Wall Street just spent $20 billion on AI, and now analysts want receipts.
The Summary
- JPMorgan spends nearly $20 billion annually on technology, Bank of America $13 billion, and earnings calls are now battlegrounds over whether any of it is working
- Anthropic's Mythos model triggered an emergency Treasury meeting with Wall Street CEOs scrambling to assess cybersecurity implications
- The pressure isn't just about productivity anymore. It's about proving banks aren't getting disrupted by the tools they're buying.
The Signal
Wall Street's AI experiment just entered its accountability phase. During Q1 earnings, analysts stopped asking *if* banks are using AI and started demanding proof it's actually doing something. Bank of America CEO Brian Moynihan had to explain why his company is a beneficiary, not a victim. Citi got grilled on hiring philosophy "in this new AI regime." These aren't softball questions. They're the market asking: show me the money or show me the layoffs.
The timing matters. These earnings calls came one week after an impromptu Treasury meeting about Anthropic's Mythos model, which apparently scared banks, tech giants, and government officials enough to gather in the same room. The cybersecurity angle isn't theoretical anymore. When a model can potentially crack systems that banks built their entire security posture around, "billions invested in AI" suddenly looks less like innovation and more like exposure.
"JPMorgan Chase CEO Jamie Dimon defended his bank's spending, saying he is trying to keep the company from falling behind."
Here's what the numbers tell us:
- JPMorgan: $20 billion annual tech budget
- Bank of America: $13 billion in 2025
- Neither CEO could point to specific productivity gains or headcount reductions when pressed
The silence on metrics is the story. These institutions track everything. Average handle time. Loans per analyst. Errors per thousand transactions. If AI were moving the needle on efficiency, they'd lead with it. Instead, Dimon is talking about "not falling behind," which is defensive language. That's the stance you take when you're spending to avoid obsolescence, not because you've cracked the ROI equation.
Bloomberg reports banks are scrambling to understand what Mythos means for cybersecurity and the internet's future. Translation: the same AI capabilities banks are deploying internally could be weaponized externally. You're building agents to analyze portfolios while someone else is building agents to probe your perimeter. The Treasury meeting wasn't about innovation. It was about threat modeling.
The Implication
Watch the next two quarters. If banks can't show concrete wins (faster loan approvals, reduced compliance costs, measurably better fraud detection), the spending justification gets harder. The Mythos wildcard changes the calculation too. Banks might have to spend just to defend against AI, separate from any productivity play.
For anyone building in the agent economy: financial services is the test case. If institutions with unlimited budgets and desperate need can't make AI pencil out, that's a signal. If they *can* make it work, the playbook will be visible in headcount, product velocity, and margin expansion. Either way, we'll know by Q3.