Insights on strategy, websites, UX, and digital growth

A reckoning around AI is coming, but it’s probably not what you might imagine.

Most discussions about where AI will take the human race eventually end up in the same place. Machines replacing humans. Entire professions disappearing overnight. Some kind of slow-motion collapse where algorithms gradually assume control while the rest of us become economically irrelevant.

Somewhere between Terminator and 1984, or a plethora of other dystopian sci-fi visions.

Perhaps that future arrives one day. Perhaps it does not.

But there is no doubt another reckoning coming first, and it is already beginning to show itself. Not in science fiction, but in something much less fantastical and far more grounded, if not a little mundane:

• Balance sheets
• Infrastructure costs
• Pricing models
• Operational reality

Just as a sidebar at this point, let me be clear: we are not sceptics of AI. Quite the opposite. We use it ourselves. We experiment constantly with low-code workflows, automation and tooling that can accelerate delivery and remove repetitive work. Like most agencies operating seriously in this space, we would be foolish not to.

So while not AI sceptics, we would consider ourselves to AI realists, and there is a good reason why for that position. There is a meaningful difference between using AI to support skilled people and believing AI removes the need for them altogether. Increasingly, those two ideas are being treated as though they are the same thing, and they are not.

Here’s why…

AI is useful, but businesses could be confusing “useful” with “sustainable”

What concerns us is not AI itself. It is the growing assumption that AI somehow changes the economic fundamentals of digital work permanently. That businesses no longer need developers, agencies or experienced technical people because software can simply be prompted into existence at high quality and low cost.

For certain small tasks, that can appear true. You can generate snippets of code, spin up rough prototypes or automate repetitive admin work remarkably quickly. Some of the gains are very real. But a great deal of what the market currently perceives as “cheap AI” is being distorted by economics that are nowhere near settled.

From conversations we are having and behaviours we are seeing, it appears many organisations do not seem to fully comprehend the financial model that sits underneath the AI products they are buying. The majority of companies promoting “AI-powered” platforms are not building their own proprietary frontier AI models themselves. They are connecting into somebody else’s model, usually through APIs provided by companies such as OpenAI, Anthropic or Google, then layering interfaces, prompts and workflows on top.

That in itself is probably not a surprise to most. Nor is this model inherently dishonest. There is real value in workflow design, integration and usability. But it does mean the underlying intelligence is rented, not owned. And rented intelligence comes with a meter attached to it.

What AI tokens actually are and why they matter

That meter is measured in tokens. Every prompt, every response, every generated image, every coding request and every automated task consumes compute resources.

What appears on the surface as a frictionless conversation with an AI assistant is, underneath, an extraordinarily resource-intensive process involving huge data centres, specialist chips, energy consumption, cooling infrastructure and massive capital expenditure.

This is a major part of the financial equation that many businesses still have not fully absorbed. AI currently feels cheap because, in many cases, it is being heavily subsidised by infrastructure investment and aggressive market expansion.

The reality is tokens are currently not being charged at anywhere close to their actual cost. They are purely a lost leader in an attempt to grab market share.

The economics of AI are still unresolved

The numbers involved in building AI models and their supporting operating costs are so staggering it is easy for the human brain to gloss over them.

Reuters reported that OpenAI surpassed $20 billion in annualised revenue in 2025, while simultaneously facing computing expenditure measured of $50 billion in 2026, with over $1.4 trillion of infrastructure investment in the next few years. Reuters also reported extraordinary infrastructure commitments from Anthropic as the industry races to secure cloud capacity and compute power.

Meanwhile, the wider infrastructure race continues to escalate. Combined AI investment by Amazon, Microsoft, Alphabet and Meta is reportedly approaching $700 billion.

None of this means AI has no future. Far from it. But it does suggest that the current economics are not stable in the way many businesses assume they are.

Eventually, the real bill is going to arrive.

Why AI pricing models are already changing

We are already starting to see the first signs of that transition. GitHub Copilot recently began moving towards usage-based billing because the previous model had become, in GitHub’s own words, “no longer sustainable”.

That should matter to businesses more than it currently seems to. Many organisations are building operational dependencies around AI tooling while assuming today’s pricing structures remain broadly unchanged forever.

The “free” magic box is about to become an expensive magic box. And much like a magic box, a lot of the magic is just illusion and sleight of hand…more on that later.

Why questioning AI economics does not make you a Luddite

Sometimes, criticism of AI economics is often dismissed with the accusation that anyone questioning the movement is simply a Luddite resisting inevitable technological progress. It is a lazy comparison and historically inaccurate. Under scrutiny, it simply doesn’t hold water.

The Luddites were not irrational people terrified of innovation. They were skilled workers responding to industrial mechanisation that threatened their livelihoods and bargaining power. Sadly, that was not enough to protect their livelihoods.

The Industrial Revolution ultimately worked and changed the world because the economics behind it were stable. Mechanisation produced greater output at predictable and commercially sustainable cost.

That is the critical distinction.

AI may well deliver long-term productivity gains on a similar scale. But the economics underpinning are not the same. The market is currently operating inside an environment of subsidised access, infrastructure races and evolving pricing models.

This is different from the economics of the industrial revolution. To illustrate:

Had the cost of operating industrial machinery suddenly increased tenfold a few years after a factory owner mechanised production, the consequences would have been severe. Prices would have risen dramatically, margins would have collapsed, or workers would have been rehired.