Bain says AI needs $6 trillion a year in new revenue by 2031 to pay for itself

The AI boom has a revenue problem, and Bain and Company has put a number on it. In its seventh annual Global Technology Report, released September 29, the consulting firm estimates that artificial intelligence will need to bring in about $6 trillion in new annual revenue by 2031 to justify what companies are pouring into chips, data centers and power. Cheaper office work will not get the industry there on its own. Most of that money, Bain argues, has to come from products that barely exist today.

That is a striking conclusion from a firm that advises many of the companies doing the spending. It does not say the AI build out is a mistake. It says the build out is a wager, and the wager only pays if customers start buying very different things than they buy now.

Where Bain says $6 trillion in AI revenue has to come from

Bain broke the $6 trillion figure into three parts:

• About $4.2 trillion from new products and categories that are only starting to take shape.
• Between $1 trillion and $1.4 trillion from enterprise customers.
• Between $200 billion and $400 billion from consumers.

The firm said infrastructure, meaning the chips, data centers and power behind AI, accounts for about $1.5 trillion of that opportunity.

The Next Web, which went through the report in detail, reported that Bain expects existing AI products to generate only $1.2 trillion to $1.8 trillion by 2031. The rest of the gap is what makes the $4.2 trillion line so important. It is the part of the forecast that depends on markets that have not proven themselves yet.

According to The Next Web, Bain pointed to a few places where that new money could come from: chatbots that carry advertising and take share from traditional web search, self driving vehicles and industrial automation, and what the industry now calls physical AI, which covers robots, digital twins and simulation. Each of those is plausible. None is a sure thing on Bain's timeline.

"The economics of AI infrastructure demand trillions in new revenue beyond productivity gains," said David Crawford, chairman of Bain's Global Technology Practice, in the firm's announcement.

Why productivity is not enough

The core of the report is a simple piece of arithmetic. If AI mostly helps companies do existing work with fewer hours, the savings show up as lower costs for the buyer. Some of that value flows to the AI vendor as subscription fees, but much of it stays with the customer, and some of it disappears into lower prices for end consumers. That is good for the economy. It is not enough to pay for the hardware.

The spending side keeps climbing. The Next Web reported that Bain estimates Microsoft, Google, Amazon, Meta and Oracle will spend roughly $780 billion on capital projects in 2026 alone, and that the largest AI data centers are doubling in size and cost every 12 to 16 months. A data center is not a one time cost either. Chips age quickly, power bills arrive every month, and the next generation of hardware is already on order before the current one pays off.

So the industry needs to sell things people and businesses did not buy before. Bain's examples include AI agents that complete whole tasks rather than drafting a paragraph, new kinds of software built on top of models, and services that would not be economical without them. If you want a sense of how far today's agents are from that goal, our explainer on what AI agents can do and where they still fail is a useful reality check.

The physical limits

Money is only one constraint. Data centers need land, grid connections, cooling, skilled electricians and local permits. The Next Web reported that Bain counted at least 75 projects worth about $130 billion that were blocked or delayed by local opposition in the first quarter of 2026, close to the total for all of 2025. Power, chips, labor and permits all face shortages, according to the report.

That matters for the revenue question. A company that cannot get a facility online on schedule cannot sell the services it was built to run, which stretches out the payback period on everything else.

The security warning

Bain also flagged a sharp change in cyber risk. "AI has massively compressed the time required by a typical cyberattack from an estimated four weeks to about 18 hours," the report states.

Defenders have felt that squeeze all year. When a critical flaw lands in widely used gear, the window between disclosure and mass exploitation is now measured in days, as the recent Citrix NetScaler zero days showed. For ordinary users the lesson is less dramatic but still real: the basics, like unique passwords and strong sign in protection, matter more when attackers can automate the boring parts of their work. Our guide to protecting your online accounts in 10 minutes covers the quick wins.

Who this affects

Investors. The report is a reminder that the AI build out is a bet on future demand, not a guaranteed return. Companies spending at this pace need new categories to take off within a few years. If adoption runs slower than planned, the pressure will show up first in depreciation charges and in questions on earnings calls about when the spending turns into profit.

Businesses buying AI. Vendors need revenue, and they will push hard to get it. Expect more bundling, more usage based pricing, and more products pitched as agents that replace a workflow rather than assist with it. That can be a good deal, but it is worth asking for clear measures of what the tool actually saves before signing multiyear contracts.

Everyday users. The report explains why AI features keep appearing in nearly every app and subscription, often whether you asked for them or not. The industry needs paying customers at a scale it has not reached yet. Some of those features will be useful. Others exist mainly to justify a price increase.

Policymakers. Large spending commitments tend to bring requests for faster permits, cheaper power and lighter rules. The debate over voluntary commitments, which we covered in our piece on why AI safety pledges need public proof, will only get louder as the financial stakes rise.

What to watch next

A few signals will show whether Bain's math is holding up:

1. Paid adoption, not trial use. Free users do not cover chip bills. Watch for companies reporting paying seats and revenue per customer for AI products.
2. Agent products that work reliably. The $4.2 trillion bucket depends heavily on software that completes tasks end to end. Reliability, not demos, will decide whether businesses pay for it.
3. Ad supported chatbots. If assistants start carrying ads at scale, that is a sign the industry is reaching for the advertising model that funded web search.
4. Data center delays. More local pushback or power shortages would slow the build out and lengthen payback periods.
5. Capital spending guidance. Any pause or cut from the largest cloud companies would say more about expectations than any forecast.

You can follow all of these threads in our AI coverage.

Frequently asked questions

Does Bain think AI is a bubble?

Not in so many words. The report frames the spending as a bet that needs new revenue to pay off. It does not predict a crash, but it makes clear that today's products alone are not expected to cover the bill.

What counts as "new revenue" in the report?

Money from products and categories that are only starting to form, such as AI agents that handle full tasks, ad supported chatbots, self driving and industrial automation, and robotics. It is separate from savings companies get by making existing work cheaper.

Why does the cyberattack figure appear in a revenue report?

Because AI changes costs on both sides. Bain says AI has cut the time a typical attack takes from about four weeks to about 18 hours, which raises security spending and risk for every company adopting the technology.