Top Banks Move From AI Pilots to Profits, as Capital One Narrows Gap With JPMorgan
By Yamini Kalra
The New AI Arms Race in Banking Has a Surprise Contender
If there’s a poster child for artificial intelligence in banking, it’s JPMorgan Chase, ranking first among 50 of the world’s largest banks in the Evident AI Index 2025. But the margin is thinning rapidly. Capital One is tugging at the title, coming in at a close second.
Royal Bank of Canada, Commonwealth Bank of Australia – notably the only bank from the Asia Pacific – and Morgan Stanley round out a cohort that has moved from experimentation to industrial strength. What’s common between them, according to the index, is aggressive talent buildouts, AI governance baked into the C-suite and real dollars behind AI use cases that touch everything from legal, trading and credit to customer-facing agents.
Below, we profile the frontrunners: how they build their lead and the AI use cases that underpin it.
JPMorgan Chase
2025 may end with a banger for JPMorgan, which outperformed Wall Street expectations and reported a net income of US$14.4 billion, marking a double-digit increase. The bank claims to have over 450 AI use cases in motion, spending more than US$2 billion a year on AI – and getting returns. On his part, CEO Jamie Dimon spent much of the year advocating for AI use cases in public forums. “There will be no job, no process, no function that won’t be affected by AI – mostly for the positive,” he said at an industry event. At another, he admitted curtly: “It will eliminate jobs.”
This is perhaps why JPMorgan is still the only bank to score 100 per cent in the leadership pillar in the index. It leads across the innovation and transparency metric as well, but closed in on an even split with Capital One in talent.
JPMorgan’s backbone is its internal AI ecosystem, labeled OmniAI, that claims to enable data scientists and business teams to move models from ideation to production with governance and reuse. In-house AI-driven applications such as the ChatCFO – a large language model developed specifically for the company’s finance team – automate workflows, provide data-driven insights and generate financial models. “You would have thought Santa had come down the chimney with the finance team,” said CEO of asset and wealth management Mary Callahan Erdoes.
The organization has also developed the LLM Suite – a proprietary generative AI platform that functions similar to OpenAI’s ChatGPT – for nearly 50,000 employees to enhance productivity through tasks such as summarizing documents, generating insights and solving complex analytical challenges.
Another one of its initiatives, IndexGPT, personalizes investment advice in real time. For a bank that processes over US$6 trillion in consumer payments annually, this is more than just add-on advisory services. It also introduced the Cash Flow Intelligence AI tool that claims to cut manual work by 90 per cent.
The Evident AI Index quotes JPMorgan Chase as declaring that “many users regularly gain 2 to 4 hours of productivity per week and select software development teams have seen productivity improve by double digits.”
Capital One
Capital One is a quarter the size of JPMorgan Chase – the largest bank in the index. Yet, it leads with the largest growth in AI talent volume of any bank year-over-year. Post the company’s merger with Discover, it expanded its AI talent by 73 per cent, preserving its status as the bank with the highest ratio of AI professionals per headcount.
To quote the report, “There were lingering questions about whether Capital One was going to be able to keep its historically unmatched levels of AI talent density after the mega merger with Discover. But the combined enterprise increased its overall headcount by 47 per cent … which resulted in a new high score in terms of AI talent density: 6.7 per cent of total employee headcount.”
That density matters because it means fewer organizational silos, stronger internal peer communities and faster iteration cycles. The result is a company that talks in patents and executions.
Technically, the Virginia-based bank has adopted a pragmatic architecture: Unlike JPMorgan’s proprietary LLM suite, Capital One builds on open-source foundations – primarily Meta’s Llama models – which it fine-tunes and constrains using its proprietary data. The advantage, its chief scientist and head of enterprise AI Prem Natarajan told Euromoney, is “control, explainability and cost.” By using open models and layering safety and policy systems around them, the company believes it can customize without losing oversight.
This architecture – open foundations, private finetuning, agentic orchestration and tokenization for privacy – is what forms the bank’s emerging AI portfolio.
Chat Concierge, launched in early 2025, is the banking industry’s first agentic AI tool deployed in a live customer environment. It is marketed as powering intelligent conversations across car dealership websites and helping users compare vehicles, explore financing options, estimate trade-in values and schedule test drives.
Agent servicing brings gen AI to Capital One’s 20,000 customer service representatives. The assistant has reportedly improved average handling time and consistency in service quality while cutting repetitive search and scripting tasks. The design principle is “AI-assisted, not AI-replaced,” where each model recommendation is reviewed by a human before any customer-visible action.
Databolt, developed by Capital One Software, underpins these deployments with vaultless tokenization. It replaces personal data with reversible tokens, allowing AI systems to analyze sensitive information safely.
But what Capital One has not yet matched is ROI transparency. The Evident AI Index notes that while Capital One reports robust AI hiring, patent growth and published use cases, it does not yet disclose fully realized vs projected ROI across its AI deployments. Only JPMorgan, BNP Paribas and DBS Bank currently do.
