The gap is $4.2 trillion
Bain & Company’s seventh annual Global Technology Report, published on Monday, puts a number on the question the industry keeps deferring. To pay for the compute it is building, AI needs to generate about $6 trillion in annual revenue by 2031.
Bain’s own tally of where that could plausibly come from today — consumer subscriptions and advertising, plus enterprise use in software development, sales, marketing, customer service and IT operations — reaches $1.2 trillion to $1.8 trillion.
That leaves roughly $4.2 trillion that does not yet exist in any recognisable form.
Where Bain thinks it comes from
The report does not treat the gap as unfillable. It sketches where new revenue would have to originate, and the categories are revealing because none of them are chatbots.
Physical AI — simulations, digital twins and robots — is put at $900 billion. Self-driving cars, trucks and drones, together with other industrial automation, are a $400 billion opportunity. Advertising-carrying chatbots that replace much of web search add $100 billion to $200 billion or more. The remainder would have to come from areas Bain lists but does not size individually: drug discovery, mental health applications, energy generation.
Annual spending on AI infrastructure, on Bain’s numbers, could reach $1.5 trillion by 2031 — covering new data centres and compute plus upgrades to chips, memory and networking already installed.
“The economics of AI infrastructure demand trillions in new revenue beyond productivity gains,” said David Crawford, chairman of Bain’s global technology practice.
Hardware is growing four times faster than software
One figure in the report explains the shape of the problem. Between 2020 and 2026, hardware and semiconductors grew at 24 per cent a year against 6 per cent for software.
That is an industry building capacity far faster than it is building the businesses that would consume it. Anne Hoecker, who heads Bain’s global technology practice, frames the consequence in procurement terms: “Supply chain and procurement strategies are increasingly sources of competitive advantage.”
The report also notes that AI has compressed the timeline of a cyberattack from about four weeks to 18 hours, and puts leading labs’ investment in forward-deployed engineering — consultants who embed with customers to make deployments work — at $9.75 billion.
How it fits with the other forecasts
Bain’s is the third large estimate in a week, and they do not contradict each other so much as measure different things. Goldman Sachs put hyperscaler capital spending at $1.2 trillion for next year. A Brookings paper priced the full US build-out at $10.3 trillion through 2032. Bain is asking the demand-side question the other two leave open: who pays the bill, and for what.
The number to watch is not $6 trillion. It is whether any of the $4.2 trillion categories — robotics, autonomous vehicles, drug discovery — produces a business with real revenue before the depreciation schedules on the current build-out come due.
