
Most U.S. banks — 80% in fact — increased their artificial intelligence spending for 2025, according to American Banker's
In a survey fielded earlier this year, 11% of bankers said they expect to make a significant increase (25% or more) in this technology category; 25% plan a moderate increase (10% to 24%); 44% project a slight increase (less than 10%); 7% expect no change and 13% intend to decrease AI spending.
This is good news for AI software providers. Experts say banks are focusing this spending in a few key areas this year.
Enterprise-grade generative AI models
A major area of bank tech investment is in enterprise licenses for generative AI, such as corporate versions of Anthropic's Claude or OpenAI's GPT-4, according to Michael Abbott, global banking lead at Accenture. Corporate licenses come with security and safety that are not a part of consumer-facing versions of generative AI like ChatGPT, he noted.
JPMorganChase, for instance, provides employees access to several popular large language models through its
"We built the platform using our own engineers and AI practitioners, so the only variable cost is compute," Chief Analytics Officer Derek Waldron told American Banker. "That means that if you don't use it, you don't pay."
"That value proposition turned out to be very, very desirable to business leaders, because then they know that it's only they're only paying for what they're getting value from," Waldron said. The bank does not disclose how much it spent on developing LLM Suite and on AI in general. Its overall 2025 technology budget is $18 billion.
Embedded AI and AI applications
Software with AI embedded in it is a top area of investment for banks, according to Abbott. An example could be Salesforce with its baked-in Einstein Copilot, a conversational AI assistant for Salesforce CRM, designed to augment employees and help them complete tasks more efficiently.
Banks are investing in AI for fraud prevention and mitigation, cyber security breach alerts, and operational risk alerts, said Celent analyst Alenka Grealish.
They're also spending on generative AI for code development, she said.
"That spans everything from refactoring to code writing and debugging," Grealish said. "Those have been prime spots." Risk-oriented applications like credit portfolio monitoring will probably grow with economic uncertainty, she said.
"There's always a rise in fraud as well as credit losses if the economy gets shaky," Grealish said. "Already, we're seeing consumer confidence, small business confidence decline. So I think those areas will receive a disproportionate share of investment."
The need to better engage with customers will lead to investment in natural language processing, personalization and agent support, which all involve AI, she said. Generative AI is already helping front office employees with intent recognition, sourcing information, references, useful links, and a greater ability to answer questions, including complex questions.
Bank of America's Erica is a case in point — it uses natural language understanding to translate customer's questions into queries a computer can understand and find an answer for. The bank has said it's spending $4 billion on AI and new tech initiatives this year.
Data management, data governance
Banks are just starting to invest in data management software, Grealish said.
"Before, data just sat in a database," she said. "It wasn't asked to perform in an AI predictive model, or a generative AI employee-facing AI assistant." Now, it has to be able to read structured data, like information in databases, and unstructured data, like PDFs.
"It's an awakening of, oh, we have to spend more here to get up to speed," Grealish said.
A lot of the key data required for AI solutions hasn't been prioritized in the past, Brian Gibbons, principal at EY, told American Banker.
"It's not the same data you use to calculate your liquidity capital and do regulatory reporting," he said. "It's different types of data that maybe haven't attracted the same level of investment. So we're seeing continued investment in data capabilities."
Data platform providers Databricks and Snowflake, for instance, have been getting a lot of investment.
Abbott said he's "seeing a cold war for data, because everyone wants to have the customer's data so they can run their AI agents on it and charge for it. But the banks are pushing back, saying, 'I don't want my data everywhere, I want to own my data.'"
AI model governance and testing
A related AI expense is model testing and governance.
"Testing AI solutions is very different than testing other pieces of software," Gibbons said. "A hot topic for us in talking to clients is around building a different type of testing process and set of tools for AI solutions that addresses governance needs, but also makes sure you're churning out effective products. We know there's investment in that space."
Pre-AI modernization projects
Some of the AI investment is going into core banking upgrades and other modernization efforts, Gibbons said.
AI requires modern technology infrastructure, data, models, governance and more, Grealish said.
"Those are the key building blocks," she said. "You can't have AI lay on top of a fragile infrastructure. It needs APIs. It needs cleansed data. It needs governance around the data. It needs explainability, et cetera. And that does require this long journey for many banks, to go from legacy core, legacy data, data that's not really consumable by models, because it was meant just to record a transaction."
Bespoke software-hardware combos
Some large banks are spending on private AI supercomputers, for instance BNY's purchase of an Nvidia DGX SuperPOD last year. The bank said it
LOOKING FOR MORE?