PwC Puts $31.6 Trillion Price Tag on AI Data Center Build Through 2050

The number that leads PwC's inaugural Global Data Centre Outlook isn't the most important thing in it. That would be what comes after: the finding that AI infrastructure spending, unlike every prior infrastructure build in history, isn't expected to taper.

Released September 2, the report, modeled by Oxford Economics and covering 46 countries and territories, projects $31.6 trillion in cumulative data center capital expenditure through 2050, according to PwC's press release reviewed by ETL Newswire. Annual spend rises from roughly $800 billion this year to $1.1 trillion in 2030 and $1.8 trillion by 2050. The upside scenario, if AI adoption moves faster than the central forecast, reaches nearly $50 trillion.

The structural argument PwC is making is the one worth sitting with. Past infrastructure cycles, railways, electrification, the fiber internet rollout, front-loaded their capital and leveled off once the physical plant was built. AI data centers don't work that way. The report notes that servers, GPUs, and other ICT equipment require replacement every four to six years, meaning the capital cycle resets continuously rather than decaying. PwC projects that ICT equipment will account for 93% of data center investment by 2050, up from 70% today.

That's the architecture decision embedded in the headline number. It's not a construction forecast. It's a hardware refresh forecast.

On geography, the U.S. is projected to capture close to half of cumulative investment, $15.1 trillion, given its central position in the advanced-chip ecosystem, according to PwC's own release. Asia Pacific follows at $8.2 trillion, led by China and India. Europe comes in at $5.6 trillion, the Middle East at $1.1 trillion, and Africa at $255 billion.

But PwC's report doesn't treat those regional splits as settled. The analysis tested two alternative scenarios around trade policy and digital sovereignty. In the downside case, tighter export controls on chips disrupt global supply chains: annual investment falls toward half the central forecast by 2030 before partially recovering, and cumulative spending through 2050 drops to roughly $25.5 trillion, about $6 trillion below the baseline. The sovereignty scenario produces a different result: capital gets redistributed rather than destroyed, as more countries build domestic infrastructure to avoid dependence on foreign supply chains.

The factor that shows up consistently across all three scenarios as the binding variable isn't capital availability. It's power. PwC and the Oxford Economics modeling both flag power as the decisive siting factor in every region. Affordable, reliable, and increasingly low-carbon electricity at the scale data centers require is described in the report as the hardest requirement for most markets to actually meet, harder than financing, harder than permitting, harder than chip procurement.

That framing matters for how you read the regional numbers. Europe's $5.6 trillion share isn't just a function of demand or policy; it's a function of whether enough grid capacity exists to host the load. Same logic applies to markets in Southeast Asia, the Middle East, and Latin America that don't show up prominently in the central scenario but could move substantially under the sovereignty case.

The report is also the first long-range forecast of its kind to extend through 2050, most data center analyses stop at five or ten years. The extended horizon changes what the numbers mean. They're not a demand projection; they're a capital planning framework. Infrastructure investors, utilities, and sovereign wealth funds are the actual audience, and they care less about whether AI revenues justify next quarter's capex than whether the grid, the chip supply chain, and the regulatory environment line up over a 25-year horizon.

PwC's central scenario says they will. The downside scenario says they might not, and names the specific mechanism, chip export controls, by which a $6 trillion hole could open up. That's a cleaner warning than most infrastructure reports bother to write.

Sources cited:
- PwC Global Data Centre Outlook (press release) (https://www.pwc.com/gx/en/news-room/press-releases/2026/global-investment-in-ai-infrastructure.html)
- PwC Global Data Centre Outlook (full report page) (https://www.pwc.com/gx/en/1/services/consulting/technology/data-centre-outlook.html)
- Data Center Frontier (https://www.datacenterfrontier.com/machine-learning/article/55403120/pwc-maps-316-trillion-ai-data-center-buildout-through-2050)
- Quartz (https://qz.com/pwc-data-center-ai-investment-forecast-2050-090226)
- PwC Indonesia press release (https://www.pwc.com/id/en/media-centre/press-release/2026/english/global-data-center-outlook.html)