You can’t make this stuff up (but we’ll try anyway): The economic viability of AI

Let’s start with a headline that sounds like satire: Sam Altman, OpenAI’s charismatic charlatan of a CEO, is apparently asking the US government for a trillion-dollar loan to bankroll his company’s AI ambitions. Yes, that’s “trillion” with a “t”.

According to recent reporting in the FT and Sifted, OpenAI has entered computing infrastructure deals worth roughly a trillion dollars, including a $300 billion partnership with Oracle and a $500 billion “Stargate” project with SoftBank. The problem? Projected revenues this year are in the tens of billions. Which means if you need a trillion to make ten billion, your logic might need to have a wee re-think.

Of course, we’re told that high finance is complicated. You have to spend money to make money. Scale, volume, synergy, blah blah blah. But here’s the thing, you can’t make up in volume what you lose on every unit. If you lose a pound on each widget you sell, selling a billion of them doesn’t make you a genius, it makes you broke and in the hole for a billion quid!

And yet, here we are, in the middle of what might be the most over-financed technology boom in history. Around 95% of the recent stock market rise is said to be attributable to AI investments, and infrastructure firms like NVIDIA and Cisco are raking it in. At least they sell real products that real people use for real things, unlike some in this space who are currently selling a dream wrapped in a press release. But the central issue isn’t hype or hubris. It’s economics. The story we keep hearing is that AI, particularly large language models, will replace swathes of human workers and drive the greatest societal transformation in history. The catch? That transformation only works if the numbers add up.

Leaving aside the question of what AI can actually do as well as (or better than) humans, there’s a more basic question: can it do it more cheaply? Because if not, it’s a non-starter. There’s no doubt that building AI into software products makes sense. Productivity tools, cybersecurity platforms, and network management systems can all be turbo-charged by machine learning. In those cases, AI can add genuine value, like spotting patterns, accelerating analysis, or generating ideas at scale. But when the infrastructure and energy costs of replacing humans end up being three or five times higher than simply keeping the humans, that transformation suddenly looks more like a mirage than a revolution.

And that’s where the UK story gets interesting. The press and the government would say the UK’s AI sector is booming. Numbers don’t lie right? More than 5,800 companies, nearly £24 billion in annual revenue, and over £15 billion in inward investment last year alone. Microsoft has pledged £22 billion into UK infrastructure, while Google’s DeepMind is putting another £5 billion into data centres (all from the official UK Gov AI sector study). On paper, this should be a golden age for British AI.

But let’s not lose our footing on the hype ladder. AI infrastructure is eye-wateringly expensive, not just the chips, but the data centres, the energy, the cooling, the people to run it all. The UK remains heavily reliant on imported hardware and cloud capacity. The economic return from most deployed AI systems is, at best, uneven. And if those infrastructure costs keep rising faster than the revenue generated by AI-enabled products, the whole model could start to wobble.

To make AI economically viable, the industry needs to focus on cost discipline, measurable ROI, and a credible path to profitability. These are certainly the questions I’ve been asking of the AI businesses I’ve been seeing of late. Governments and investors alike will demand evidence of real returns, not just moon-shot press releases. Energy and infrastructure costs will also decide who stays standing; if your data centre burns through more electricity than a small city, your cost base might eat your margin before your model finishes training.

In the UK, the challenge is particularly acute. We’re betting heavily on AI as a cornerstone of our national prosperity, national security and strategic autonomy. But the test will be whether that bet delivers value in pounds and pence, not just headlines. AI will undoubtedly transform industries, but it must do so within the laws of economic gravity.

Which brings us back to Altman’s trillion-dollar ask. The request itself is breath-taking, but it’s also telling. It reflects the uneasy truth that for all AI’s promise, the economics are not yet settled. If it costs more to build an “intelligent” system than it can ever hope to earn back, then the logic doesn’t just falter, it collapses.

As former US Senator Jeff Flake once said, “Clinging to a logic that defies economic reality is clinging to failure, guaranteed.” You can make money in a bubble, at the beginning. When you toss a ball, it goes up…at first. But economic reality, like gravity, always has the last word.

And if you really do need a trillion pounds to make ten billion, it might be time to check your maths, or your business model.

12, November 2025  |  Ben Addley