Major financial institutions, including JPMorgan Chase, Citigroup, and CIBC, are implementing stricter controls on AI-related expenditures. The shift toward usage-based billing models, which charge based on compute and inference rather than fixed subscriptions, has made technology budgets increasingly difficult to forecast for bank executives.
JPMorgan Chase reports annual AI-related costs of approximately $700 million. The bank has introduced a monthly spending limit of $2,000 per engineer for premium models like Anthropic’s Claude Code. Similarly, CIBC has restricted access to premium tools, noting that costs can reach CAD 3,000 per employee each month.
Banks are adopting strategies to manage these rising costs, such as routing simpler tasks to less expensive models and limiting access to high-end AI tools. These measures come as the industry accelerates the deployment of agentic AI, with banks announcing 19 new use cases in the first quarter of 2026, compared to eight in the previous quarter.
Despite the rising costs, some banks report efficiency gains. Citi attributes a 30–40% improvement in product launch speed to AI and a 30% year-over-year reduction in vendor spending. BBVA demonstrated significant optimization by reducing the cost of a loan approval process from €5 to €0.03 through efficient AI implementation.
