AI infrastructure expansion faces a revenue test: the assumptions behind a $31.6 trillion forecast

The expansion of artificial intelligence infrastructure is facing a more direct test: whether revenue from future applications can arrive in time to support the capital being invested today. A Reuters analysis on October 3 brought together several studies, highlighting a gap between technological potential, corporate valuations and actual productivity gains that still needs to be tested. Understanding this wave of investment begins with distinguishing long-term capital spending, future annual revenue and employment changes already observed.¹

Figure: AI-generated conceptual illustration: data centers, power facilities and infrastructure under construction.

$31.6 trillion is a projected path

PwC's study, published on September 2 with modeling by Oxford Economics, covers 46 countries and territories. Its baseline scenario projects about $31.6 trillion in cumulative global data-center capital spending between 2026 and 2050, with annual spending rising from around $800 billion in 2026 to $1.1 trillion in 2030 and $1.8 trillion in 2050. The figures include buildings and information and communications equipment, expressed in real US dollars at 2025 exchange rates.²

This is neither spending that has already occurred nor the total value of orders companies have signed. The report assumes that equipment is replaced every four to six years; keeping data centers operating therefore means repeated investment in servers and accelerator chips. Different rates of AI adoption produce a cumulative range of roughly $22 trillion to $50 trillion, and the study does not assign probabilities to the scenarios. Electricity supplies and chip trade could also alter the actual path.²

The revenue gap is measured annually

Bain's annual technology report, released on September 29, sets out another set of conditions from the revenue side: supporting the AI computing demand it projects would require about $6 trillion in annual revenue by 2031, while existing consumer and enterprise AI applications could contribute $1.2 trillion to $1.8 trillion. Even at the upper end of that range, new businesses would still need to fill a $4.2 trillion annual revenue gap. That figure should not be understood as a cumulative revenue shortfall over the next five years.³

Bain's potential sources include new advertising models, autonomous systems such as self-driving vehicles, robotics, and applications in drug development and energy that have yet to scale. These are possible avenues for revenue, not markets whose potential has already been realized. The $31.6 trillion figure describes capital investment over many years, while the $6 trillion figure describes the revenue needed in one year. Their time horizons and scope differ, so dividing one by the other does not produce an investment return.²³

The central question is therefore not only whether models can perform more tasks, but also who captures the additional value, how much becomes supplier revenue and whether that happens fast enough to keep pace with equipment replacement and financing arrangements. Customer cost savings, suppliers' revenue and eventual profits are also different measures. Effective technology does not automatically guarantee that every project's financing structure is sustainable.

Employment signals also require careful qualification

An updated study published by Stanford's Digital Economy Lab on August 12, using ADP payroll data through June 2026, found that employment among 22- to 25-year-olds in occupations highly exposed to AI was about 19% below a comparison path in which it grew in line with that of same-age workers in low-exposure occupations. The study did not find broad job displacement across the economy; the gap primarily reflected reduced hiring of young people rather than increased separations. The authors explicitly state that this is a descriptive association and does not allow the entire gap to be attributed to AI.⁴

The findings suggest that localized hiring contractions can coexist with stable employment overall, and job changes alone cannot establish that macroeconomic productivity has already jumped sharply. Reuters quoted Cambridge economist Diane Coyle's observation that, historically, the productivity effects of major technologies often take a long time to spread.¹

For this expansion, actual paying demand, utilization after delivery and cash flow from ongoing operations will be more informative measures. Long-term forecasts show a possible scale; operating data will gradually reveal which investments can support themselves. Large distant-future figures cannot substitute for actual returns, but the fact that returns have not yet fully emerged does not establish that all infrastructure has lost its value.

Source notes

1. 2026-10-03 · Analysis: AI's race to transform the world before the money runs out · Reuters, republished by CNA ↩ ↩²

2. 2026-09-02 · Where $31.6 trillion of capex flows in the era-defining AI build-out · PwC, Global Data Centre Outlook 2026–50 ↩ ↩² ↩³

3. 2026-09-29 · Global AI market could hit $6 trillion annually by 2031 through unlocking value and innovation – Bain & Co's 7th Global Technology Report · Bain & Company, published by PR Newswire ↩ ↩²

4. 2026-08-12 · No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19% · Stanford Digital Economy Lab; data through June 2026 ↩