British investor and author – not to mention amateur DJ – Azeem Azhar is the founder of research platform Exponential View, which recently released a report, The State of the AI Economy. He tells us about some of the shifts he sees heading our way as AI becomes ever more deeply embedded in our business and personal lives.
-
A real AI economy is emerging at speed
“There is a lot of confusion about the size of the AI economy – what are customers really spending? Are we counting things twice? Are we missing things out? Think about when a customer spends a dollar with an AI company and that company then spends some of that dollar on, say, Azure cloud services from Microsoft. Maybe they spend 50 cents of it. So should that be counted as a dollar for the AI economy or $1.50? If you don’t de-duplicate your revenue, you get a much larger number. So we went through the painstaking exercise of de-duplicating. In the 12 months to June 2026, there was $110 billion of what we call ‘de-duplicated revenue’ in the AI sector worldwide, excluding revenues in China. That has tripled in a year and it is still growing very fast, three times faster than things like the internet, mobile apps or online advertising did. It’s a really speedy technology wave.”
-
There’s still major room for growth
“$110 billion is huge in one way, but small in another. It’s 0.4% of U.S. GDP, so it’s a large industry but a small economic force thus far. And what we need to understand is that many companies are only just now starting to get the results that might get them excited. It takes a bit of time. It was only in 2023 that AI tools were widely available to enterprises from companies like Microsoft and Google. Three years is about the time that you should see the early adopters seeing meaningful success: you need about 18 months to get the first successful pilots and then a year and a half to scale those out. We are starting to see this, but I think it has been slower to provide a return on investment than the media narrative has presented – and slower I think than the technology companies thought it would be for their customers.”
-
Expectations are becoming more realistic – but ambition remains high
“The thing to look at is how Main Street, the broad brush of American and global business, is talking about AI. They are talking about their AI projects more and more, increasingly in quantified terms. When I talk to bosses of those types of companies in the United States and in Europe, they tell me that they’ve had some success, but not enormous success, and yet they’re going to continue to spend more money. They’re confident that they’re going to get the results over time, but they now recognize that it’s not straightforward. So I would expect that the amount being spent by businesses will increase pretty significantly. There’s data from Ramp, a payments processor, which shows that on average, their customers – which tend to be the more forward-leaning digital companies in the United States – are spending $11 to $12 per employee per month on AI. That’s not a huge amount. That is around the value of the free coffees they offer in the staff canteen. I would expect that number to rise reasonably quickly. The full 12-month revenues by our measure for 2026 will still be less than $200 billion. But that’s extraordinary given it was zero in 2022. And I think there’s a chance that you might see them go up by 80 to 100% in 2027.”
“Companies are going to get more ambitious about what they can do”
-
Companies will get crisper on AI spending
“As companies spend more, they move from the CFO saying, ‘Okay, guys, just experiment!’ to saying, ‘I would like to know how many dollars I’m getting for the tokens you’re buying.’ So we’re going to get that cost discipline, and that’s actually good news for everyone involved because once you can prove you’re getting a return, you’ll allocate resources. And if you’re not getting a return, you won’t waste your money.”
-
Businesses are going to aim higher
“Companies are going to get more ambitious about what they can do. They will move from the lightweight entry point of Copilot summaries to trying to get workflows done, taking advantage of the fact that the models and the harnesses around them are getting more reliable and can undertake tasks of greater length with greater accuracy.”
-
New models will shake things up
“Highly capable open-source models will reduce the relative and average cost of tokens to a business. I think that the average per-token cost will decline and the bulk of workloads will move to lower-cost models, including open-source models because, ultimately, if you just need to get your blood pressure taken, you go to your family physician, you don’t go to a tier-four university teaching hospital. Lots of things in business are as simple as that, but the volumes and the fact that more companies will be moving through the maturity curve will more than make up for the price competition.”
“When AI offers an infinite range of possibilities, the way you articulate yourself as a human really matters”
-
An agent-mediated world could mean richer lives
“We can’t yet know precisely what could be unlocked in our societies by mass adoption of AI agents. But one thing that could happen is that people will be able to get more done, whatever that more is – it might be more gardening, or more time finding exactly the dress they want to buy. The reason for that change is that lots of the friction around knowing what to do with AI will start to disappear. That has really been my experience as an early adopter. I’m going deeper into things than I otherwise would have done, and it is eliminating the points where I’m a bit incompetent. What happens when there are loads of those agents flying around? We might see many more people experimenting with quirky, curious personal projects.”
-
Technology will lower the access cost of getting into anything
“The moment that really hit me was about two years ago with a piece of software called DJ Studio which lets you make mixes without needing the equipment – the CDJs, mixer and so on – and uses AI to help with transitions. That was a really easy way for me to get into mixing, and it was quite helpful because I could just build playlists and identify energy points and vocals, which used to be quite a manual process. That makes a really big difference. Of course, what happens is that you get a feel for it, so then the next step you take is to get the hardware. So I have a pair of decks in my office now.”
-
Human judgment is still needed
“What the software hasn’t done is provide any taste. I built a piece of software to go through my 4,500 tracks to build a playlist that would tell a story. Technically it worked, but the algorithm chose tracks that didn’t work well together. Fine for the machine, a cacophony for the human ear. What I came away with is that when AI offers an infinite range of possibilities, the way you articulate yourself as a human really matters. You have these tools that augment you, but you need to do the virtual equivalent of piling through the vinyl in a record store, trying to find the bootleg cut that is going to fit a given mix.”
“Companies are only just now starting to get the results that might get them excited. It takes a bit of time”
-
We’ll be forced to think about what we mean by skills and skilling
“If we go back to 2016, when everyone first got excited about AI, the jobs that we thought were going to be taken on by AI were routine office jobs. But there’s recent data from the Philippines, which was going to be in the eye of this storm, that the contact center and offshoring market is actually growing in terms of number of employees despite the AI wave. And what has actually turned out to be the case is that the cost advantage to get AI to reliably answer queries that are handled by routine office staff is not huge. It’s not a 10 times improvement; it’s currently about 40 percent and that may not give companies enough of an incentive to make that change. But with higher value, higher wage work like entry-level investment banking analysis and strategy consultant jobs, it could be really worthwhile because those jobs are expensive and the AI can do them for considerably less.”
-
The best CEOs will focus on loops
“Most CEOs have a very short tenure, and I think many of them probably just don’t want to be fired, so it’s much easier for them to say they’re doing the AI thing but play it safe than to take a genuine risk. But for others, the main thing they will want to do in this new world is be very explicit about their company’s capacity, ability and the rate at which it learns from its experience in the world. A company has experience in the world because it gets new orders from customers, new complaints from customers, competitive pressures, suppliers behaving well or badly. All of those are new signals that don’t exist in the company’s existing pool of knowledge or indeed the training data of any AI. The faster and better a CEO can effectively identify signals, learn from them and adapt, the better. So the question is, where can you learn, with the help of AI, to improve the speed and quality and cost of what you do? The best bosses will focus relentlessly on this question.” ●
You can read The State of the AI Economy report from Exponential View here