Bain & Company said in its annual global technology report that the AI industry must reach $6 trillion in annual revenue by 2031 if the current wave of data center investment is to hold up financially.
The firm estimated that existing consumer and enterprise AI services can contribute at most $1.8 trillion of that total. That leaves another $4.2 trillion to be created by new applications and business models that do not yet exist. Bain said the pace of AI infrastructure construction is already running far ahead of the demand curve.
Bain sets a $6 trillion revenue threshold for 2031
In the report, Bain set out a clear benchmark: AI would need to generate $6 trillion in annual revenue in 2031 to justify the data centers now being built around the world.
According to the report, current consumer and enterprise AI services can only account for up to $1.8 trillion, meaning the remaining $4.2 trillion would need to come from revenue streams that have yet to be built.
Crawford, Bain’s global technology media chair, said the industry needs a wave of innovation far larger than what mobile internet and cloud computing produced. He added that AI infrastructure is being built at a pace that is visibly ahead of demand. To support that investment on a sustained basis, global GDP growth would need to be about 1 percentage point higher per year.
The hurdle rose from $2 trillion to $6 trillion in a year
In Bain’s sixth global technology report, published in September 2025, the firm estimated that meeting AI computing demand by 2030 would require about $500 billion a year in data center investment, equivalent to roughly $2 trillion in additional cloud revenue. Even after accounting for AI-driven cost savings, the annual shortfall was still about $800 billion.
At that time, the report assumed global AI computing power demand would require 200 GW of electricity.
One year later, the threshold has been lifted to $6 trillion and the shortfall has widened to $4.2 trillion. The timeline has moved back by only one year, but the overall scale has expanded by about three times. The report said the larger gap comes mainly from swelling costs and buildout scale, not from a sudden weakening in demand expectations, because industry spending is already far above current revenue.
Data center spending could reach $5 trillion to $6.5 trillion by 2030
Bain estimated that global data center spending could total $5 trillion to $6.5 trillion by 2030, while adding at least 150 GW of new capacity. That expansion would also put pressure on national energy systems.
If data centers, computing capacity, and upgrades to accelerators and memory are all included, annual AI infrastructure spending could reach as much as $1.5 trillion by 2031.
The report said one reason costs keep rising is that chips from Nvidia and SK Hynix, along with networking gear and other components, continue to get more expensive. As a result, the scale and cost of data centers are roughly doubling every 12 to 16 months.
Microsoft, Alphabet’s Google, Amazon, Meta and Oracle are still investing in data center construction at a trillion-dollar scale.
Where the missing $4.2 trillion could come from
Bain said the opportunity to close the gap may come from early-stage sectors such as autonomous equipment and robotics, as well as emerging uses in drug development, mental health and energy production.
The report also made a point it said is often overlooked: much of the market’s discussion around AI value is centered on workforce productivity, but the economics of AI infrastructure require trillions of dollars in entirely new revenue beyond productivity gains. Doing the same work faster and cheaper is not enough to pay for these data centers.
Physical constraints include transformers, water, power and local opposition
Beyond financing, Bain said developers are already running into shortages of transformers, as well as limited water and electricity supplies.
The report said that in the second quarter alone, $68 billion worth of U.S. data center projects were blocked or delayed because of strong local opposition. Those disputes have largely centered on electricity use, water consumption, noise and heat emissions, and Bain said the pushback is spreading.
The report also said rising prices for memory and accelerators are both a sign of supply tightness and a direct driver of higher revenue thresholds for the industry. That, in turn, lengthens the payback period on investment.
Still, if revenue growth from AI applications keeps pace, the investment cycle can continue. If growth falls short of expectations, the highest-cost parts of the stack may be the first to come under review. Bain’s latest upward revision is a sign that this risk is becoming harder to ignore.