Finance's AI Cost Problem Is Getting Real
Banks and asset managers across Europe are adopting artificial intelligence at scale, but a sharp problem is emerging: the bill is becoming unsustainable. Computing costs for AI tools are climbing fast, forcing finance firms to choose between expensive external models and building cheaper alternatives in-house.
Fund managers, bankers, and traders report growing AI use across their operations. The applications range from compiling analyst recommendations into automated rating systems to training chatbots for portfolio allocation ideas and writing code for quantitative traders. Yet the economics are shifting.
The Price Squeeze
Users of Claude, the AI assistant from Anthropic, are facing steep price increases. Even Anthropic's deal with SpaceX to boost processing capacity hasn't kept pace with demand for computation-heavy tasks.
Finance firms report costs climbing from tens of thousands of dollars to several million annually for a single organization. This spending surge mirrors complaints from the tech industry about the same vendor.
Anthropic is on track for its first profitable quarter and is considering a public listing as early as October. Higher prices help the company justify its losses and valuations, but they're creating pressure elsewhere in finance.
The Hidden Cost of Outsourcing
AI agents that perform tasks independently deliver real advantages. But their expense is driving cost cuts elsewhere in financial services. Firms also face a new risk: dependence on American AI suppliers as core IT skills migrate outside.
One analyst asked Anthropic directly: "The AI companies of today are not making any money and are spending a lot. Is it a problem that we don't really know how they'll be charging in a year or two from now?"
Standard Chartered's chief executive touched this nerve when he discussed "lower-value" human capital being replaced by financial and investment capital as part of the bank's AI push. The comment sparked backlash because it suggested salaried employees would bear the cost of expensive technology.
A Shift Toward In-House Models
Finance firms are moving away from what some call "token-maxxing" - spending heavily on AI processing regardless of efficiency - toward more disciplined spending. Several financial services firms are now building in-house models for tasks that don't require a state-of-the-art external large language model.
This approach has real value. Cheaper, custom chatbots built internally may cost less than external options. Workers who can build and maintain these systems become valuable, not "lower-value" as some executives suggested.
Finance has historically guarded technology and data closely, with banks and asset managers competing rather than collaborating. But AI costs might change that calculus. Industry consolidation, especially in Europe's overbanked regions, could help firms absorb costs and share expertise on in-house models.
The Bigger Picture
Few expect Anthropic or OpenAI to lose their competitive edge soon. Fear of missing out remains strong enough to keep finance spending. Tech spending across governments and companies is projected to rise nearly 8% in 2026, the largest increase in years.
The pattern resembles cloud computing adoption, where companies built some systems in-house without meaningfully hurting Silicon Valley's profits. Yet AI for finance deepens Europe's dependence on American tech giants like Amazon.
Some users have successfully created their own models, reducing but not eliminating spending. That experience suggests finance may face more pressure this time to avoid outsourcing too quickly. The more cautious voices in banking - those saying they're being "very, very thoughtful on the cost of AI" - may point the way forward.
Your membership also unlocks:
