The AI industry will need to generate $6 trillion in annual revenue by 2031 to cover its compute bills, but existing products can only deliver about $1.8 trillion of that, according to Bain & Company's seventh annual Global Technology Report released Tuesday. That leaves a $4.2 trillion hole that will have to be filled by entirely new markets, from robotaxis to AI-designed drugs, or the buildout could stall.
The consultancy's math assumes AI infrastructure spending climbs to $1.5 trillion a year by 2031, with capital expenditures equaling roughly 25% of industry revenue. That's a level of sustained investment that has no real precedent outside of government wartime spending.
Where the money is supposed to come from
Bain breaks down the $1.8 trillion in likely existing revenue first. Consumer subscriptions and advertising, the stuff most people associate with ChatGPT and its rivals, could bring in $200 billion to $400 billion by 2031. Enterprise uses — software development, customer service, back-office automation — are projected to add $1 trillion to $1.4 trillion.
That's the floor. To reach $6 trillion, Bain identifies four categories that don't exist yet at meaningful scale: advertising inside chatbots and AI-powered search ($100 billion to $200 billion-plus), autonomous vehicles and industrial automation ($400 billion), physical AI like robotics and digital twins ($900 billion), and new applications including drug discovery, mental health support, and materials science.
Some of those bets are already being placed. Waymo operates commercial robotaxis in several U.S. cities. Pharmaceutical companies have signed deals with AI firms to hunt for new molecules. But none of these markets is close to generating the kind of revenue Bain says will be needed within seven years.
The spending is running ahead of the demand
Hyperscaler capital spending tells the story of how fast the buildout has moved. Bain estimates Microsoft, Google, Amazon, Meta, and Oracle will spend a combined $780 billion in 2026 — nearly five times what they spent three years earlier.
The physical infrastructure is scaling just as aggressively. Leading AI data centers now approach 1 gigawatt of power capacity, with many expected to near 2 GW by 2027 and 9 GW campuses by the end of the decade. According to Epoch AI figures cited in the report, data center size and cost are doubling roughly every 12 to 16 months.
David Crawford, chairman of Bain's global Technology practice, said efficiency gains alone won't cover the spending. He noted that AI infrastructure is being built well ahead of demand, and that sustainably funding it would require adding about 1% to annual global GDP growth. That's a tall order. Global GDP growth has averaged around 3% in recent years, meaning AI would need to contribute roughly a third of all future growth.
What happens if the gap doesn't close
The report doesn't predict a crash. It lays out the arithmetic and leaves the implications hanging. But the numbers suggest that if the new revenue categories don't materialize — if robotaxis stay confined to a handful of cities, if AI drug discovery produces more hype than approved medicines, if consumers balk at paying $200 a year for a chatbot — the infrastructure that's being built now won't pay for itself.
The alternative is that AI companies find a way to charge for something people and businesses actually want at a price they'll pay. That's not a technology problem. It's a business model problem, and Bain's figures suggest the industry has about seven years to solve it.
The full report, including regional breakdowns and sector-specific forecasts, is available through Bain's website.
