What happens when AI stops being cheap?

what-happens-when-ai-stops-being-cheap-helloyes.jpg
Ben Marshall
July 2026

“AI won’t replace designers, it will help them. AI won’t eliminate creative jobs, it will make creative people more productive. AI is just another tool. AI should be embraced, not feared. AI is inevitable.”

Whether those claims turn out to be true remains to be seen. Some may prove accurate, others may not. That’s not really what interests me… What interests me is something else entirely.

Why is so little attention seemingly being paid to the economics of the AI systems we’re rapidly becoming dependent on?

Every week seems to bring another AI-powered tool aimed squarely at creative professionals. Midjourney, Runway, Gamma, Cursor, Lovable, Firefly and countless others promise to help us work faster, generate more ideas, create more content and deliver more output. Presentation builders promise polished decks in minutes. Image generators can help designers churn out endless concepts and moodboards. Video tools allow production at a fraction of the usual cost and time. AI assistants have already appeared inside the software many of us already use, quietly (or irritatingly) becoming part of our day-to-day workflow.

The pitch is almost always some variation of the same idea: faster, easier, cheaper - and crucially, allowing us to use our creative time more efficiently! 

To be fair, many of these tools are genuinely impressive. The conversation that follows is usually focused on capability. What can the tool do? How much time can it save? How much faster can it make us? Alongside this runs another familiar discussion: that AI won’t replace creatives, but augment them. That designers who embrace AI will outperform those who don’t. That the future belongs to people who learn to work alongside these systems rather than compete against them.

I don’t argue that any of this may be false. What I see far less often is a discussion about sustainability. Not environmental sustainability (which is a separate conversation), but economic sustainability. What does this tool depend on to exist? And perhaps more importantly, what happens if the economics underpinning it begin to change?

Many of the AI products currently being integrated into creative workflows feel remarkably affordable. Some are free. Others cost little more than a streaming subscription. Increasingly, AI features are simply being bundled into software that agencies and designers are already paying for.

Viewed individually, none of these costs seem particularly significant, yet behind many of these products sit some of the largest infrastructure investments in corporate history.

A thing I find notable is how often discussions around AI focus on what the technology can do for us, and how rarely they focus on what the companies behind it expect in return.

Why are some of the world’s largest companies spending hundreds of billions of dollars building these systems? The answer is obvious: they believe there will eventually be a return. The question is, where does that return come from? Nobody invests billions of dollars in infrastructure because they’re hoping to break even. The expectation, sooner or later, is that those investments generate returns - that isn’t a criticism, it’s simply how investment works.

The bit I find most concerning is that many creative businesses appear to be building increasingly important parts of their workflow around technologies whose long-term economics remain uncertain.

Feed AI Laptop 02

Take a look at the wider industry. According to estimates compiled by Is AI Profitable Yet?, OpenAI has spent approximately $55 billion against around $28 billion in revenue. Microsoft is estimated to have spent roughly $266 billion against around $31 billion in revenue. Google has spent around $287 billion against approximately $25 billion in revenue. Amazon’s figures stand at roughly $313 billion against $22 billion in revenue, while Meta is estimated to have spent around $230 billion against approximately $3 billion in revenue (!).

The precise figures can be debated, but that’s not really the point - What matters is the scale. These are not the economics of a mature industry quietly generating steady profits, they’re the economics of a sector still investing enormous sums in pursuit of future returns. It’s the economics of a race.

Put bluntly: much of the AI industry is still trying to work out how to make money. That doesn’t mean it won’t - It may become one of the most profitable industries in history, but if that happens, it’s worth asking where those profits ultimately come from.

History is full of products and services that were initially subsidised, aggressively priced or funded through years of investment in pursuit of market share and future profitability (remember when Ubers were £5?). Investors fund growth, adoption accelerates, dependency develops and eventually the economics have to stand on their own. Why should AI be any different?

All this matters because AI is rapidly moving from experimentation to infrastructure. Many agencies are no longer testing these tools, they’re embedding them into everyday workflows. What starts as a useful image-generation tool becomes part of the concept development process. What begins as an AI copy assistant becomes part of content production. Presentation generators become part of client delivery. Video tools become part of production pipelines. Over time, convenience becomes infrastructure, and once something becomes infrastructure, dependency follows.

Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.

Frank Herbert
Dune (1965)

Creative agencies are generally very good at identifying risk elsewhere in their businesses - If a single client accounted for 70% of an agency’s revenue, most owners would recognise the danger immediately. The client might be wonderful, the work might be profitable, the relationship might have lasted years. None of that changes the fact that the business has become dependent on something it doesn’t control. The same principle applies to suppliers, platforms and revenue streams. Agencies diversify because concentration creates risk.

Yet when it comes to AI, many of the same agencies seem remarkably comfortable becoming increasingly dependent on platforms whose future pricing, terms and economics are entirely outside their control.

One dependency is recognised as risk, the other is often celebrated as innovation. This strikes me as a contradiction.

The question isn’t whether AI tools are useful, as many clearly are. The question is whether we’re paying enough attention to the dependencies we’re actively choosing to create.

If the AI tools your agency relies on become significantly more expensive in five years’ time, what happens then? Will your clients absorb the additional cost? Will your margins shrink? Could you switch providers easily? Or will those tools have become so deeply embedded in your workflow that changing course would be difficult, disruptive or commercially unviable?

I’m not suggesting agencies should avoid AI, and I’m certainly not suggesting the technology isn’t useful in many circumstances. I’m also not suggesting that every optimistic prediction about AI is wrong. My concern is much simpler:

Much of the conversation around AI assumes that today’s economics are permanent, which strikes me as overly optimistic.

The biggest question facing creative businesses may not be what AI can do today. It may be what happens tomorrow if the foundations we’ve built upon turn out to be more expensive, more concentrated, or more fragile than we expected. Because the question that keeps coming back to me is a simple one:

That’s all very interesting. But who’s paying for this, and what happens when they want their money back?

Read more
aviation-communication-brand-systems-helloyes.png
19.06.2026

Why are pilots so good at talking to each other?

There are few subjects I’ve been interested in for as long as aviation. I grew up around it, and at one point came surprisingly close to making it my own career.

Read more
london-underground-map-information-design-helloyes.jpg
26.06.2026

Why the most accurate map was the wrong one

If you’ve ever travelled on the London Underground, you’ve probably used one of the most famous pieces of graphic design ever created.

Read more