Global data center spending is on track to hit a staggering $31.6 trillion by 2050, driven by the AI buildout, according to a new report from PwC and Oxford Economics. The number is so big it barely registers: it is roughly the size of the entire U.S. economy in a single year. But the headline figure only tells part of the story. What matters for decision makers is how the money flows, what it gets spent on, and where the real bottlenecks sit.
Before this week, the most ambitious published estimates for AI infrastructure spending topped out in the tens of trillions over a few decades. PwC's Global Data Centre Outlook narrows the fog with a concrete, country-by-country model, and the practical question for anyone planning around AI is how to actually use it. You do not need to digest 46 markets at once. You need the handful of numbers that drive every downstream decision.
Break Down the Forecast Before You Panic
The projection runs from 2026 through 2050. The central scenario lands at $31.6 trillion, with an upside case approaching $50 trillion and a downside case around $22 trillion. Before acting on any single figure, anchor yourself in the pieces that actually matter. Here is the short version of how to read it.
- Annual pace: about $800 billion in 2026, climbing to $1.1 trillion by 2030 and $1.8 trillion by 2050. It does not taper; it compounds.
- The United States captures nearly half the total, roughly $15.1 trillion or about 48 percent of global spending.
- APAC, driven by China and India, accounts for $8.2 trillion; Europe comes next at $5.6 trillion, then the Middle East at $1.1 trillion and Africa at $255 billion.
The biggest surprise is what the money buys. ICT equipment, meaning GPUs, servers and networking, grows from roughly 70 percent of capex today to 93 percent by 2050. The physical buildings become almost incidental next to the silicon inside them.
Why This Cycle Refuses to Slow Down
Earlier infrastructure booms did not behave like this. Railways, electrification and the internet all front-loaded their spending, then tapered off once the network existed. PwC makes the contrast explicit: each of those eras required enormous capital and defined its time, yet the AI infrastructure cycle dwarfs all three combined.
The engine behind this is hardware turnover. AI accelerators, including Nvidia-class GPUs, obsolete every four to six years and get swapped out inside facilities that still have decades of useful life left.
For every dollar spent on construction, the report estimates roughly twelve dollars in follow-on ICT equipment follows over the asset's lifetime. Read that again: it is not a construction boom. It is a subscription that auto-renews, and the bill grows every cycle.
Track the Three Things That Could Derail It
No forecast is guaranteed, and PwC is candid about the risks. If you are planning around this number, keep these three variables on your watch list.
- Power availability. PwC flags it as the primary bottleneck. Reliable, affordable, low-carbon electricity at scale decides where data centers actually get built.
- Chip supply chains. Disruptions here could cut total global investment by nearly 20 percent, per the report's downside scenario. The whole cycle is held together by a fragile silicon supply chain.
- Local opposition. Data Center Watch tracked at least 75 projects worth roughly $130 billion as blocked or delayed by community pushback in the first quarter of this year alone, over land use and water concerns.
The practical takeaway is to treat PwC's figures as a working baseline rather than a consensus forecast. The $22 trillion to $50 trillion spread is wide for a reason, because the outcome genuinely depends on decisions that have not been made yet.
For anyone building, investing in or staffing around AI infrastructure, the lesson is straightforward: bet less on the $31.6 trillion headline and more on the pace of annual spend, the refresh cycle and the power supply that actually feeds it. What flows downstream is a more capable cloud AI layer over time, sustained pressure on energy infrastructure and, eventually, the pricing of AI services consumers use daily.
The most expensive infrastructure build in history is already underway, and it is just getting started. Understanding the parts that make up the whole is the only sane way to navigate it.
Comments