Big banks say AI is paying off, and their software vendors may foot the bill
JPMorgan, Citi and Wells Fargo opened the third quarter season with very different stories about artificial intelligence. The common thread is cost, and it points at the suppliers.
Three of America's largest banks reported third quarter results on 13 October, and artificial intelligence ran through all three earnings days without once appearing as a line in the accounts. JPMorgan Chase said the benefits of its AI programme are not yet visible in its quarterly earnings. Citigroup said its gains from AI and automation have already outrun what it has spent on them. Wells Fargo cut its headcount by a further 8% over twelve months and credited efficiency work for lower costs.
The differences in tone matter less than what the three banks share. Each is treating AI as a cost discipline rather than a revenue story, each wants more of the work done in-house, and each is becoming a harder customer for the fintech and enterprise software companies that sell to it. For startups whose business plan depends on large bank contracts, the season opened with a clear message about who is expected to absorb the cost of the transition.
JPMorgan: large ambitions, modest numbers
JPMorgan earned $15.3bn in the quarter, or $5.57 per share, on managed revenue of $50.9bn, up 8% from a year earlier. Excluding a securities item, net income was $15.8bn. Expenses rose 12% to $27.3bn, with technology investment among the drivers, and the bank guided to roughly $27.5bn for the fourth quarter. Chief financial officer Jeremy Barnum told analysts that the market's consensus of around $114bn for 2027 expenses was not far from the bank's preliminary view.
On AI, the most revealing exchange came when Mike Mayo of Wells Fargo said he could not find the benefits in the earnings release. Barnum replied that the AI expense figure circulating in the market overstated what the bank actually spends, and that "the benefits aren't observable in our current quarterly earnings." He said he was moving away from putting a number on the gains, pointed to software engineering as the area where generative and agentic tools are changing productivity, and warned that some of the savings would be offset by token costs. Crypto Briefing reported that Barnum had also said the spending "has not spiraled".
The bank's response to token costs is telling. Barnum described a policy of model agnosticism, using proprietary and open-weight models, some of them running in JPMorgan's own data centres. Jamie Dimon added that model routing tools were coming, observing that the bank does not need the most expensive model for every query. That is the language of a buyer intent on commoditising its suppliers.
JPMorgan's scale explains why it can talk this way. According to Crypto Briefing, the bank runs a technology budget of about $20bn a year, has more than 1,000 AI use cases in production and sees around 150,000 employees use its internal language models every week. Dimon also cautioned against outside agents. Asked about consumer AI agents that could move deposits in search of yield, he said the bank would not let just any agent into its systems, citing the risk of rogue agents. Fintechs building agentic payment and savings tools should read that as a statement about access to bank data, not merely about security.
Citi: the savings are already larger than the spend
Citigroup told a more confident story. Net income rose to $5.2bn, or $2.85 per share, on revenue up 8% to $23.8bn, the best third quarter revenue in a decade according to chief executive Jane Fraser. Operating expenses fell 2% to $14.0bn as reported, or rose 3% excluding a goodwill impairment booked a year earlier, and the efficiency ratio improved to 58.5% from 64.7%. Fraser wrote in the results release that the bank was continuing to invest in growth "and prepare for an AI-driven future," and said Citi was pulling forward investments toward its return target of 11% to 13%. Severance added to expenses in several divisions.
On the call, as reported by Benzinga, Fraser said the benefits of AI and automation had significantly exceeded the related investments so far this year. The supporting figures were specific. About 14,000 engineers use Devin, the agentic coding tool Citi has adopted as its main platform. AI has helped the bank launch new offerings 30% to 40% faster. Around 100 manual processes are being redesigned with AI and automation, and AI-powered customer service tools now operate in 82 countries, improving client response times by 21%.
The number that should concern vendors is a different one. Citi said spending on technology vendors had fallen 30% from a year earlier as it brings more work in-house. Fraser also said she saw no slowdown in investment across the wider AI ecosystem, arguing that "there's simply not enough compute." Citi is therefore bullish on AI as an industry while spending less with the outside firms that helped it build its technology estate.
Wells Fargo: efficiency by subtraction
Wells Fargo reported net income of $6.0bn, or $1.87 per share, up 13% on a per share basis, on revenue of $21.9bn, up 2%. Noninterest expense fell 1% to $13.7bn, and the bank attributed the decline mainly to efficiency initiatives, including the 8% cut in headcount and lower lease costs, along with lower severance than a year earlier. Technology and equipment expense rose.
Chief executive Charlie Scharf framed the quarter around self-funding: "Expense savings have funded much of the investments that are making the company stronger." He cited "the investment in AI infrastructure" as one of the positive forces in the American economy that outweigh risks from rates, oil prices and inflation. Wells Fargo did not attribute its headcount reduction to AI, and its efficiency ratio of 63% remains well above Citi's. Still, the pattern is the same: costs held flat or lower while technology spending rises inside the envelope.
What the calls mean for fintech and AI suppliers
Taken together, the three calls describe a market in which the largest buyers of financial technology have concluded that AI is a tool for doing more with their own engineers. Citi's 30% reduction in vendor spending and JPMorgan's insistence on model agnosticism point in the same direction. Banks want to own the orchestration layer, rent models as cheaply as possible and route work between them. That leaves less room for the middleware and point solutions that many fintech startups have sold to large institutions over the past decade.
There are exceptions. Tools that become the standard inside a bank, as Devin appears to have done at Citi, can win very large seat counts. Security, identity and compliance products benefit from Dimon's caution about outside agents. And the compute shortage Fraser described keeps capital flowing toward infrastructure. But the procurement mood has changed: the banks are measuring AI by expense lines and headcount, and they expect suppliers to show savings that survive a chief financial officer's scrutiny.
Our view
The divergence between JPMorgan and Citi is less a disagreement about AI than about disclosure. Citi has a return target to defend and every reason to show that technology is lowering its cost base; JPMorgan, with the stronger franchise, can afford to say the benefits are hard to see. Both are building internal capability and both are pushing token and vendor costs down.
For founders selling into banks, that is the central fact of this earnings season. The pitch that worked a few years ago, a clever AI feature priced per seat, now meets buyers that run their own models and count every outside contract. The winners will be companies that become infrastructure inside the bank, or that solve problems the bank does not want to own, such as fraud, identity and agent security. Investors should watch whether other large banks echo Citi's vendor numbers in the coming days, and whether JPMorgan ever chooses to quantify its returns. If the largest buyers keep in-sourcing, valuations for bank-focused AI software will have to adjust to a narrower market.
Sources
- Webull (Benzinga): Transcript: JPMorgan Chase Q3 2026 Earnings Conference Call
- Crypto Briefing: JPMorgan CFO says the bank's AI exposure remains modest as sentiment shifts
- Pulse 2.0: JPMorgan Chase reports $15.3 billion Q3 profit as revenue rises 8%
- SEC: Citigroup Inc. third quarter 2026 results, Form 8-K exhibit 99.1
- Webull (Benzinga): Citigroup CEO says AI spending isn't slowing down despite higher interest rates
- SEC: Wells Fargo & Company third quarter 2026 earnings release, Form 8-K exhibit 99.1
Filed under AI. Written by Invest Times Editorial under our editorial standards. Commentary and analysis for information only, not investment advice.
