Is The AI Curve Actually Cooling?

AI has rarely left the headlines over the last two years. At The Interline we’ve written two dedicated reports on the subject, and every week our news desk fills with so many AI flavoured stories that it’s become a routine challenge to find stories to analyse that aren’t about AI, because the level of investment, speculation, storytelling, and hype have essentially sucked all the other air out of the room.

But being ubiquitous is not the same as being universally accepted, and part of the reason that AI topics have become so popular is that both the underlying technology and the intended use cases are contentious. There are strong arguments being made for the case that AI (generative AI specifically) represents the next wave of consumer and enterprise technology, on a scale equal to or larger than the impact of the internet, which has become the world’s dominant application platform. There are equally strong arguments being advanced that the ceiling for general AI capabilities has already been reached, and that investments in the technology will deliver diminishing returns.

There is also, it has to be said, an alarming amount of money tied up in the question of which perspective will be proven correct – to the extent that Deloitte analysis from earlier this month concluded that “investment in gen AI drove the economy in the first half of this year”. And the data both backs up this assessment and exposes the precarity of it; analysts have identified that, today, the S&P 500 is more heavily weighted towards the top 10 companies than it has been in the last 60 years. Of those ten companies, at least eight are companies that are either AI pioneers or that have pivoted to AI.

Without putting too fine a point on it: what happens with AI, as a class of technologies being positioned to create a gigantic reservoir of value through automation, individual empowerment, and the promise of general intelligence and capability on tap, is what happens to the wider economy. Which is why any indication that AI will create all that unprecedented value is so celebrated, and why any suggestion that it can’t translate into what feels like an outsize sense of doom.

And this week the news, the markets, and the big AI companies themselves have all been reckoning with a signal suggesting that the value creation potential of AI is overblown. Or at least that return on investment in AI could be harder to come by than the optimists hoped.

Just days after ChatGPT-5’s launch, and in the shadow of Meta’s latest billion-dollar sweep of AI talent, MIT released a study suggesting that ninety five percent of businesses experimenting with AI have not been able to realise significant value from those investments.  Accordingly, markets responded, with tech stocks dipping across the board, and investors floating the idea of an “Altman bubble”. And the OpenAI CEO has certainly had a lot to say this summer that put the wind in AI’s sails, including writing in June that “we have recently built systems that are smarter than people in many ways, and are able to significantly amplify the output of people using them”. Which certainly sounds like something we should be able to measure, if not now, then soon…

Which is why the MIT study deserves to be taken seriously, even if it shouldn’t necessarily be read at face value. Its scope was fairly modest: 150 executive interviews, 350 employee surveys, and an analysis of 300 public AI deployments. That makes it valuable as a snapshot, but hardly sweeping and industry-defining. In our own Fashion and Technology survey (created in partnership with Fashion By Informa) we collected perspectives from  more than 160 businesses across a single sector, fashion, whereas the MIT survey set out to capture a much broader, cross-industry perspective. Scale aside, though, the important point is that the MIT numbers have landed at a moment when the market is already jittery, and Sam Altman himself has acknowledged, in the same week no less, that AI feels like a bubble where “smart people get excited over a kernel of truth”. And while the focal point for these conversations is the market performance of AI companies, the underlying question is the same one we’ve already enumerated here: “is AI actually going to create value for people and companies, and if so how can we measure it, and how soon?”

To answer that question, it’s important to look at the headline statistic in context. The data demonstrating that only 5% of companies are seeing real value from their AI initiatives refers specifically to integrated AI pilots, and it comes at the end of a funnel that already filters heavily.  Around eighty percent of enterprises have experimented with AI in some form. Of those, forty percent have deployed a pilot into a live setting. And at the far end, just five percent can point to measurable returns. Seen through that lens, the picture is maybe less catastrophic than the headlines make it sound. Still, it highlights how stubbornly difficult it is to turn generative AI into a quantifiable business benefit – let alone a repeatable one.

The study also points to two dynamics behind the figures that feel particularly relevant for fashion and beauty. 

The first is the gap between building internally and buying from outside vendors. Vendor led projects according to the report, succeed roughly two thirds of the time, while in house builds limp along at about one third. 

The second is the rise of “Shadow AI”, where employees sidestep official deployments and use consumer tools they’re already familiar with, like ChatGPT, Claude or Midjourney, to get their everyday work done. This kind of usage is inherently difficult to measure, and it also speaks to something fundamental: a belief among individuals that AI is the future of computing, contrasted with companies who are clearly convinced of the same thing (hence the widespread investment) but who are struggling to evidence it.

To cut to the chase: the real tension here exists between the promise of general-purpose AI models, sold and deployed ready to do work across any domain (which is an applicable promise in people’s personal lives, because individuals do a lot of different things every day!) and the reality that successful AI initiatives seem to be basically the opposite, being narrow, focused deployments aimed at solving specific and measurable problems.

This is also something we’ve been tracking over the last seven months. In the Fashion and Technology In 2025 survey, more than half of businesses (around 57 percent at the close of 2024) already had AI projects either deployed or in active development, and  that figure was expected to rise to 80 percent by the end of this year. A similar amount also expected AI to deliver dramatic additional value, but the concrete examples we found from industry professionals were almost entirely focused on discrete challenges rather than generally-capable chatbots.

As one survey respondent put it, they’d like to see “less focus on shiny possibilities for AI, and more focus on the nuts and bolts that machine learning might help us with.” It’s a simple line, but it captures the sentiment perfectly. 

The other lesson from this week’s cooling of markets and user sentiment is the importance of setting realistic near and mid-term expectations that people can bet on – whether they’re making those bets with capital, or whether they’re simply ‘finding religion’ around the idea that AI will transform the way they live and work. 

Obviously the major AI labs are incentivised to keep pursuing general models: systems that can handle any question, any task, across any domain, because their customer bases are incredibly broad. Their valuations depend on the promise that generative AI is not going to become a scattering of different solutions, but rather a single monolithic entity that can take on anything we choose to throw at it. Enterprises, however, are clearly discovering that value primarily emerges when projects get hyper specific and pare way, way back on that general promise. This is what we referred to in this year’s AI Report as the “product era” – a transition from AI has a nebulous cloud of possibilities, to AI as applications that can be compartmentalised, bought, implemented, and then assessed according to measurable as-is and to-be criteria.

All of this leaves the sector without a clear answer. The hype might finally be cooling where the big AI labs are concerned, but that same optimism isn’t vanishing but rather turning inwards, towards industry experts creating specific tools (with AI in them) to address real pain points.

The idea of a bubble stretched large and about to burst paints a dramatic, headline-ready narrative. That doesn’t mean it’s an incorrect story, of course; it’s almost inevitable that a stock market over-indexed on a single promise will contract when that promise gets tempered.What’s less immediately interesting to read is that the time of AI-as-experiment may be over, and that what remains is the embedding of AI where it actually helps. For The Interline’s writers and audience, who spend every day thinking about implementations, integrations, and practical reality, that’s catnip. For people whose pocketbooks and personalities are tied up in the idea that AI will change the world on a gigantic scale at speed, this week suggests they should instead be thinking about a more predictable technology adoption and maturity curve – one that ends with AI as a major force, but one that has more than its share of bumps in the road ahead. 

Best from The Interline:

Kicking off this week, Darya Badiei Khorsand on the rise of AI-generated personas. And whether, in a world of curated illusion, does it matter who, or what, is speaking to us?

Next, we announced a new downloadable briefing for the wider community of current fashion and beauty professionals, and the incoming generation of hybrid talent and digital decision-makers – coming this Autumn.

In our first news analysis of the week, we delve into the ever changing threat of tariffs. When trade rules change overnight, planning takes on a different definition, and early investments in building supply-side flexibility are already starting to pay off.

Next, The Interline’s own Dan Butt on how fashion could harness the potential of AI in determining what to make, and when, but without compromising its originality.