AI needs $6tn a year by 2031 to pay for its data centres, Bain says

The AI industry will need about $6tn in annual revenue by 2031 to pay for the data centres being built today. Bain & Company makes the estimate in its Global Technology Report, published on Tuesday. Existing AI products could bring in only $1.2tn to $1.8tn of that.

“The debate today is fixated on employee productivity. The economics of AI infrastructure demand trillions in new revenue beyond productivity gains,” David Crawford, chairman of Bain’s global technology practice, said in a statement.

AI infrastructure is being built well ahead of demand, Crawford said. Funding it would require adding about 1% to the annual global GDP growth rate.

How Bain gets to $6tn

Annual spending on AI infrastructure could reach $1.5tn by 2031, the report says. That covers new data centres and compute, plus upgrades to the chips, memory and networking already installed.

Bain assumes capital spending runs at about 25% of industry revenue, in line with trends among cloud providers. At that ratio, the AI market would need to approach $6tn a year.

Microsoft, Google, Amazon, Meta and Oracle could spend $780bn on capital expenditure in 2026, Bain estimates. That is nearly five times the level of three years earlier. The largest AI data centres double in size and cost every 12 to 16 months or so, according to Epoch AI data in the report.

Where the other $4.2tn could come from

Consumer AI, through subscriptions and advertising, could bring in $200bn to $400bn by 2031. Enterprise AI could add $1tn to $1.4tn for providers, from software development, sales, marketing, customer service and IT. That leaves about $4.2tn still to find.

Bain names four sources for it. Chatbots that carry ads and replace much of web search could add $100bn to $200bn or more. Self-driving cars, trucks and drones, along with other industrial automation, are a $400bn opportunity.

Physical AI, meaning simulations, digital twins and robots, could be worth $900bn. That assumes a 10% cut in R&D and manufacturing costs. Products that do not exist yet could help close the gap. Bain gives AI-driven drug discovery, mental health support and new materials for batteries and chips as examples.

Built ahead of demand

Bain’s data centre model projects $5tn to $6.5tn of spending on nearly 150 gigawatts of new capacity by 2030. That would almost triple global capacity in five years. Power, chips, skilled labour and permits are all short at the same time, the report says.

Local opposition blocked or delayed at least 75 projects worth $130bn in the first quarter, Bain says. That nearly matches the $156bn of projects disrupted in all of 2025. In July, New York became the first US state to pause giant data centres. Fourteen other states have introduced similar bills.

On Tuesday, OpenAI, Blackstone and US unions formed an alliance to head off state moratoriums. Earlier this month, PwC projected $31tn of AI infrastructure investment by 2050.

The report also finds that AI has cut a typical cyberattack from about four weeks to 18 hours.