What technology stack does Bank of America run its AI on?
Bank of America builds most of its stack in-house and says so in numbers. It spends about $14 billion a year on technology with about 60,000 technology employees. Its private cloud cut servers to about 70,000 from more than 200,000 and data centers to 23 from 67, saving $2 billion a year, Moynihan said in 2019, while the bank still buys external cloud where third parties are cheaper and was first major user of IBM's financial-services cloud. For AI, CTIO Hari Gopalkrishnan said on October 1, 2026 that an orchestration layer called Orchestra sends simple classification to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models, partly to control token costs. Erica was rebuilt in 2025 as a second-generation, model-agnostic platform for reuse. Employees use Microsoft 365 Copilot (200,000 teammates by January 2026); 19,000 developers use coding assistance; Redis supports caching, event streaming and AI infrastructure. In 2026 the stack also had to absorb Anthropic's Mythos: 1,000 contractors joined 3,000 staff to patch flaws in internal and vendor code.[1][2][3][4][5][6][7]
Data as of Oct 16, 2026 · BofA profile updated Oct 13, 2026 · 20 sources cited on this page · 140 banks profiled on this site · Markdown
What are the layers of Bank of America's AI and technology stack?
As executives, filings and partners describe them. The bank has not published an architecture or named its proprietary model providers or GPU vendors.
| Layer | What is on the record | Source | Date |
|---|---|---|---|
| Budget and people | About $14 billion a year, more than $4.1 billion for new initiatives; about 60,000 technology employees | Fast Facts | Retrieved October 2026[1] |
| Infrastructure | Private cloud: about 70,000 servers from more than 200,000, 23 data centers from 67, $2 billion a year saved; external cloud used where 23%-30% cheaper | Moynihan via Reuters | November 2019[2] |
| Public cloud | IBM's financial-services-specific cloud, built with Bank of America as first major bank user | Reuters | November 6, 2019[2] |
| Model routing | Orchestra sends simple classification to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models; controls token costs | Gopalkrishnan | October 1, 2026[3] |
| Assistant platform | Erica rebuilt as a second-generation, model-agnostic platform in 2025 after first-generation BERT and OpenNLP; 'build once' reuse across lines of business | Fortune | March 17, 2026[4] |
| Data layer | Redis for caching, session management, event streaming and AI infrastructure | Redis | October 10, 2024[6] |
| Employee AI | Microsoft 365 Copilot to 200,000 teammates; about 200,000 active users of general-purpose AI tools and 2,000-plus departmental templates | Q4 2025 call; Q2 2026 deck | January and July 2026[5][8] |
| Software delivery | 19,000 developers with real-time coding assistance, efficiency gains over 20%; AI took 30% out of the coding stream in 2025 | Fast Facts; Q4 2025 call | 2026[1][5] |
| Security engineering | 1,000 contractors added in Q2 2026 to 3,000 staff to patch internal and vendor software flaws exposed by Mythos; engineering workforce about 10% more efficient | Moynihan via CNBC and The Register | October 14, 2026[7][9] |
| Partner pipeline | Technology Innovation Summit: AI, data and analytics, enterprise infrastructure, security and software development; looking for tools to automate infrastructure management | Gopalkrishnan via Banking Dive | September 30, 2026[10] |
Does Bank of America build or buy its AI technology?
Mostly build, with selective buying. Erica, EricaAssist and the Orchestra routing layer are in-house; the bank calls itself model-agnostic and has not named the proprietary model suppliers behind Orchestra. In 2019 Moynihan said 'We don't need to own the hardware. We just need to find out who can provide it the right way', even as the private cloud saved $2 billion a year.[3][4][2]
Bought technology enters through the summit: Gopalkrishnan says about one-fifth of the 7,000 companies that have attended became vendors, and the bank uses the event to tell startups what scale, security and multilingual requirements a large regulated buyer has.[10]
Moynihan's warning on Mythos also shapes vendor policy: 'we've got to be mindful that the people doing the software development are the people trying to sell us the services'.[7]
What does the BofA profile record on this topic?
The rows below are the entries of the Bank of America AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 45 dated moves, 18 use cases and 68 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Frontier-model testing | Part of Anthropic's Glasswing programme from the start; testing Mythos internally, per Reuters sources. | Pilot | [11] |
| AI-assisted software development | About 19,000 developers use real-time coding assistance, raising productivity by more than 20%; AI took 30% out of the coding stream in 2025. | In production | [8][5] |
| Orchestra model-routing layer | Sends simple classification tasks to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models, which the CTIO says also controls token costs. | In production | [3] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Oct 1, 2026 | CTIO at Fortune AIQ Summit: Orchestra routes work to open-weight or proprietary models | Gopalkrishnan says AI is often rejected in favour of deterministic rules, every project is reviewed against 16 risk pillars, simple classification runs on approved open-weight models on the bank's own GPUs, and agents stay assistive with humans in the loop. He gives Erica's total as 3.6 billion interactions. | [3] |
| Sep 23, 2026 | DeMare: coding is the clearest AI return | At a BofA Securities conference the co-president says about 20,000 developers use coding agents for a 15% to 20% gain, about 95% of staff have AI access, and the bank is assessing third-party providers; Gopalkrishnan says AI investment interest is more than twice last year's and spending will probably double. | [12] |
| Jul 16, 2026 | Moynihan on Mythos: 'a big change' | The CEO joins bank chiefs warning that frontier models such as Anthropic's Mythos change how fast system vulnerabilities must be fixed. | [13] |
| Jul 14, 2026 | Q2 2026 deck adds a dedicated AI slide | Three levels of AI: general-purpose tools (~200,000 users), functional tools for wealth, banking and developers, and multi-process systems for call centres, payments and document intelligence; more than 600,000 AI training courses completed. | [8][14][15] |
| Mar 17, 2026 | 'Build once' strategy explained: second-generation Erica platform in 2025 | Fortune reports that an AI council with the head of strategy ran about 15 proofs of concept outside the Erica platform, found teams 'reinventing the wheel', and the bank rebuilt Erica as a second-generation, model-agnostic platform in 2025; the first generation started on BERT and OpenNLP. | [4] |
| Jan 14, 2026 | Q4 2025 call: coding 30% faster, Copilot for 200,000 | Moynihan says AI took 30% out of the coding stream for the bank's 18,000 coders, saving about 2,000 people, that Microsoft 365 Copilot is out to 200,000 teammates, and that direct AI spend is 'several hundred million dollars'. | [5] |
| Nov 5, 2025 | Investor day: 270 models, fraud losses halved, 18,000 coders on agents | Moynihan says the bank runs 270 AI and machine-learning models; the CTIO reports 18,000 developers on coding agents and a 20% productivity gain; Holly O'Neill says AI-enabled processes cut fraud losses in half since 2018. Technology spend is about $100 billion over a decade, $4 billion a year on new initiatives and $1 billion on cybersecurity. | [16][17][18] |
| Sep 30, 2025 | CashPro gets a generative-AI assistant | The bank restates its four AI domains: intelligent agents, search and summarisation, content generation, operations and coding. | [19] |
| Jul 16, 2025 | Q2 2025 deck maps AI across four domains | Erica for Employees used by more than 90% of teammates, cutting service-desk calls by about half; an AI assistant for about 17,000 developers; more than 50 AI-enabled fraud-detection models; about 1,400 AI and machine-learning patents granted and pending. | [20] |
Does Bank of America use the public cloud?
It relies mainly on a private cloud, which Moynihan said in 2019 had cut servers to about 70,000 from more than 200,000 and data centers to 23 from 67, saving $2 billion a year. It still buys external cloud where third parties are 23%-30% cheaper and was the first major bank on IBM's financial-services cloud.
Which AI models does Bank of America run?
It says it is model-agnostic. CTIO Hari Gopalkrishnan said in October 2026 that Orchestra routes simple classification to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models it has not named. Erica's first generation was built on BERT and OpenNLP.
How many technologists does Bank of America employ?
About 60,000 people work in its global technology team, according to the bank's Fast Facts page, which also puts annual technology spend at about $14 billion, more than $4.1 billion of it for new initiatives including AI.
How has Anthropic's Mythos affected Bank of America's technology work?
Moynihan told CNBC on October 14, 2026 that the bank added 1,000 contractors in the second quarter to the 3,000 people already patching software, to fix internal and vendor flaws exposed by frontier models including Mythos; remediation and AML costs together were 'a couple hundred million plus' in Q3.
- Bank of America Fast Facts
- IBM, Bank of America team up to launch financial services-specific cloud
- AI in regulated industries: BofA and S&P on the huge gains to be had, and the risks of rushing in
- Inside Bank of America's 'build once' AI strategy
- BofA (BAC) Q4 2025 Earnings Call Transcript
- Redis Honored by BofA for its Leading Real-Time Data, AI Infrastructure
- Bank of America hired 1,000 contractors to patch software flaws exposed by Anthropic's Mythos
- Second Quarter 2026 Financial Results presentation (Form 8-K, Exhibit 99.2)
- Bank of America had to bring in 1,000 contractors to patch all the bugs Mythos identified
- How BofA's tech summit influences strategy
- Banking industry scrambles for Anthropic's Mythos as global regulators review risks
- Bank of America plans to double AI budget next year
- BofA's Moynihan Joins Bank CEOs Warning About Mythos AI Risks
- Bank of America (BAC) Q2 2026 Earnings Call Transcript
- Banks report operational changes driven by AI adoption
- BofA's Moynihan pledges growth, 'no excuses'
- BofA co-presidents pledge rigor on investor day targets
- Bank of America runs 270 AI models across operations
- BofA's New GenAI Assistant Transforms Global Payments Solutions
- Second Quarter 2025 Financial Results presentation (Form 8-K, Exhibit 99.2)
When BofA 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