UK BUSINESSES formally integrating AI agents into their systems and automating key workflows are sleepwalking into a financial black hole due to the growing problem of ‘token burn’, experts have warned.
Currently, many AI agents are being subsidised by gargantuan venture capital investments, but one industry insider has cautioned that “a few £20 AI subscriptions could become £2,000 business black holes once the subsidy smoke clears”.
Another said: “Every impressive autonomous workflow still has a meter running underneath it, and many users have not yet seen the real bill because venture money is softening the landing.”
An exponentially growing number of UK businesses in all sectors are integrating AI into their systems in an effort to streamline their operations and potentially reduce headcount.
But the vast majority, according to specialists, are unaware of the underlying economics powering the world’s leading AI firms and the growing issue of ‘token burn’ — and have not properly modelled how it will impact their cost base and bottom line.
Token burn
Every time someone uses AI agents, whether spontaneously or as part of an automated workflow, they are burning tokens. Each word inputted, each punctuation mark and even the space between words is a token, and that token costs money, as it involves energy and compute.
So a prompt or series of prompts, a reply, an uploaded file, a generated image or video, dataset analysis or automated flow, is burning huge volumes of tokens.
Essentially, a token is the fundamental computational resource that powers AI agents, specifically the massive GPU (graphics processing unit) and electricity costs required to process inputs and generate outputs.
To date, experts say, the true cost of these tokens has been hidden due to venture capital money subsidising a still nascent industry in an effort to get users “hooked”.
Reflecting this, people and businesses are, on the whole, paying fixed monthly subs that are concealing the actual raw compute and energy costs.
But at some point, they say, investors will want to see a return on their money, which could see subscription costs skyrocket as AI agents have no choice but to firstly cover their costs and secondly target profitability. That’s when the AI dream has the potential to turn into an operational nightmare.
Compute costs
Colette Mason, Author & AI Consultant at London-based Clever Clogs AI, gave an insight into the true economics of AI. She also warned that even if businesses are prepared to cover the true cost of tokens, which is the AI industry’s elephant in the room, the underlying infrastructure may not be there anyway.
She said: “A £160 AI subscription can consume £3,500 in actual compute, which is a 25x subsidy. Raw electricity alone runs 15% to 20% of API [usage-based] pricing, meaning some users cost providers more in power than they pay in fees. And that’s before GPUs, training or staff are counted.
“That subsidy is ending. For example, Microsoft’s GitHub Copilot paused all individual signups on April 20th, not to increase revenue but because they ran out of capacity.
“Microsoft is shifting to token-based billing because charging per message made as much sense as a supermarket charging per item in your trolley whether you picked up crisps or a widescreen TV. When tokens cost 25x as much, will business owners still want AI embedded everywhere?
“Another problem with the AI growth story is that, even if every business paid the full price for tokens, there are not enough chips to serve them. Microsoft, Google, Meta and Amazon have forward-ordered most of Nvidia’s allocation through 2027, crowding out everyone else.”
Margin burn
Rohit Parmar-Mistry, Founder at Burton-on-Trent-based Pattrn Data, said ‘token burn’ risks becoming ‘margin burn’ for businesses.
He added: “Token burn is the bit of the AI agent story that gets hidden behind the demo. Every impressive autonomous workflow still has a meter running underneath it, and many users have not yet seen the real bill because venture money is softening the landing.
“That matters for businesses and investors because an agent that looks cheap in a subsidised market may become expensive once usage scales and pricing normalises. The risk is not just technical failure. It is building operations around a cost base nobody has properly modelled.
“The sensible question is not whether agents are useful. Some clearly will be. It is whether the value of each task is greater than the compute, supervision and error-management cost behind it. If that maths is vague, token burn becomes margin burn.”
Black hole
Mitali Deypurkaystha, AI Strategist & Author at Newcastle upon Tyne-based Impact Icon AI, cautioned that many businesses are unaware they are swapping wage bills for meter bills.
She said: “A few £20 AI subscriptions could become £2,000 business black holes once the subsidy smoke clears. Right now, firms are treating AI like NHS dentistry: manageable while the real costs are heavily cushioned, but a nasty shock when the true bill finally lands.
“Businesses may think they are cutting wage bills, but many are quietly swapping them for meter bills that could rocket with every click, prompt and piece of AI slop once the subsidies disappear.
“Investors are betting firms will become so hooked on AI that ripping it out feels even more painful than paying the soaring costs. The danger is that many businesses may already be walking into a trap without realising how expensive the escape route could become.”
Trapped
Another AI industry observer, Kanchan Satyal of #123Sudo, warned the problem gets worse the more seriously businesses are using AI and embedding it into their workflows and wider models.
She said: “If you are using AI casually, you might never feel it in a meaningful way. But if you are using AI daily across real work writing, strategy, operations, research, support, planning, documentation and ideation, the cost stacks very quickly. This is where teams get trapped.
“AI feels cheap at the start because they measure it in moments. But the real cost shows up in systems. A few prompts are cheap. A few hundred workflows are not. A whole team using AI all day inside messy workflows? That is where the burn becomes serious.”
It’s the serious nature of token usage, or ‘burn’, that makes it a governance problem, according to Katrina Young, Chief Technology Officer and Digital Transformation Strategist at KYC Digital.
She said: “Token burn is less a procurement problem and more a governance problem. Most organisations deploying AI agents still do not fully understand what automated workflows cost to run at scale.
“Tokens are consumed operationally while budgets are still being viewed through flat subscription thinking. That disconnect is where financial surprises become operational problems.
“The real question is not whether AI is useful but whether businesses have properly modelled the long-term infrastructure economics of embedding it into everyday operations. Many still have not.”
