Designs GPUs and AI systems and sells them, with a licensed software stack for training, customising and serving models, to cloud providers, AI labs and enterprises; banks run the stack in their own data centres or rent it in the cloud.
What is NVIDIA, and which banks use it?
NVIDIA, founded in 1993 and listed on Nasdaq, makes the GPUs, DGX systems and NVIDIA AI Enterprise software (NeMo, NIM, Triton) that banks use to train and serve models in their own data centres or in the cloud. Bank customers on our pages include BNY (a DGX SuperPOD since March 2024), American Express (fraud models on DGX and Triton), Capital One, PNC and, at group level, Santander and Deutsche Bank. NVIDIA AI Enterprise lists at $4,500 per GPU per year; hardware prices are not published.
Category
Founded
1993
Headquarters
Santa Clara, California, US
Ownership
Public (NASDAQ: NVDA); founded in 1993 by Jensen Huang, who has been president and chief executive since its inception.[14][20]
Deployment
On-premises GPU servers and DGX systems, cloud GPU instances, and NVIDIA AI Enterprise software licensed per GPU or by the hour on cloud marketplaces
Pricing
GPUs and DGX systems through NVIDIA and its partners on quotes; NVIDIA AI Enterprise software per GPU per year, as a perpetual licence, or per GPU-hour on cloud marketplaces
Banks among the 120
6
- BNY said in March 2024 it was the first major bank to deploy an NVIDIA DGX SuperPOD with DGX H100 systems, used for deposit forecasting, payment automation, predictive trade analytics and end-of-day cash balances, with dozens of applications in development on NVIDIA AI Enterprise software such as NeMo and Triton.
- American Express trains LSTM fraud models on NVIDIA DGX, optimises them with TensorRT and serves them on Triton Inference Server within a two-millisecond latency budget; NVIDIA's customer story reports 50 times the throughput of the CPU-based setup it replaced (vendor-reported).
- PNC is buying Nvidia chips for its own data centres as part of what CEO Bill Demchak calls an 'AI factory' (June 2026), to rely less on usage-priced external models; by October 2026 GPU chips were installed in PNC data centres, adding depreciation.
- Santander said on 19 October 2026 that its AI Lab specialises models with NVIDIA NeMo Customizer, uses NVIDIA Kumo models for structured data and serves models with NVIDIA Dynamo-Triton, cutting serving costs in selected experiments from about €5 to about €0.03 per interaction; the two set a three-year collaboration framework.
- Outside the banks tracked here, NVIDIA names RBC Capital Markets (agentic AI built with NVIDIA AI Enterprise for more than 8,000 users), Lloyds Banking Group, NatWest and Revolut among its financial-services users, and HSBC through Starburst's GPU engine; RBC's US bank, City National, is not named.
- Scale: fiscal 2026 revenue was $215.9 billion; second-quarter fiscal 2027 revenue was $96.2 billion, of which data centre $89.0 billion, and NVIDIA's third-quarter outlook assumes no data-centre compute revenue from China.
- NVIDIA AI Enterprise list prices are published: $4,500 per GPU for a one-year self-managed subscription or $1 per GPU per hour on cloud marketplaces; prices for GPUs and DGX systems are not published.
What does NVIDIA sell to banks?
| Product | What it does | Status |
|---|---|---|
| NVIDIA AI Enterprise[16][19] | Licensed software suite for developing and deploying AI on self-managed systems or cloud GPU instances; generally available since August 2021. | Generally available |
| DGX systems and DGX SuperPOD[1][2] | Integrated AI servers and the reference architecture for clusters of them with InfiniBand networking; the systems behind BNY's 2024 deployment and American Express's fraud-model training. | Generally available |
| NVIDIA NeMo[1][7][11] | Platform for building, customising and guardrailing generative models, including NeMo Customizer (used by Santander), NeMo Retriever and the NeMo Agent Toolkit (used by RBC Capital Markets). | Generally available |
| NVIDIA NIM[17] | Inference microservices that package models as optimised containers for clouds, data centres or workstations; downloadable since June 2024. | Generally available |
| Triton Inference Server and TensorRT[2][4][7] | Model-serving and inference-optimisation software; American Express serves its fraud models on Triton, Capital One uses Triton and TensorRT-LLM, and Santander serves models with NVIDIA Dynamo-Triton. | Generally available |
| Data-centre GPUs (Hopper, Blackwell, Vera Rubin)[1][14] | The accelerators inside DGX and partner servers; BNY's SuperPOD uses DGX H100 systems, and NVIDIA said in August 2026 that the Vera Rubin platform was ramping into full production. | Generally available |
How much does NVIDIA cost?
How it is charged: GPUs and DGX systems through NVIDIA and its partners on quotes; NVIDIA AI Enterprise software per GPU per year, as a perpetual licence, or per GPU-hour on cloud marketplaces.
| Product | Plan | List price | Unit | Checked |
|---|---|---|---|---|
| NVIDIA AI Enterprise | Self-managed subscription, 1 year (includes Business Standard support)[19] | $4,500 | per GPU | Oct 26, 2026 |
| NVIDIA AI Enterprise | Perpetual licence with 5 years of support[19] | $22,500 | per GPU | Oct 26, 2026 |
| NVIDIA AI Enterprise | Cloud marketplace, production, pay as you go[19] | $1 | per GPU per hour, plus the cloud provider's instance cost | Oct 26, 2026 |
NVIDIA AI Enterprise list prices are published in NVIDIA's licensing guide, which says a five-year subscription is included with NVIDIA H100, H200 NVL and A800 PCIe GPUs. Prices for GPUs, DGX systems and DGX SuperPOD clusters are not published, and none of the bank deployments on this page disclosed a value.
Which banks use NVIDIA?
| Bank | What | Status |
|---|---|---|
| BNY (Bank of New York Mellon) | On-premises NVIDIA DGX SuperPOD with DGX H100 systems for deposit forecasting, payment automation, predictive trade analytics and end-of-day cash balances; AI applications built with NVIDIA AI Enterprise software (NeMo, Triton, Base Command)[1] | In production |
| American Express | Real-time fraud detection: LSTM models trained on NVIDIA DGX, optimised with TensorRT and served on Triton Inference Server alongside a gradient boosting model[2] | In production |
| Capital One | Agent servicing tool for thousands of customer-service agents built with the NVIDIA enterprise AI ecosystem; NVIDIA's inference stack (Triton, TensorRT-LLM) in its multi-agent workflows; NVIDIA RAPIDS for GPU-accelerated Spark[3][4] | In production |
| PNC Financial Services | Nvidia chips bought for PNC's own GPU compute in its data centres (the 'AI factory'), to run in-house large and small language models[5][6] | Rolling out |
| Santander US | Group level: Santander AI Lab specialises models with NeMo Customizer, uses NVIDIA Kumo models and serves models with Dynamo-Triton inside Santander's own architecture in experimental use cases, under a three-year collaboration framework (October 2026); the release does not break out the US bank[7] | Pilot |
| Deutsche Bank (U.S.) | Group level: multi-year innovation partnership (December 2022) under which Deutsche Bank planned to use NVIDIA AI Enterprise for risk-model development and high-performance computing, on-premises and on Google Cloud; US scope not stated[8] | Announced |
Among the 120 largest US banks on this site, from sources that name the bank: the bank, NVIDIA or the press. Status is the furthest stage a source confirms.
What is NVIDIA AI Enterprise?
NVIDIA's licensed software suite for building and deploying AI, including NeMo for building and customising generative models and the Triton Inference Server for serving them; it is licensed per GPU and runs on self-managed systems or on GPU instances bought through the major cloud marketplaces.[19][16][1]
What has NVIDIA done, and when?
- Oct 22, 2026
European financial-services users at GTC Berlin[13]
NVIDIA names Revolut, Lloyds Banking Group, NatWest, Santander and HSBC (through Starburst's GPU engine) among firms running models on its accelerated computing.
- Oct 19, 2026
Santander three-year collaboration framework[7]
Santander reports specialised models built with NeMo Customizer and NVIDIA Kumo and served with Dynamo-Triton, and agrees a framework covering foundation models, agents, fraud and AML, real-time decisioning and a banking digital twin.
- Sep 28, 2026
JPMorgan Chase joins Nvidia-led agent-safety effort[10]
JPMorgan Chase joins an Nvidia effort to develop shared, open-source safety tools for AI agents; no purchase commitment, investment or deployment timetable was announced.
- Aug 26, 2026
Second-quarter fiscal 2027: $96.2 billion revenue[14]
Data-centre revenue reaches $89.0 billion; Vera Rubin is ramping into full production, and the outlook assumes no data-centre compute revenue from China.
- Jul 9, 2026
JPMorgan Chase picks SambaNova for on-premises inference[12]
JPMorgan Chase picks AI chip startup SambaNova as an infrastructure partner, deploying its SN40 and SN50 systems (RDU architecture) for on-premises AI inference.
- Jun 11, 2026
PNC is buying Nvidia chips for an 'AI factory'[5]
American Banker reports PNC is acquiring data centres and buying Nvidia chips to run its own large and small language models, to avoid vendor lock-in and usage-based pricing.
- Apr 27, 2026
Federated-learning fraud proof of concept[9]
Kinexys by J.P. Morgan, BNY, RBC, DeepTempo and NVIDIA train fraud models without pooling data, using NVIDIA FLARE on DGX Cloud; a joint test, not a purchase.
- Feb 25, 2026
Fiscal 2026 revenue of $215.9 billion[15]
Revenue for the year to January 25, 2026 rises 65%; fourth-quarter data-centre revenue is $62.3 billion.
- Oct 28, 2025
Palantir collaboration[18]
NVIDIA and Palantir announce an integrated stack that brings NVIDIA accelerated computing, CUDA-X libraries and Nemotron models into Palantir's AI Platform.
- Mar 17, 2025
Capital One presents NVIDIA-based builds at GTC[3]
Capital One engineers describe an agent servicing tool built with the NVIDIA enterprise AI ecosystem, a GPU high-performance computing cluster on Kubernetes and GPU-accelerated Spark with NVIDIA RAPIDS.
- Mar 18, 2024
BNY deploys a DGX SuperPOD[1]
BNY becomes the first major bank to deploy a DGX SuperPOD with DGX H100 systems, installed with NVIDIA Professional Services ahead of typical timelines.
- Dec 7, 2022
Deutsche Bank innovation partnership[8]
Deutsche Bank announces a multi-year partnership and plans to use NVIDIA AI Enterprise for risk-model development and high-performance computing, on-premises and on Google Cloud, plus an Omniverse avatar and 'Financial Transformers' language models.
- Aug 24, 2021
NVIDIA AI Enterprise generally available[16]
NVIDIA makes its AI Enterprise software suite available for VMware vSphere on NVIDIA-Certified Systems from server makers including Dell, HPE, Lenovo and Supermicro.
- Jan 1, 1993
Founded[20]
Jensen Huang founds NVIDIA and has led it as president and chief executive since (exact date not given in the source).
What do banks use NVIDIA for?
| Use | What it does | Status | Use case |
|---|---|---|---|
| Real-time card fraud scoring[2] | American Express scores transactions with LSTM models trained on DGX and served on Triton Inference Server inside a two-millisecond latency requirement, combined with its gradient boosting model. | In production | Fraud detection |
| Agent assistants and multi-agent workflows[3][4] | Capital One's agent servicing tool for thousands of customer-service agents uses the NVIDIA enterprise AI ecosystem, and its multi-agent workflows run on an NVIDIA inference stack with Triton and TensorRT-LLM. | In production | Generative & agentic AI |
| Forecasting and trade analytics on a DGX SuperPOD[1] | BNY's SuperPOD supports deposit forecasting, payment automation, predictive trade analytics and end-of-day cash balances. | In production | Trading & capital markets |
| On-premises GPU compute for in-house models[1][5][6] | BNY uses its SuperPOD to expand on-premises AI infrastructure; PNC is installing GPUs in its own data centres so models run on 'our token inside of our own data centers'. | Rolling out | Data & privacy |
| Evaluating and specialising models for banking tasks[7] | Santander AI Lab's Model Arena evaluates open-weight models against banking tasks before specialising them with NeMo Customizer; the work is in experimental use cases. | Pilot | Model risk management |
| Reducing dependence on usage-priced external models[5] | PNC says owning GPU compute is meant to reduce long-term reliance on external, usage-based pricing and avoid vendor lock-in, while still using third-party models where they fit. | Announced | Third-party & vendor AI |
What has NVIDIA disclosed in numbers?
| Metric | Value | As of |
|---|---|---|
| Revenue, fiscal 2026 | $215.9 billion, up 65%[15] | Jan 25, 2026 |
| Revenue, second quarter of fiscal 2027 | $96.2 billion, of which data centre $89.0 billion[14] | Jul 26, 2026 |
| Employees | more than 42,000 in 38 countries[22] | Oct 26, 2026 |
| American Express fraud-model throughput on GPUs versus CPUs (vendor-reported, NVIDIA customer story) | 50x[2] | Oct 15, 2026 |
| Santander model-serving cost per interaction in selected experimental use cases (bank-reported) | from around €5 to approximately €0.03[7] | Oct 19, 2026 |
“By deploying NVIDIA’s AI supercomputer, we can accelerate our processing capacity to innovate and launch AI-enabled capabilities that help us manage, move and keep our clients’ assets safe.”[1]
Bridget Engle · Chief Information Officer, BNY (at the time) · Mar 18, 2024
“We will have our own GPU compute.”[5]
Bill Demchak · CEO, PNC · Jun 11, 2026
“NVIDIA is a key player in the infrastructure, software and research that is powering the latest advances in artificial intelligence.”[7]
Iñaki Bernal · Chief Artificial Intelligence Officer, Santander · Oct 19, 2026
Who runs NVIDIA?
Which rules apply to a bank using NVIDIA?
| Authority | Why it matters here | Documents |
|---|---|---|
| OCC | Buying GPUs, DGX systems and NVIDIA AI Enterprise subscriptions with support is a third-party relationship under the 2023 interagency guidance: due diligence, contract terms, ongoing monitoring and exit planning for hardware and software a bank runs for several years. | OCC Bulletin 2023-17 FDIC FIL-29-2023 |
| Federal Reserve | The Fed's third-party guidance covers the supplier; SR 26-2 covers quantitative models such as fraud models served on NVIDIA hardware but leaves generative and agentic AI, including open models a bank fine-tunes with NeMo, to broader governance. | SR 23-4 SR 26-2 |
| Basel Committee | When on-premises GPU clusters support critical operations, the Basel operational-resilience principles and the 2025 third-party principles on supply chains and concentration apply, including the concentration risk of sourcing accelerators from one supplier. | BCBS Principles for Operational Resilience (2021) BCBS Third-Party Risk Principles (Dec 2025) |
| FSB | The FSB's August 2026 letter asks firms to ensure resilience among critical third-party technology providers as frontier AI raises cyber risk, which reaches the hardware and software layer banks build their AI on. | FSB Chair's letter to G20 (Aug 2026) |
| NIST | NIST's Generative AI Profile lists value-chain and information-security risks for third-party AI components, relevant when a bank serves open-weight models through NIM or Triton in its own data centres. | NIST AI 600-1 (Generative AI Profile) |
Linked authorities have a page in the Regulation tracker; the others are named for completeness.
What should a bank note before buying NVIDIA?
Analysis by BankingNewsAI from the sources cited on this page. Observations, not advice or a recommendation.
01
Most named bank uses are in production
BNY, American Express and Capital One describe NVIDIA systems already in use; PNC's compute is being installed, and the Santander and Deutsche Bank entries are group-level, one experimental and one a 2022 plan.
02
Software is priced, hardware is not
NVIDIA AI Enterprise list prices are public; the GPUs and DGX systems that make up most of a deployment are sold on quotes through NVIDIA and partners, and no bank deal on this page disclosed a value.
03
Results are reported by the vendor or the bank
The 50x throughput figure comes from NVIDIA's American Express customer story and Santander's cost reduction from selected experiments; neither has been independently measured.
04
Partnerships are not purchases
JPMorgan Chase's agent-safety work with Nvidia, Goldman Sachs's compute-financing memorandum and the Kinexys federated-learning test involve NVIDIA without a disclosed purchase, so they are not counted as customer relationships here.
05
Supplier concentration is a live question
NVIDIA's outlook excludes China data-centre revenue under export controls, supervisors are asking about critical technology providers, and JPMorgan Chase chose SambaNova hardware for on-premises inference in 2026.
What are the alternatives to NVIDIA?
SambaNova
AI-inference chip maker selling RDU systems for on-premises use (SambaRack, SambaStack) and a per-token open-model API (SambaCloud).
How much does NVIDIA AI Enterprise cost?
NVIDIA's licensing guide lists NVIDIA AI Enterprise at $4,500 per GPU for a one-year self-managed subscription with standard support, $22,500 per GPU for a perpetual licence with five years of support, or $1 per GPU per hour plus the cloud instance cost on cloud marketplaces, as checked on October 26, 2026. A five-year subscription is included with H100, H200 NVL and A800 PCIe GPUs. GPU and DGX system prices are not published.
Which banks use NVIDIA?
Among the banks tracked on this site, BNY runs an on-premises DGX SuperPOD (since March 2024), American Express serves fraud models on DGX and Triton, Capital One builds agent tools on NVIDIA's software and inference stack, and PNC is installing Nvidia GPUs for its own models. At group level, Santander specialises and serves models with NVIDIA software and Deutsche Bank announced plans to use NVIDIA AI Enterprise in 2022.
Who owns NVIDIA?
NVIDIA is a public company listed on Nasdaq under the ticker NVDA. Jensen Huang founded it in 1993 and has served as president and chief executive since its inception. Its fiscal 2026 revenue, for the year to January 25, 2026, was $215.9 billion, and its second quarter of fiscal 2027 brought in $96.2 billion.
Can a bank run NVIDIA AI on-premises?
Yes. NVIDIA AI Enterprise is licensed per GPU for self-managed systems as well as by the hour on cloud marketplaces, and BNY's DGX SuperPOD, PNC's own data-centre GPUs and Santander's in-house model serving with Dynamo-Triton are on-premises examples. Banks still buy the hardware itself on quotes from NVIDIA or its server partners.
Does JPMorgan Chase use NVIDIA?
Not on the sources this site tracks. JPMorgan Chase joined an Nvidia-led effort on open-source safety tools for AI agents in September 2026, with no purchase announced, and its Kinexys unit ran a federated-learning fraud test on NVIDIA FLARE and DGX Cloud with BNY, RBC and DeepTempo. For on-premises inference it selected SambaNova systems in July 2026.
NVIDIA vs SambaNova for bank AI inference?
Both sell hardware and software for running models on infrastructure a bank controls. NVIDIA's GPUs and AI Enterprise software are used by several banks on this site and its software prices are published; SambaNova sells its own RDU chips as on-premises racks and a per-token cloud API, with JPMorgan Chase as its named bank customer. Neither publishes on-premises system prices.
- BNY First Global Bank to Deploy AI Supercomputer Powered by NVIDIA DGX SuperPOD With DGX H100
- AI to Combat Financial Fraud: American Express customer story (undated page, retrieved 2026-10-15)
- Highlights from NVIDIA GTC AI Conference 2025 (Capital One sessions and speakers)
- How Capital One built production multi-agent AI workflows to power enterprise use cases
- Why PNC is building its own 'AI factory'
- PNC (PNC) Q3 2026 Earnings Call Transcript
- Santander builds AI capabilities that improve performance and lower costs using NVIDIA AI infrastructure
- Deutsche Bank Partners with NVIDIA to embed AI into Financial Services
- Kinexys Brings Federated Learning for Fraud Detection
- JPMorgan Stock Slips 1% as Bank Agents Get an Open Safety Layer
- RBC Capital Markets Advances Research and Banking Capabilities With Agentic AI (undated page, retrieved 2026-10-05)
- JPMorgan Chase picks SambaNova for on-prem AI inference
- Leading European Financial Services Organizations Embrace Open Models to Build Proprietary Intelligence
- NVIDIA Announces Financial Results for Second Quarter Fiscal 2027
- NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2026
- Global Availability of NVIDIA AI Enterprise Makes AI Accessible for Every Industry
- NVIDIA NIM Revolutionizes Model Deployment, Now Available to Transform World's Millions of Developers Into Generative AI Developers
- Palantir and NVIDIA Team Up to Operationalize AI — Turning Enterprise Data Into Dynamic Decision Intelligence
- Pricing — NVIDIA Enterprise Licensing Guide (undated page, retrieved 2026-10-26)
- Jensen Huang, Founder & CEO (undated biography, retrieved 2026-10-26)
- Colette Kress, EVP and Chief Financial Officer (undated biography, retrieved 2026-10-26)
- NVIDIA in Brief 2026 (retrieved 2026-10-26)
When a bank signs with NVIDIA, you'll read it here first.
when a bank picks an AI vendor, or a vendor ships for banks, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning