JPMorgan sets $2K monthly spending limits for Claude Code use

JPMorgan Chase is putting a price tag on how much AI its engineers can consume. The bank has implemented a $2,000 monthly spending cap on Anthropic’s Claude Code for select engineers, a move that signals even the most aggressive AI adopters in finance are hitting the brakes on unchecked usage.

The spending limit applies within JPMorgan’s internal secure development environment called Devspace, a fortified sandbox where engineers interact with AI coding tools under tighter security protocols.

The math behind the cap

JPMorgan required roughly 65,000 Global Technology employees to adopt approved AI coding tools as of March 2026, a mandate designed to boost productivity across the bank’s sprawling engineering operation.

Even if only a fraction of those engineers are subject to the Claude Code cap, the potential monthly spend is enormous. At $2,000 per head across just 10,000 engineers, that’s $20 million a month, or $240 million annually, on a single AI tool.

JPMorgan’s projected technology budget for 2026 sits at approximately $20 billion, with AI investments consuming a significant chunk.

Devspace and the security equation

The decision to route Claude Code through Devspace rather than letting engineers access it through standard channels reflects a security calculus that’s become standard at major financial institutions. Devspace appears to function as a controlled gateway, allowing JPMorgan to monitor what data flows to and from the AI model while maintaining compliance with internal security policies.

Goldman Sachs has taken similarly cautious steps in managing AI tool access across its engineering teams.

JPMorgan restricted access to Anthropic’s Claude models entirely for employees based in Hong Kong starting in June 2026, a decision driven by licensing and legal restrictions in the jurisdiction.