpace's models, optimized for NVIDIA's latest GPU architecture, process this data faster and with higher precision than conventional machine learning pipelines. The Oracle Cloud layer provides scalable storage and compute resources, enabling banks to run these models across massive customer datasets without building their own data centers.

For Latin American banks, the impact is significant. Many lenders in the region rely on limited credit bureau data, which can exclude large segments of the population. The NeoSpace-NVIDIA-Oracle system allows banks to analyze alternative data sources and generate credit scores for customers who lack traditional credit histories. This expands access to installment loans, credit cards, and other financial products.

The technology is already being tested by several major banks in Brazil, Mexico, and Colombia. Early adopters report faster loan approvals and lower default rates. NeoSpace plans to roll out the solution more broadly across the region in the coming months, with potential expansion into other emerging markets.