The AI industry is learning a lesson media owners already know
Opinion
AI models are commoditising in real time, at enormous cost. Media owners have seen this film before, and they’re about to star in the sequel, writes Omar Oakes.
AI will replace humans because they can do many tasks better, faster and more cheaply.”
You’ve likely heard some version of this claim many times in recent years; sadly, without challenge, it’s been accepted as common sense. Indeed, this conventional wisdom has encouraged the “frontier AI labs” OpenAI (ChatGPT) and Anthropic (Claude) to prepare to go public on massive valuations, while recent earnings are revealing how Meta, Google and Amazon’s investment in AI has been even more gargantuan than investors had thought.
That’s the thing about a gold rush: it’s about gambling more than speed. You are betting that moving quickly in an unsafe or risky fashion is better than moving cautiously and safely, such is the urgency of the opportunity at hand.
Which is why using a huge amount of AI became a badge of honour. Amazon built an internal leaderboard, KiroRank, to rank engineering teams by the number of tokens they burned. Meta built its own version, reportedly called Claudeonomics. The more you spent, the more innovative you looked.
Then the bill arrived.
Uber burned through its entire 2026 AI budget in four months, after handing thousands of engineers unrestricted access to AI coding tools. Its COO went on a podcast and admitted the company couldn’t actually draw a line between the spend and anything customers noticed.
The leaderboard, meanwhile, was quietly taken down after staff started burning tokens on nothing in particular, just to climb the rankings.
Oopsie.
The great model shop-around
Here’s what’s actually happening, and it should feel familiar to anyone who sells media for a living: buyers are shopping the same job across multiple suppliers, in real time, based almost entirely on price. Not quality.
The Wall Street Journal reported last week that companies are routinely running the same task on several AI models and picking whichever is cheapest that clears the bar.
Cursor, the AI coding startup, tested building a web browser from scratch on OpenAI’s flagship model: a little over $10,000. The same job, split across a cheaper coding model and Anthropic’s Opus: $1,339.
Marty Kausas, CEO of AI platform Pylon, put it more bluntly to the Journal: “There’s zero loyalty that I’m seeing. It really feels like a bloodbath right now.”
His company has picked up more than $1m in free tokens this year alone, doled out by AI vendors terrified of losing accounts. Because every supplier knows the customer is one click from somewhere cheaper.
Sound familiar? It should. It’s the exact mechanism that has hollowed out CPM as a meaningful metric: buyers treating a differentiated product as an interchangeable commodity because, somewhere, there is always a cheaper unit that will technically do the job.
The industry that’s still celebrating the wrong thing
This tension was visible at Cannes Lions just weeks ago: an ad industry still in its investment-euphoria phase about AI, but realising the gains won’t come cheaply.
Havas’ press conference was almost a perfect example. It touted €1m in new funding for three separate AI ventures: Vermeer, its creative production platform; Ava, its internal model-access layer; and Verve Labs, a start-up building AI-driven “people models” of consumer behaviour. The room applauded.
But then Dan Hagen, Havas’s chief data and technology officer, warned “the cost of AI is going to go up significantly” – pointing to the upcoming OpenAI and Anthropic IPOs as the trigger. All those public investors will soon demand that they deliver the wild revenues they continually promise, so prices of AI tokens must surely go up.
Hagen cited BCG research that, among companies actually realising ROI from AI transformation today, 70% of that value comes from people and change management, 20% from the technology stack, and just 10% from the specific models used.
So the value comes from people using AI, not the AI itself? And it’s getting more expensive? Hmm. Wasn’t this hyper-efficient tech supposed to get rid of all these expensive humans?
Zenith’s UK CEO Sannah Rogers was asked this very question on a panel with Bauer Media that same day.
“I think people who think that you can deliver just more for less with AI, they fundamentally misunderstand the technology and the cost that comes with it,” Rogers said.
Uh oh.
What media owners are about to get wrong, again
If you sell media for a living, you’ve seen this film before, and you already know how it ends.
Digital infrastructure – the programmatic pipes, first-party data platforms, “AI-powered” contextual targeting – was supposed to be the thing that let media owners compete with platforms on the platforms’ own terms. Instead, each of those capabilities became table stakes within a couple of product cycles, commoditised by scale players who could build them cheaper and faster.
Meanwhile, the actual value – the audience relationship, the addressable data, the margin – came from somewhere else.
AI is running that exact playbook, except the commoditisation that took a decade to consume digital media is compressing into 18 months.
This is only going to get worse. Cheaper Chinese models – DeepSeek, Kimi, models from Z.AI and Moonshot – are already considered ‘good enough’ for most tasks, and companies are routing work to whichever model is cheapest for that specific job.
So when a media owner leads a client pitch with “we do AI-powered targeting” or “our AI tools give you an edge,” any buyer who’s been paying attention should ask: for how long?
Every AI capability you can buy off the shelf, your competitor can buy too, at a lower price, next quarter.
The thing that can’t be commoditised
None of this is an argument against using AI.
It’s a warning about building your business on AI hype, which goes criminally unchallenged by investors, business leaders, and, I’m sorry to say, business journalists.
The capability itself was never going to be the differentiator; not for Havas, Uber, nor a mid-sized media owner selling against Google. If big companies, the ones that can negotiate proprietary deals with AI providers, are worried about spiralling costs, what hope is there for the many boutique consultancies whose founders, I fear, have bet big on “agentic” solutions?
The true value is always the thing sitting underneath the technology: the editorial judgement about what’s worth making, the trust an audience places in verified human output, the relationship a sales team has built over years that a cheaper competitor can’t simply undercut into existence.
That’s the asset media owners have been chronically underpricing since before generative AI existed.
It’s also the one thing in this entire story that hasn’t got cheaper or faster to replicate. Good.
Omar Oakes was the founding editor of The Media Leader and continues to write a column as a freelance journalist and communications consultant for advertising and media companies. He has reported on advertising and media for 10 years.