AI spending could require roughly $3.7 trillion in annual industry revenue by 2032 to justify current infrastructure plans, according to one economic model.
AI spending is reaching a scale where the hardest question is no longer how much companies can invest. It is how much revenue the industry must eventually generate to make that investment worthwhile.
One economic model puts the hurdle at roughly $3.7 trillion a year by 2032.
That would equal about 9.2% of projected U.S. gross domestic product.
According to The Wall Street Journal, the figure comes from analysis by Columbia Business School professor Stijn Van Nieuwerburgh.
It is an investment-return scenario, not a prediction that Americans will definitely spend 9% of GDP on artificial intelligence.
That distinction matters.
AI spending is creating an enormous revenue hurdle
Van Nieuwerburgh is studying the infrastructure required to train and operate increasingly powerful AI systems.
His estimates include data centers, specialized chips, networking equipment and power infrastructure.
The Brookings Institution summary of his research estimates that the U.S. AI infrastructure buildout could require about $10.3 trillion in investment between 2025 and 2032.
That averages roughly 3.63% of U.S. GDP each year during the period.
The 9% figure answers a different question.
If investors want the infrastructure to produce a 10% unlevered return, the analysis estimates that AI providers would eventually need about $3.7 trillion in annual revenue.
That works out to roughly 9.2% of projected U.S. GDP in 2032.
Current AI spending therefore creates a large gap between the money going into infrastructure and the revenue today’s leading AI companies generate.
Closing that gap would require enormous growth.
The infrastructure buildout could reach $10.3 trillion
Much of the AI spending story is physical.
AI may look like software to consumers, but a ChatGPT query or Claude coding task ultimately runs through buildings filled with expensive chips.
Those facilities need electricity, cooling, networking and backup systems.
Van Nieuwerburgh estimates the U.S. could add about 183 gigawatts of data center capacity through 2032 under the central scenario in his research.
Brookings says the scale of investment would be larger relative to the economy than major U.S. infrastructure booms involving canals, railroads, highways and telecommunications.
That does not mean every project currently proposed will be built.
Some facilities may be delayed or cancelled. Chip prices can change. New models may become more efficient.
Those variables make long-range AI spending estimates unusually sensitive to assumptions.
The AI Decode has already examined the financial pressure around the data center IPO boom, where investors are asking harder questions about power availability, debt and whether customer contracts are strong enough to support expensive projects.
The same concerns apply across the wider AI economy.
Falling AI prices could make the math harder
There is another problem for companies trying to earn a return on AI spending.
AI services may become cheaper.
Competition between OpenAI, Anthropic, Google and other developers has repeatedly pushed down the cost of using models.
Hardware is also improving.
A company may therefore process far more AI queries in 2032 while earning less money from each query.
This is good for customers. It complicates the investment case.
If the cost of AI falls quickly, usage must rise even faster for revenue to reach the levels implied by today’s infrastructure commitments.
The industry’s optimistic argument is that cheaper AI will create new demand.
Companies may use agents for coding, research, customer service, logistics, medicine and other work that is still largely performed by people.
If those applications create substantial economic value, very high usage could be possible.
What remains uncertain is how much of that value AI providers will capture as revenue.
AI spending is increasingly being financed outside Big Tech
The way the infrastructure is financed is also changing.
Early AI spending could largely be supported by the enormous cash flows of companies such as Microsoft, Alphabet, Amazon and Meta.
The planned buildout is becoming too large for corporate cash alone.
Van Nieuwerburgh’s research says financing is increasingly moving into joint ventures, private credit, project debt, securitization and special-purpose vehicles.
These structures can bring more capital into the industry.
They can also make financial exposure harder to see.
A data center may depend on a single large AI customer. Its debt may assume that the customer keeps renting computing capacity for years.
If demand disappoints, the loss can spread beyond the technology company that originally ordered the infrastructure.
The AI Decode’s reporting on OpenAI’s $38.5 billion loss warning illustrates the other side of that equation. Fast revenue growth can exist alongside enormous research and infrastructure costs.
For the wider market, that makes profitability as important as adoption.
9% of GDP is a test, not a forecast
The headline number needs careful interpretation.
The analysis does not say AI spending will definitely reach 9% of GDP every year.
It says annual AI revenues may need to reach roughly that share of projected GDP by 2032 under specific assumptions for today’s infrastructure investment to produce the modeled return.
Change those assumptions and the result changes.
Equipment may last longer than expected. Infrastructure could cost less. Some announced data centers may never be constructed. AI companies might operate at higher margins.
The opposite can also happen.
Chips can become obsolete faster. Electricity can become more expensive. Competition can reduce prices. Companies may need more computing power than current projections assume.
That is why the 9% figure is useful as a stress test rather than a destination.
The AI industry does not simply need millions of people to use AI. Current AI spending assumes that those users, businesses and governments eventually produce enough revenue to support one of the largest technology infrastructure expansions in U.S. history.
AI adoption is already growing quickly.
The unanswered question is whether economic value and industry revenue can grow quickly enough to keep pace with the infrastructure being built to serve it.
That is the calculation behind the next phase of AI spending, and the answer will determine how much of today’s enormous buildout ultimately makes financial sense.
