What is Capital One's technology stack?
Capital One runs on the public cloud: it closed its last of eight data centers in 2020 after an eight-year migration begun in 2012, the first US bank to announce going all in, and its 2025 10-K says it has 'substantially migrated primarily all aspects' of core infrastructure and customer applications to third-party cloud platforms, 'principally AWS'. On top sit a serverless-first architecture with tens of thousands of Lambda functions across thousands of AWS accounts, a One Pipeline CI/CD platform that checks governance and risk controls, Snowflake (more than 6,000 analysts in 2022) and Databricks (4,000-plus users, 5 million-plus jobs in 2025) for data, and in-house AI built by customising open-weight models on more than 20 trillion tokens of training data. The technology organisation grew from 11,000 (85% engineers) in 2020 to about 12,000 in 2022-23 and more than 14,000 by 2025, 88% of engineers holding AWS certifications, trained through an internal Tech College since 2017. Tools built for itself became products: Capital One Software (June 2022) sells Slingshot and Databolt to other enterprises, and Data Profiler, rubicon-ml and VulnHunter are open source. CEO Richard Fairbank says the stack 'was built from the outset working backwards from the AI revolution'.[1][2][3][4][5][6][7][8][9][10][11][12][13]
Data as of Oct 15, 2026 · Capital One profile updated Oct 1, 2026 · 31 sources cited on this page · 140 banks profiled on this site
What runs where in Capital One's stack?
Layer by layer, as the bank and its vendors describe it. 'Since' is the earliest public date for that layer, not necessarily first use.
| Layer | What Capital One uses | Evidence | Since |
|---|---|---|---|
| Cloud infrastructure | AWS, all in: eight on-premises data centers exited; nearly 2,000 applications in the cloud, 80% built from the ground up; more than 30 AWS services including EC2, S3, RDS and Amazon Connect; active-active with automated failover | AWS case study; the 10-K names AWS as the principal platform and a dependency risk | 2020 (exit); 2012 (start)[1][3] |
| Compute pattern | Serverless-first: thousands of AWS accounts, tens of thousands of Lambda functions, a Serverless Center of Excellence; DynamoDB for key-value, DocumentDB for documents, Amazon WorkSpaces for virtual desktops | Capital One serverless hub; AWS re:Invent 2022 recap | 2022[4][5] |
| Delivery | One Pipeline, a single CI/CD platform that checks governance policies, risk controls and standards; agile and DevOps since the migration; Operational Readiness Reviews by distinguished engineers | AWS; Capital One Software blog | 2022-2024[5][12][1] |
| Data platforms | Snowflake: more than 6,000 analysts running millions of queries (2022), Slingshot cut costs 27% against projections; Databricks: 4,000-plus users, more than 5 million jobs (2025), used 'to power advanced analytics and machine learning' | Capital One Software releases | 2022; 2025[6][7] |
| AI models | Open-source LLMs customised in-house on synthetic and public data (20 trillion-plus tokens); Meta Llama base for Chat Concierge; multi-agent MACAW on fine-tuned open-weight models; smaller specialised models to cut latency and token cost | Bank disclosures | 2025-2026[8][14][15][16] |
| Contact centre | Amazon Connect; thousands of agents worked remotely on it during 2020 | AWS and Capital One | 2020[12][1] |
| Security tooling | VulnHunter agentic code-security tool (open-sourced); Databolt tokenisation for sensitive data; patents on AI cyber-threat detection and threat-model statements | Bank, VentureBeat, Google Patents | 2025-2026[17][18][19][20] |
| Core processing (outside the cloud story) | TSYS for card processing in North America and the UK; FIS for certain banking systems; Discover's Global Payment Network still on a mix of data centers and vendors | 2025 10-K | February 2026[3] |
| Open source | Open-source-first since 2015: Data Profiler, rubicon-ml, Global Attribution Mapping, VulnHunter | Capital One | 2015-2026[12][21][22][17] |
How big is Capital One's technology organisation?
Headcount figures as published, in date order; the bank uses 'technologists' and 'engineers' loosely, so each row keeps the original wording.
| Figure | Wording | Who | Date |
|---|---|---|---|
| 11,000 | 'An 11,000-person tech organization with 85% of that being engineers' | Melanie Frank, managing VP of technology, at AWS re:Invent | December 1, 2020[2][1] |
| 12,000 | 'Capital One's 12,000-strong technology team' | Forbes, on the Capital One Software launch | June 1, 2022[9] |
| ~12,000; 88% | Tech College serves '~12,000 engineers'; '88% of Capital One's engineers currently hold an AWS certification' | Capital One Tech | December 7, 2023[10] |
| 14,000-plus | '14,000-plus army of engineering and technology professionals' tapped to build Databolt | Banking Dive | August 6, 2025[11] |
| More than 14,000 | 'A tech workforce of more than 14,000 people, mostly engineers'; 'one of the largest users of AWS' | Capital One serverless page | Retrieved October 2026[4] |
| 50,000 | Associates overall at the time of the migration | AWS case study; Capital One Software | 2020-2024[1][12] |
| Tech College | Internal technology school since 2017; monthly 'Invest in Yourself' days for AWS certification; a Machine Learning Engineering Training Program | Capital One | 2017; 2023[12][10] |
How did Capital One get to this stack?
| Date | Event | Detail |
|---|---|---|
| 2012 | Technology transformation begins | Goal: 'a technology company that does banking instead of a bank that just uses technology'; cloud chosen so the bank could use machine learning at scale[1][12] |
| 2015 | Open-source first | Policy adopted; later projects include Data Profiler and rubicon-ml[12] |
| 2017 | Tech College launches | Internal training in cloud, security, data and machine learning[12] |
| 2020 | Last data center closed | Eight data centers exited; 103 tons of copper and steel recycled; an expected 10 megawatts a year saved; 'first US bank to announce going all in on the cloud'[1][2][23] |
| June 1, 2022 | Capital One Software launches | Enterprise B2B software business under Ravi Raghu; first product Slingshot for Snowflake, after 27% cost savings in-house[6][9] |
| November 2022 | Serverless-first on stage | One Pipeline and Lambda standardisation presented at AWS re:Invent[5] |
| April 2025 | Databolt commercial rollout completes | Tokenisation product born from an eight-year internal effort; integrated with Snowflake and Databricks in May 2025[11] |
| June 10, 2025 | Databricks partnership deepens | Databolt integrates with Unity Catalog; Slingshot adds Databricks optimisation; Sync Computing team joins[7] |
| January 22, 2026 | 'Working backwards from the AI revolution' | Fairbank on the Q4 2025 call: the stack was built from the outset for AI[13] |
| February 19, 2026 | 10-K: cloud dependence disclosed | 'Substantially migrated primarily all aspects' of core IT to third-party cloud platforms, 'principally AWS'; AWS outages named as a risk; Discover's network still on data centers and vendors[3] |
| March 23, 2026 | Databolt expands for AI | Capabilities for sensitive unstructured data used by AI[18] |
| June 16, 2026 | Databolt Connect | Secure multi-party data sharing inside Databricks Clean Rooms, on Databricks Marketplace[24] |
| July 16, 2026 | VulnHunter open-sourced | Internally built agentic code-security tool released under Apache 2.0[17] |
What does the Capital One profile record on this topic?
The rows below are the entries of the Capital One AI strategy profile that bear on this page, as of Oct 1, 2026; the profile holds the full record of 31 dated moves, 13 use cases and 46 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Digital and customer-service AI | Part of the enterprise AI program presented at re:Invent. | In production | [25] |
| Databolt (Capital One Software) | Tokenisation and data-security tooling sold to enterprises for secure AI adoption. | In production | [18] |
| MACAW fraud-call summaries | Understanding, reasoning, validation and explanation agents on fine-tuned open-weight models summarise long fraud calls for several hundred specialist agents; the same architecture runs Chat Concierge. | In production | [15] |
| Proprietary LLMs customised from open-source models | Generative AI models built by customising open-source LLMs with synthetic and licensed open-source data; the disclosed training set exceeds 20 trillion tokens. | In production | [8][15] |
| VulnHunter | Agentic code-security tool with attacker-first analysis and a falsification engine; open-sourced in July 2026. | In production | [17] |
| AI and model risk in the 10-K | The 2025 10-K names models and AI in pricing, fraud detection, loan grading, credit decisions and software development as a risk, plus agentic-commerce fraud and AI intellectual-property uncertainty. | In production | [3] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Sep 29, 2026 | Natarajan: cutting latency and token costs with smaller, open-weight models | Chief enterprise AI officer Prem Natarajan says the bank has put more emphasis on agentic AI infrastructure to reduce latency and lower token costs, building customised open-weight and specialised models with less context targeted at the desired output; customer-service agents now edit only one-fifth of AI-drafted copy. | [16] |
| Sep 28, 2026 | Shetty: 'agentic AI now must be treated as an end-to-end system' | VP of enterprise AI Rashmi Shetty says 14 years of data-infrastructure work gives agents a strong foundation and context, and that the bank adopts platform-first governance — rule-based validation, sandbox simulations, observability of agent trajectories, standardised harnesses and human-in-the-loop controls for high-risk actions — before developers build agents, rather than retrofitting it later. | [26] |
| Aug 13, 2026 | MACAW and open-weight models, described | A machine-learning engineering leader explains the multi-agent fraud-call workflow and fine-tuning open-weight models on proprietary data. | [15] |
| Jul 21, 2026 | Q2 2026 call: AI infrastructure investment continues; Brex models to be calibrated | Investment focus stays on the technology transformation and AI infrastructure; full Brex acceleration needs data pipelines and model calibration. | [27][28] |
| Jul 16, 2026 | VulnHunter open-sourced | An internally built agentic AI code-security tool released on GitHub under Apache 2.0; the CISO cites the brief window before AI attack tools spread. | [17] |
| Apr 21, 2026 | Q1 2026 call: AI leverage is 'vastly greater' when embedded | Fairbank: 14th year of the technology transformation; continued investment in AI infrastructure and specific AI experiences. | [29] |
| Mar 23, 2026 | Databolt expands for secure AI adoption | Capital One Software's data-security product adds capabilities for sensitive unstructured data. | [18] |
| Mar 17, 2026 | 'Building AI moats': the chief scientist on Bloomberg | Why the bank builds rather than buys its AI stack; AI as a capacity multiplier. | [30] |
| Feb 19, 2026 | 2025 10-K: 'our evolving use of AI' is a risk factor | Models and AI in pricing, fraud detection, loan grading, credit and software development; agentic-commerce fraud liability; uncertain IP protection for AI output. | [3] |
| Jan 22, 2026 | Q4 2025 call: a stack built 'working backwards from the AI revolution' | Fairbank says the bank is 'now building AI solutions across our businesses' and that Brex has deployed in-house AI agents for expense management and audit. | [13] |
| Jan 22, 2026 | Agrees to buy Brex for $5.15 billion | Corporate cards, expense software and nearly $13 billion in deposits at partner banks. | [31] |
| Dec 1, 2024 | Enterprise AI at scale, presented at AWS re:Invent | The bank lays out use cases across AML, cybersecurity, digital and customer service, and its enterprise AI program. | [25] |
Is Capital One really all in on the cloud?
Yes, with one disclosed exception. Capital One closed the last of its eight data centers in 2020 after an eight-year migration and describes itself as the first US bank to announce going all in on the public cloud. The 2025 10-K says it has 'substantially migrated primarily all aspects' of its core IT infrastructure and customer-facing applications to third-party cloud platforms, 'principally AWS', and lists AWS outages as a risk. The exception is Discover, acquired in 2025, whose Global Payment Network is still processed through a combination of data centers and third-party vendors during integration.
What does Capital One use AWS for?
Nearly everything: close to 2,000 applications ran in the cloud at the time of the exit, 80% built from the ground up, across more than 30 AWS services including EC2, S3, RDS, Lambda, DynamoDB, DocumentDB, WorkSpaces and Amazon Connect for the contact centre. The architecture is serverless-first, with thousands of AWS accounts and tens of thousands of Lambda functions governed by a Serverless Center of Excellence and a One Pipeline CI/CD platform. Eno and the mobile app are built on EC2, Lambda and DynamoDB. The bank calls itself one of the largest users of AWS.
What is Capital One's data platform?
Two commercial platforms that Capital One both uses and sells tooling for. Snowflake: more than 6,000 analysts ran millions of queries in 2022, and the bank's internal provisioning tool, Slingshot, saved 27% against projected costs before it was sold to others. Databricks: more than 4,000 users running over 5 million jobs by June 2025, described as foundational infrastructure for AI/ML; Capital One Ventures is also an investor. Databolt tokenises sensitive data in both, and Databolt Connect (June 2026) links tokenised datasets across parties inside Databricks Clean Rooms.
How many engineers does Capital One have?
More than 14,000 technology staff, 'mostly engineers', per the bank's own serverless page and a 14,000-plus figure reported by Banking Dive in August 2025. Earlier figures: 11,000 with 85% engineers at the end of the migration in 2020, 12,000 in June 2022 and about 12,000 engineers in December 2023, 88% of them AWS-certified. The bank had 50,000 associates overall around 2020-24.
What is Capital One Software?
An enterprise B2B software business launched June 1, 2022 under Ravi Raghu to sell tools the bank built for itself. Slingshot (2022) manages Snowflake and, since 2025, Databricks cost and governance; Databolt (commercial rollout April 2025) tokenises sensitive data, was extended to unstructured data for AI in March 2026 and gained Databolt Connect for clean-room data sharing in June 2026. CIO Rob Alexander told Forbes in 2022 that winning enterprise software businesses take five to seven years to reach $100 million in revenue; the bank has not disclosed the unit's revenue.
What open-source software has Capital One released?
The bank adopted an open-source-first policy in 2015. Named releases: Global Attribution Mapping, an explainability method for neural networks (2019); Data Profiler for data monitoring and sensitive-data detection and rubicon-ml for capturing model-development runs (2021); and VulnHunter, an agentic AI code-security tool released under Apache 2.0 in July 2026. Chat Concierge and the MACAW workflow are built on open-weight models (Meta's Llama among them) rather than released as open source.
Who runs technology at Capital One?
Robert M. Alexander, Chief Information Officer, who has been with the bank about 20 years and oversaw the move to the public cloud, per the 10-K; Chris Nims, EVP and Chief Information Security Officer, who as SVP of cloud and productivity engineering led the data-center exit; Andy Ozment, Chief Technology Risk Officer; Amy Lenander, Chief Data Officer; and, for AI, Prem Natarajan, Chief Scientist and Head of Enterprise AI. Founder and CEO Richard Fairbank has driven the transformation since 2012.
- Capital One Completes Migration from Data Centers to AWS, Becomes First US Bank to Announce Going All In on the Cloud (case study; page retrieved 2026-10-15)
- Capital One closes all data centers, relies on AWS on-demand infrastructure
- Capital One Financial Corporation Annual Report on Form 10-K for 2025
- Serverless at Scale (Capital One Tech hub page, retrieved 2026-10-15)
- AWS re:Invent 2022 wrap up: Capital One optimizes for productivity, efficiency, and security on AWS
- Capital One Launches Capital One Software (Slingshot for Snowflake)
- Capital One Software strengthens partnership with Databricks
- GAI Training Data Transparency Disclosure
- Exclusive: Banking Giant Capital One Enters B2B Software Industry With Launch Of New Business
- Tech Career Development: Shape Your Future (Tech College, ~12,000 engineers)
- Why Capital One's software division created its own data security platform (Databolt)
- Lessons from Capital One's cloud migration journey
- Capital One (COF) Q4 2025 Earnings Call Transcript
- AI for Car Dealerships: Smarter Car Shopping
- Why Capital One built its multi-agent AI platform around open-weight models (presented by Capital One)
- How bottom-line-focused financial firms are finding ROI in AI
- Capital One releases VulnHunter, an open-source AI tool that finds software flaws before hackers do
- Capital One Software Expands Databolt Capabilities
- US12603903B2: Methods and systems for cyber threat detection using artificial intelligence models in data-sparse environments (granted)
- US12682077B2: Generating threat statements for a threat model using artificial intelligence (granted)
- How Capital One is developing for the bank of the future
- Global Attribution Mapping (GAM): open-sourcing explainability for neural networks
- See how Capital One is using AWS machine learning at scale to transform banking and payments
- Capital One Software Introduces Databolt Connect for Secure Data Collaboration on Databricks Marketplace
- How AI is transforming financial services & banking
- Capital One's agentic AI strategy hinges on data, platform-first mindset
- Capital One Financial Corporation Q2 2026 Earnings Call Summary
- Capital One highlights Discover, Brex integration
- Capital One (COF) Q1 2026 Earnings Transcript
- Capital One on Building AI Moats in Banking (Tech Disruptors)
- Capital One Strikes $5.15 Billion Deal for Fintech Brex
When Capital One moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning