AI ROI Series | EP29 | JPMorgan Spends $700M a Year on AI | Say What?!
JPMorgan spends about $700 million a year on AI, and its CFO says the benefits "aren't observable in our current quarterly earnings." Paul Brzozowski, Co-Founder of Olakai, the Enterprise AI System of Record, breaks the math on the number one AI bank in the world. We start with the $700 million itself: what it is per day, per user, and how many hours each employee has to save to pay for it. Then the credit JPMorgan has earned, what it built, the value math on the four hours a week its people say they save, the token bill, build versus buy, and the one bank that did put a number on its AI. It closes with four questions every AI leader should be able to answer this quarter, and the one line I'd take to the board. If you own the AI budget, or defend it to a CFO, this one is for you. The math is directional, so check it and tell me where you read it differently. Chapters 0:00 JPMorgan spends $700M a year on AI. What does it return? 0:53 Welcome, and today's plan 1:27 What does $700M a year look like per day? 1:38 Why the number used to be $2B 2:22 How many hours does each user need to save to break even? 3:35 Is JPMorgan really the best AI bank? 4:23 What did JPMorgan build with it? 5:42 What is JPMorgan's AI actually worth? 6:49 Why is the token bill growing? 8:05 Should a bank build its own AI measurement? 9:10 Has any bank put a number on AI ROI? 9:29 Did JPMorgan's AI live up to the hype? 9:54 What is an Enterprise AI System of Record? 10:58 Four questions every AI leader should answer 11:23 The one line for the board Key numbers AI expense: about $700M a year (CFO, Q3 2026 call), against the $2B Jamie Dimon cited in October 2025. About $1.9M a day. Per user: about $4,700 a year, or $390 a month, across about 150,000 weekly LLM Suite users (our math). Break-even: at $100 an hour, $700M buys 7M hours, about 47 hours per user per year, roughly one hour a week. Employees report saving four (our math). Value if the four hours are real: 150,000 people x 4 hours x 46 weeks is about 27.6M hours, roughly $2.8B a year (our math). Technology budget: $19.8B, so AI is about 3.5% of it. Claude Code: a reported $2,000 monthly cap across about 8,000 licenses, a ceiling of up to $192M a year (our math; ceiling, not actual spend). GenAI gateway job posting: $204,250 to $285,000 base, including "per-tenant cost attribution." Bank of America: about $400M of AI spend generating about $800M of benefit. MIT NANDA: tools bought from specialist vendors succeeded about two-thirds of the time, internal builds about a third as often. Gartner: inference cost per agentic workflow up more than 5x through 2028. Sources: JPMorgan Chase Q3 2026 earnings call, 13 October 2026, as transcribed by The Motley Fool and Investing.com; JPMorgan Q2 2026 call; JPMorgan 2026 Company Update transcript, 23 February 2026; Dimon on Bloomberg TV, 7 October 2025; Evident AI Index, October 2026; JPMorgan technology blog and HBS case 9-325-066 on LLM Suite; Business Insider; Semafor, 3 June 2026; JPMorgan job posting, Principal Software Engineer, AI Foundations; Gartner, 17 August 2026; MIT NANDA, "The GenAI Divide," August 2025; Bank of America via Banking Dive and its Q3 2026 call. Want to see this on your own data, or explore our demo environment? DM me on LinkedIn: [your LinkedIn profile URL] Learn more about Olakai, the Enterprise AI System of Record: https://olakai.ai Your AI is an investment. Let's measure it like one. Tags JPMorgan, JPMorgan Chase, Jamie Dimon, Jeremy Barnum, JPMorgan AI, LLM Suite, AI ROI, AI spend, AI budget, bank AI, banking AI, Bank of America AI, Claude Code, AI token cost, AI FinOps, enterprise AI, build vs buy AI, AI measurement, CFO AI, Olakai, Enterprise AI System of Record, AI ROI Series, Paul Brzozowski
