As AI deployment in capital markets accelerates, firms will have to use the right tools or risk being hit with a nasty bill.
Standard Chartered CEO Bill Winters apologised last week after saying AI could replace “lower value human capital”.
Winters’ thinking was that tech has improved to such an extent that investment in AI is becoming increasingly substitutable for employing people.
AI agents are popular — systems capable of reacting to their surroundings and taking real time decisions. But agents require significantly more computing power than pre-defined AI workflows.
It’s common with agentic systems to have layers of AI — one to control the project and others to execute tasks or for quality control. The costs quickly spiral.
Capital markets firms should be careful that their investments into AI are justified by the returns.
Anecdotally, some firms are spending hundreds of pounds per week per developer or analyst. A small sum compared to the cost of human staff, but not small enough to ignore.
Examples of AI costs getting out of control are emerging in other sectors. The Information reported that Uber had spent its entire 2026 AI budget in just four months.
The risk is compounded because the underlying cost of AI could start to rise.
Energy is a key input for using the computation to run AI, and prices have risen since the start of the war in Iran. Supply chains for other data centre components have also been disrupted.
Hyperscalers have committed vast capital to building digital infrastructure. They will expect some return. The cost of AI has been subsidised for firms and consumers during adoption, but once it is embedded into workflows, providers are in a stronger position to increase their prices.
In the longer term, there are reasons for optimism. Technological breakthroughs have brought down the cost of computation over time. The spike in energy prices may be temporary if the conflict between the US and Iran is resolved.
But for now, firms need to think about how to build resilience into their AI systems. Model selection is important. Not every task requires a cutting edge, compute-intensive model.
Google CEO Sundar Pichai made this argument in a blog post on the company’s website earlier this month. He was pitching Google’s new Gemini 3.5 Flash, which he claimed, “delivers frontier-level capabilities at less than half the price of comparable frontier models”.
“If companies used a mix of Flash and other frontier models they could save a lot of money,” Pichai wrote.
But firms don’t need to use Gemini 3.5 Flash to save money on AI. Some tasks don’t require large language models at all. Others can be accomplished using open-weight models, those where users can download the model and run it on their own computers, limiting the ability of one AI lab to dictate pricing.
Managing risk — and cost — means a thoughtful approach to AI deployment.