Bankers may love using Claude but the bills are beginning to reflect the AI industry’s financial reality

Rampant demand from white-collar types is, of course, a good problem for companies like Anthropic that can impose price rises that might start to justify the AI industry’s epic losses and hype-fuelled valuations. Dario Amodei’s company is on track for its first profitable quarter and is mulling a stock-market listing as early as October.

But the rapid adoption of AI agents, which can perform tasks independently, and their soaring expense create new quandaries for finance customers even as they can see the advantages. This adds to pressure on a firm’s profitability and is driving cost cuts in other parts of these businesses. It creates fears, too, of being handcuffed to particular AI companies as core IT skills shift to outside suppliers, usually American.

No wonder the boss of Standard Chartered touched a nerve with his ill-judged comments about “lower-value” human capital giving way to financial and investment capital as part of the bank’s AI push. It could be construed as salaried employees paying the price for the technology’s increasingly high costs.

As one analyst put it to the bank: “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?”

The behaviour of banks and other financial users is likely to evolve. Indeed, we may already be moving away from a ‘token-maxxing’ mindset—whereby heavy spending on AI processing power is a badge of honour regardless of whether it’s wasted—into a more mature phase.

I have been told of several examples where financial services are shifting to building inhouse models for tasks that do not need an all-knowing external LLM. “Not every task needs a frontier model,” says Christopher Tozzi, author of a history on open-source software.

That might encourage some surprise corporate thinking. Maybe ‘lower-value’ humans who can build and maintain a chatbot that is cheaper than an outside one will be found in possession of a valuable skill.

Finance firms might even do the unimaginable and club together, to better absorb AI costs from without and to share expertise on in-house models. This industry has never been good at sharing tech or data, with banks and asset managers usually trying to get one-up on their rivals. Nor has it been good at cross-border mergers that make sense but are politically difficult, such as UniCredit’s tilt at Commerzbank.

Maybe AI is the catalyst that will revolutionize capital allocation and finally consolidate the industry, especially in Europe’s over-banked parts. Perhaps UniCredit’s boss Andrea Orcel should show German Chancellor Friedrich Merz the banks’ expected AI budgets.

For now, few anticipate that Anthropic or OpenAI’s competitive moats will be crossed anytime soon. AI evangelists are probably right that the fear of missing out is so strong today that any chance of an edge against a competitor will keep the finance crowd spending. Tech expenditure by governments and companies is set to rise almost 8% in 2026, the biggest increase in years.

There are some parallels with cloud computing, where firms started building some stuff in-house without doing too much harm to the profits of Silicon Valley and Seattle. And, of course, AI is deepening Europe’s geopolitical dependence on US tech giants such as Amazon.

But having seen some users successfully creating their own models, reducing though not eliminating their spending burden, I do wonder if this time there will be more pressure to avoid outsourcing too much too quickly. Lost in the furore around Bill Winters’ AI comment was a more cautious line: “We are... being very, very thoughtful on the cost of AI.” If that encourages more European corporate self-help, so much the better. ©Bloomberg

The author is a Bloomberg Opinion columnist writing about the future of money and the future of Europe.