As an early adopter of cuTile, machine learning engineers at Jane Street implemented Mamba-2, a state-of-the-art sequence model, in cuTile and partnered with NVIDIA to optimize the kernels. This blog post walks you through the three custom kernels powering the Mamba-2 forward pass and the cuTile optimizations behind them.
AI Research for Financial Services
NVIDIA researchers and collaborators study financial foundation models, market structure, portfolio methods, and agentic AI for quantitative research. Explore papers, code, datasets, and talks that connect research questions to methods and evaluation.
How does NVIDIA approach AI research in financial services?
Research at NVIDIA examines financial data, market assumptions, and uncertainty in portfolio methods. Open models offer adaptable starting points; agentic examples show how quant teams can propose, code, backtest, and refine candidate signals against their own data and evaluation criteria.
Related NVIDIA AI security research examines how to evaluate open-model risks, secure agents with access to data and tools, and validate protections through transparent, reproducible methods.
Research contributions
Selected papers and technical blogs from NVIDIA researchers and collaborators. Explore the methods, evidence, and implementation details behind each contribution.
Open source codes and developer examples
Inspect research implementations and explore GPU accelerated workflows for financial problems.
Events and workshops
Explore talks, speaking sessions, and workshops led by NVIDIA experts in financial services.
Financial services research contributions
Selected papers and technical blogs from NVIDIA researchers and collaborators. Explore the methods, evidence, and implementation details behind each contribution.
Examine how hidden order-queue assumptions can change passive-execution outcomes in a trading backtest, even when the observed aggregate order book is identical.
Work through dependence matrices, factorization, and changing instrument groupings in a reproducible financial clustering workflow.
Explore GPU methods for identifying shared risks across financial instruments by learning structure in correlation and extreme-event dependence.
Use NVIDIA NeMo Data Designer, NeMo Curator, and Nemotron models to build large-scale, diverse synthetic datasets through an iterative generation-deduplication loop, with the released FinHeadlineMix dataset.
Using synthetic financial headlines generated with NVIDIA Nemotron 3 Nano, researchers study how to distill an open language model for financial-event classification. The paper maps how training data, model size, and reasoning supervision affect task accuracy and the retention of broader knowledge.
PRAGMA learns reusable representations from banking event histories for fraud detection, credit scoring, and personalization. The research tests an open NVIDIA Nemotron model to enrich transaction text. Those representations could provide predictive context for future agentic banking workflows.
Explore an end-to-end developer example that connects financial data preparation, scenario generation, and portfolio optimization on GPUs with CUDA-X Data Science open-source libraries like NVIDIA cuOpt and cuML.
Explore a framework that combines learned financial representations, forecasts, and uncertainty estimates to identify useful factors for portfolio construction.
Explore neural representations of financial market structures that combine multiscale decomposition with calibration to observed financial data.
Open-Source Codes and Developer Examples
Inspect research implementations and explore GPU workflows for financial problems.
Financial services on build.nvidia.com
Developer examples designed for quick-start AI development in financial services, including artifacts like Docker containers and Jupyter Notebooks, allowing for fast deployment with tools like Docker compose and Brev Launchable.
model-distillation-finetuning-recipes
Reference PyTorch recipes for LoRA supervised fine-tuning and logit-based knowledge distillation with FSDP2. The code is intentionally compact: Hugging Face models, PEFT adapters, torch.distributed, and explicit training loops without a trainer framework.
Security in the Age of AI
Explore NVIDIA research on strengthening cyber defense with AI and securing AI systems, including open-model evaluation and safeguards for agentic workflows.
Events and workshops
Agenthon 2026: Verifiable AI for Quantitative Finance
Test financial agents across coding, forecasting, market simulation, and evidence-grounded prediction. Explore the competition and its concluding workshop at NeurIPS Atlanta.
The 2nd Workshop on Advances in Financial AI: Towards Agentic and Responsible Systems
Cheap Inference, High Fidelity: Scaling Laws for LLM Distillation in Finance
Generative AI in Finance
Invited Talk 2: Ioana Boier (NVIDIA)
Financial AI research FAQs
What financial services topics does this page cover?
This page covers financial foundation models, synthetic data and model distillation, market structure and trading execution, and portfolio construction and risk. It also features agentic AI for quantitative research and links to related AI security work.
What does NVIDIA contribute to financial services research?
NVIDIA contributes research methods, open models and tools, and GPU-accelerated workflows. With academic and industry collaborators, its researchers publish papers and technical blogs, share code and datasets, and support workshops that help others test and build on the work.
Where can I find published NVIDIA Research papers?
Find published papers in the NVIDIA Research publications directory. Individual research entries should also link directly to their paper, technical blog, code, data, or other public primary source.
Where can I find information about NVIDIA technologies for financial services?
Explore AI factories for financial services, including open AI models and libraries, secure infrastructure, blogs, and success stories.