PwC Puts $31.6 Trillion Price Tag on AI Data Center Build Through 2050, Flags Power as Binding Constraint

The number that will dominate infrastructure conversations this fall is $31.6 trillion. That's how much PwC projects will flow into data center construction and equipment globally through 2050, according to its inaugural Global Data Centre Outlook, released September 2.

The report, which PwC commissioned Oxford Economics to model across 46 countries and territories, isn't a market-size estimate dressed up as analysis. It's a capital expenditure forecast, and the distinction matters. It counts what actually gets built and refreshed, not what gets announced or funded. According to the PwC press release reviewed for this piece, annual data center capex is expected to climb from roughly $800 billion in 2026 to $1.8 trillion per year by 2050.

That trajectory is unusual for infrastructure. Most big build cycles peak and taper. This one doesn't, and PwC's explanation for why is the most useful part of the report. Unlike railways or fiber optic networks, AI infrastructure doesn't stay put once it's built. Servers, GPUs and other compute equipment require replacement on four-to-six-year cycles, according to Data Center Frontier's analysis of the report. The result isn't a one-time buildout. It's a recurring capex obligation that compounds as the installed base grows.

The geographic split is lopsided. The U.S. is projected to capture 48 percent of cumulative investment, roughly $15.1 trillion, 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, with sovereign AI strategies accelerating spend there and in the Middle East.

But the headline figure and the regional breakdown aren't where the real story sits. PwC identifies power availability as what it calls the binding constraint in every region it examined. Grid connections take four to ten years in many markets, according to reporting by TheStreet on the PwC findings, while the data centers they need to feed are built in two to three years. That mismatch is already stalling projects, and no amount of capital closes it quickly.

The report also models what disrupted chip trade flows would do to the forecast. A scenario with constrained semiconductor access cuts projected global investment by nearly 20 percent, according to PwC's release. That's not a tail risk number. It's a reminder that the infrastructure supercycle runs on Nvidia, TSMC and a handful of packaging facilities, and that export controls or supply chain fractures hit the whole stack, not just the chip layer.

The private capital markets have already absorbed this logic. On July 21, a consortium comprising the AI Infrastructure Partnership, Abu Dhabi's MGX and BlackRock's Global Infrastructure Partners closed a $40 billion acquisition of Aligned Data Centers, the largest data center deal on record, according to reporting by Data Center Dynamics and confirmed in Aligned's own press statement. The buyers then committed an additional $5 billion in growth capital to expand the company's AI-ready capacity. Aligned operates 51 campuses with more than 6.4 gigawatts of operational and planned capacity.

That deal is the clearest evidence yet of how the PwC thesis is being priced in real time. The investors buying Aligned aren't betting on a short AI spending spike. They're betting on the recurring refresh cycle PwC describes, and on a platform that already has the power contracts, the permits and the physical footprint that take years to replicate.

What PwC's outlook doesn't resolve is whether the grid can actually keep up. The report is honest about the range of uncertainty: plausible cumulative investment runs from roughly $22 trillion to nearly $50 trillion, depending primarily on how fast AI adoption moves. That's a $28 trillion swing, and the variable it hangs on most isn't model capability or enterprise demand. It's whether utilities and regulators can site, permit and energize infrastructure fast enough to serve projects that are already in the ground.

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)
- Data Center Frontier (https://www.datacenterfrontier.com/machine-learning/article/55403120/pwc-maps-316-trillion-ai-data-center-buildout-through-2050)
- TheStreet (https://www.thestreet.com/technology/pwc-ai-data-center-capex-forecast-power-constraint)
- Data Center Dynamics (https://www.datacenterdynamics.com/en/news/aligned-data-centers-sold-to-blackrock-and-mgx-in-record-breaking-40bn-deal/)
- Aligned Data Centers (press statement) (https://aligneddc.com/press-release/aip-mgx-and-blackrocks-gip-close-acquisition-of-aligned-data-centers/)