Global spending on data centers and other digital infrastructure is accelerating amid an artificial intelligence arms race. But for companies to see a return on this massive investment, they will need to generate trillions of dollars of new AI revenue that doesn’t exist today, according to a report released this week.

The report from consulting firm Bain & Co. projects that the artificial intelligence industry will need to generate $6T in annual revenue by 2031 to justify the unprecedented wave of capital pouring into data center development.

Existing AI services — both consumer-facing products and enterprise AI applications — could generate up to $1.8T of that total, the report’s authors estimate. But that leaves a $4.2T gap in annual revenues that will have to emerge in the next four and a half years.

“Productivity gains from existing enterprise and consumer applications won't be enough,” the report’s authors wrote. “[E]ntirely new markets must emerge to close the funding gap.”

What will help bridge that gap — or whether it will be bridged at all — remains in question.

Consumer AI products like ChatGPT could generate as much as $400B in revenue by 2031 through subscriptions and advertising, according to Bain. Meanwhile, the report estimates that enterprise AI adoption could contribute as much as $1.4T in value as companies use AI to improve productivity in areas like software development, sales and IT operations.

The rest, according to Bain, will have to come from completely new sources of economic value — technologies and business models that either don’t exist today or remain in their infancy but ultimately emerge as “breakthrough AI applications that transform industries and expand the global economy.”

This conclusion echoes a concern voiced by some critics of the AI spending boom: that tech giants are betting on the emergence of a yet-to-be-invented “killer app” that will make their trillions of dollars in digital infrastructure investment worth it.

“Dramatic innovation will be required to deliver the revenue necessary to fund the gap,” the Bain report said.

Big Tech companies have ramped up annual capital expenditures on AI into the hundreds of billions. In their second-quarter earnings reports, Google, Amazon and Microsoft estimated that their full-year capex would total $205B, $220B and $190B, respectively.

Bain points to four market segments that are likely to deliver at least some of the innovation needed for AI revenue to catch up with skyrocketing capex.

It predicts around $900B in annual revenue from “physical AI”: models that allow realistic simulations and AI-powered robotics. Autonomous vehicles and industrial automation are expected to provide around $400B in additional annual revenue gains, while the integration of advertising into AI search and other tools could add $200B.

The rest will have to come from the development of new AI products that don’t exist today. These new use cases may yet emerge, but the clock is ticking.

“The industry needs a wave of application innovation comparable with what mobile and cloud unlocked, not just productivity gains on existing workflows,” the report’s authors wrote. “The question is whether the applications arrive in time to pay for it.”