- AI infrastructure spending could reach $1.5 trillion a year by 2031.
- Bain estimates the industry could need about $6 trillion in annual revenue to support that spending.
- New revenue is expected to come mainly from search, advertising, autonomous systems, physical AI and enterprise software.
- Rapidly growing data center costs, electricity needs and computing demand are increasing the financial pressure on AI companies.
The artificial intelligence boom is entering a more expensive phase, and the biggest challenge may no longer be building better models. It could be finding enough business to pay for the enormous infrastructure required to run them.
A new analysis from Bain & Company suggests the AI industry could need as much as $6 trillion in annual revenue by 2031 to support the level of infrastructure investment expected over the next several years.
That figure is based on a projected $1.5 trillion in annual spending on AI related infrastructure by 2031. The spending would cover data centers, processors, memory, networking equipment and other systems needed to keep increasingly powerful AI services running.
The calculation assumes infrastructure costs account for roughly 25 percent of industry revenue. Bain describes that assumption as ambitious, but broadly consistent with the spending patterns seen in the cloud computing industry.
The numbers highlight a growing question for AI companies. Massive amounts of capital are being committed to computing capacity today, but the commercial returns needed to justify that investment still have to materialize.
AI needs more than today’s products to close the gap
Search, advertising, autonomous systems and physical AI are expected to provide the largest share of the potential new revenue.
Bain estimates these emerging areas could eventually generate about $4.2 trillion in annual commercial value. The challenge is that many of these businesses are still developing and, in some cases, are only beginning to take shape.
Autonomous systems are one example. AI powered vehicles, robotics and other physical systems could create large new markets, but widespread adoption will take time. The same applies to AI products that could transform search and advertising.
Enterprise software is another major opportunity.
Bain estimates that enterprise productivity applications could contribute between $1 trillion and $1.4 trillion in annual revenue. Companies are already using AI for software development, sales, marketing, customer service and IT operations, but the industry would need these applications to expand dramatically to reach the projected figures.
Consumer subscriptions and advertising are expected to make a much smaller contribution. Bain puts their potential at roughly $200 billion to $400 billion, despite the growing number of people using AI services.
That difference matters because consumer AI has attracted enormous attention, but converting millions or even billions of users into sustainable revenue is a different challenge.
Data centers are becoming a financial problem
The cost of building the infrastructure itself is becoming another pressure point.
Bain estimates that the size and cost of data centers have been increasing at roughly double rates over periods of around 12 to 16 months globally. That does not mean every individual project literally doubles in price every year, but it illustrates how quickly the scale of AI infrastructure is changing.
The numbers surrounding large AI facilities show just how significant the shift has become.
Epoch AI estimates that Meta’s Prometheus data center project in Ohio had around 600MW of capacity and could cost approximately $24 billion based on its 2025 estimates.
The proposed expansion could take the facility to 2GW and around $80 billion by 2027. Under longer term projections, capacity could eventually reach several gigawatts, with total spending potentially climbing into the hundreds of billions of dollars.
These projects are not simply about buying more GPUs.
Large AI facilities require electricity generation, transmission infrastructure, grid connections, advanced semiconductor hardware, cooling systems, networking equipment and workers capable of operating highly specialized facilities.
That creates a chain reaction across the technology and energy industries.
A company can have enough money to order computing equipment and still face delays because the local grid cannot supply the required electricity. It may also struggle to secure suitable land, cooling capacity or network infrastructure.
The real test is whether AI can create enough economic value
The AI industry has already demonstrated that there is enormous demand for computing power. The more difficult question is whether that demand will translate into enough revenue.
AI companies are investing heavily before many of the expected applications have reached maturity. That strategy makes sense if future products become large enough to support the infrastructure being built today.
But if adoption is slower than expected, companies could find themselves carrying extremely expensive data centers and computing equipment without enough revenue to generate acceptable returns.
The pressure is particularly important for companies operating at the frontier of AI. Training increasingly capable models requires enormous computing resources, while running those models for millions of users creates another substantial operating expense.
That means the industry faces two challenges at once. It needs to make AI more useful and it needs to discover business models capable of turning that usefulness into recurring revenue.
The potential markets are certainly large. Drug discovery, energy systems, healthcare, autonomous machines and enterprise software could all create significant economic value if AI becomes deeply embedded in those industries.
However, those opportunities cannot be treated as guaranteed revenue.
The next stage of the AI boom will therefore be measured by more than model performance or the number of GPUs installed. Companies will increasingly have to demonstrate that their technology can support the enormous physical infrastructure behind it.
The central question is no longer simply how much AI can scale.
It is whether the economy can generate enough value to pay for that scale.
4 plain point summary
