‘I believe it’s a bubble’: Will AI ever produce enough revenue to justify the price? The maths is daunting

NEW YORK – In June, Mr Elon Musk predicted that artificial intelligence (AI) would become smarter than any human by the end of 2026. In July, OpenAI chief executive Sam Altman said his company’s product “is going to reshape the course of human history”. Mr Mark Zuckerberg has imagined creating “a personal superintelligence that helps you achieve your goals, create what you want to see in the world”.

The CEOs are backing up their words with dollars. Five of the biggest tech companies are together spending an estimated US$371 billion (S$482 billion) in 2025 on the massive data centres necessary to train and run complex AI models. That figure is expected to increase in the years to come: McKinsey & Co projects that data centres will require US$5.2 trillion in spending by 2030 to keep up with demand for AI.

All that spending raises the question: Will AI ever produce enough revenue to justify the price? The maths is daunting. In 2025, AI technology is expected to generate US$60 billion in revenue, according to one estimate by writers Azeem Azhar and Nathan Warren of AI-focused newsletter Exponential View. That number will need to increase dramatically if tech companies want to recoup costs. In September, Bain & Co calculated that Big Tech would need US$2 trillion in additional annual revenue to pay for data centre expenditures by 2030 and projected a shortfall of US$800 billion a year even under ideal circumstances. If AI giants and their investors can’t make their money back, then we’re talking about a historically large case of overbuilding and overinvestment. There might even be a word for that.

“I believe it’s a bubble,” says Mr Harris Kupperman, founder of Praetorian Capital Management, a self-described contrarian hedge fund with about US$300 million in assets under management. “Will there ever be a payback on this stuff? I think the answer is ‘highly unlikely’.”

Covering the cost of just 2025’s build-out will require additional revenue of US$480 billion, according to Mr Kupperman’s calculations. Where that will come from is unclear, especially considering that for most users, ChatGPT is currently free. “If they charged you a couple dollars every time you queried ChatGPT, I don’t know if they’d have a market,” Mr Kupperman says. 

It doesn’t help that graphics processing units (GPUs) – the essential computer chips for AI that account for a significant chunk of data centre costs – depreciate quickly. In past bubbles, such as the railroad construction of the 19th century or the telecommunications build-out of the early 2000s, the overspending at least created lasting infrastructure. Even if people didn’t use the railroad tracks right away, they remained useful for decades, as was the fibre-optic cable laid in the 1990s. GPUs, in contrast, appear to have a shelf life of just a few years before they need to be reassigned to more basic AI tasks. The maths makes AI spending seem less like a flywheel than a hamster wheel. 

Then there are the bottlenecks, which could prevent many planned AI factories from reaching completion. While data centres typically take two to three years to build, hooking them up to an energy source can take up to eight years, according to Boston Consulting Group. So it will take that much longer for data centres to begin generating revenue. And that’s if the energy is available at all: A 2024 review by the Commonwealth of Virginia, the data centre capital of the world, concluded that meeting the unconstrained energy needs of this infrastructure would be “very difficult”, and even meeting half of that demand would be “difficult”.

Sceptics also point to disappointing performance by the AI tools themselves. A much-cited study by the Massachusetts Institute of Technology Media Lab found that 95 per cent of AI projects piloted by businesses have produced no measurable return. McKinsey reported that almost eight in 10 companies that adopt generative AI see “no significant bottom-line impact”. OpenAI released GPT-5 in August to tepid reviews, raising questions about the industry’s assumption that more data is better when it comes to AI.

Finally, observers concerned about an AI bubble point to the circularity of recent deals – for example, Nvidia is selling chips to OpenAI while also investing in it – that has echoes of the telecoms bubble.

They also worry about the growing opacity of funding mechanisms. In August, Meta Platforms raised US$29 billion for data centres from private credit firms. Big start-ups such as OpenAI and CoreWeave have also leaned on private credit for data centre financing. These agreements often take the form of obscure special purpose vehicles, which keep the debt off the big companies’ balance sheets and make the health of the investments hard to track.

There are also signs that the exposure of average investors to these risky bets is growing, as private credit firms raise money from insurance companies and real-estate-focused exchange-traded funds invest in data centres, according to investor and writer Paul Kedrosky. And of course anyone who owns shares in a mutual fund is most likely betting on data centres via Big Tech stocks.

The AI bulls – which, judging by stock market trends, include most investors – aren’t too concerned. In their view, revenue from generative AI is growing and will continue to grow. And while the infrastructure build-out looks big, it’s not that big compared with certain past bubbles, write Mr Azhar and Mr Warren of Exponential View. As a proportion of gross domestic product, spending on railroads during that boom was four times greater than current spending on AI infrastructure. Most important, Big Tech companies have plenty of cash; they can afford to lose a few hundred billion dollars. Mr Altman and Mr Zuckerberg have both suggested they’re focused on the long game rather than immediate profits. 

The question is how long their investors will take the same view. The challenge of any bubble is not knowing if it will pop but when. Entrepreneurial studies clinical assistant professor Erik Gordon of the University of Michigan says the first signs of fading optimism are likely to appear in venture capital funding for AI start-ups. “You might see the size of funding rounds go down,” he says. Public stock prices would most likely follow. (For now, it’s still up, up, up.)

Until then, the AI explosion is something of a Rorschach test: What you see depends on how you view AI’s prospects broadly. Haters have been calling the boom financially unsustainable for ages. Those who believe in the technology’s potential seem to have faith that it will still grow and improve fast enough to recoup costs – even if the traditional laws of the market suggest otherwise. BLOOMBERG