$31.6 trillion. That’s roughly what the entire US economy produces in a single year. According to PwC’s Global Data Centre Outlook (2026–2050), co-modeled with Oxford Economics across 46 countries, it’s also what the world will spend building AI infrastructure between now and mid-century. Every AI assistant running on a consumer device, every cloud copilot embedded in a productivity app, every generative tool powering today’s workflows — all of it traces back to this. And unlike every infrastructure boom before it, this one doesn’t taper off. It compounds.
The Bill That Never Peaks
The spending figures behind this cycle are unlike anything infrastructure planners have modeled before.
- Global data center capex hits $31.6 trillion in PwC’s central scenario; the upside case approaches $50 trillion, the downside lands around $22 trillion
- Annual spend runs roughly $800 billion in 2026, climbs to $1.1 trillion by 2030, and hits $1.8 trillion by 2050
- The US captures nearly half — $15.1 trillion, about 48% of the global total
- APAC (driven by China and India) takes $8.2 trillion; Europe $5.6 trillion; Middle East $1.1 trillion; Africa $255 billion
- ICT equipment — GPUs, servers, networking — grows from roughly 70% of capex today to 93% by 2050; the buildings become almost incidental
Railways, electrification, the internet — all front-loaded spending that eventually tapered. AI data centers work differently. PwC puts it plainly: “Railways. Electrification. The internet. Each required enormous amounts of capital and defined an era. The AI infrastructure cycle underway dwarfs all three.”
Hardware — AI accelerators, including Nvidia-class GPUs — obsoletes every 4–6 years and gets replaced inside facilities that still have decades of useful life. Per PwC, every $1 spent on construction triggers roughly $12 in follow-on ICT equipment over the asset’s lifetime. That’s not a construction boom. It’s a subscription that auto-renews — except the bill grows every cycle instead of staying flat.
The Friction Nobody’s Advertising
Real constraints around power, chips, and community opposition could determine how much of this spending actually materializes.
Power availability is PwC’s identified primary bottleneck. Reliable, affordable, low-carbon electricity at scale determines where data centers actually get built. Chip supply chain disruptions could cut total global investment by nearly 20%, per PwC’s downside scenario.
Meanwhile, Data Center Watch tracked at least 75 projects worth roughly $130 billion blocked or delayed by local opposition in just Q1 of this year — land use, water consumption, and community anxiety about AI’s broader social footprint all feature in the resistance.
The $22T–$50T scenario range narrows or widens based on three variables:
- Grid policy
- Semiconductor access
- Community approval
PwC’s figures are a baseline for planners — not a consensus forecast, as Bloomberg’s coverage notes.
What flows downstream: more capable cloud AI services over time, but sustained pressure on energy infrastructure and, eventually, the subscription pricing on AI-backed services consumers use daily. The most expensive infrastructure build in history is already underway. It’s just getting started.
