The $31.6 Trillion Question: PwC Sizes the Biggest Capital Mobilization in History

Every era of technology has its defining number. For the AI age, PwC has just put one on the table, and it is almost too large to process: US$31.6 trillion — the amount of capital the consulting giant expects to pour into data centres globally between now and 2050 to feed the world’s appetite for artificial intelligence.

Published this week as part of PwC’s Global Data Centre Outlook 2026, the forecast describes what it calls one of the largest mobilizations of capital in modern economic history — a 25-year construction cycle, bigger than the railway booms of the nineteenth century and the interstate highway program of the twentieth, scaled to the demands of machine intelligence.

The headline numbers

The base-case trajectory is steep by any historical standard:

• $31.6 trillion in cumulative global data-centre capital expenditure through 2050
• Annual capex of roughly $800 billion in 2026, rising to $1.1 trillion by 2030 and $1.8 trillion per year by 2050
• The United States captures nearly half — about 48% — of total global spending
• Equipment (chips, servers, and networking) already accounts for 70% of data-centre costs today; by 2050 the modelling puts that share at 93%
• A plausible upside scenario of nearly $50 trillion if AI adoption runs hotter than expected

Put differently: the world is currently spending about $800 billion a year on the physical substrate of AI. Within five years PwC expects that to breach a trillion annually. By mid-century, single-year spending could approach $1.8 trillion — more than twice today’s entire global semiconductor industry revenue.

Why equipment eats the budget

The shift from 70% to 93% equipment share is the most quietly radical number in the report. It reflects a structural change in what a data centre is.

Legacy cloud facilities were buildings first — real estate, cooling, land, power contracts. The AI-era data centre inverts that logic. GPU accelerators and AI-specific silicon now dominate the bill of materials, and they depreciate on a far faster clock than concrete. A hyperscale AI campus can cost tens of billions of dollars, with the majority flowing straight to chip suppliers — a dynamic that explains why NVIDIA’s data-centre revenue has grown the way it has, and why custom-silicon programs at Google, Amazon, Meta, and Anthropic have become strategic priorities rather than cost projects.

The implication for the supply chain is enormous. If equipment consistently absorbs 90%+ of a $1.8-trillion annual budget, the downstream market for accelerators, memory, and networking approaches $1.6 trillion a year by 2050 — numbers that reframe the entire investment case for advanced packaging, HBM memory, and foundry capacity.

Geography: the U.S. takes half

PwC’s allocation of nearly half the total spend to the United States is a bet on policy continuity as much as economics. The U.S. hosts the deepest concentration of hyperscalers, the largest AI-lab customers, and a power grid that — however strained — remains more amenable to gigawatt-scale interconnects than most alternatives.

The remaining half splits across a widening map: Europe racing to reconcile AI ambitions with energy reality, the Gulf states converting oil wealth into sovereign AI capacity, and Asia’s build-out led by China’s parallel ecosystem and Japan, India, and Southeast Asia’s emergent markets. The report’s framing of “power supply and politics” as the binding constraints is telling — PwC explicitly conditions its forecast on grids keeping pace and regulatory regimes staying stable. Neither is guaranteed.

The financing question nobody can dodge

A number of this size inevitably raises the sustainability question. Several of this year’s defining AI-infrastructure stories — ByteDance’s record $29.6 billion loan, Broadcom’s mounting AI-linked debt issuances, Jane Street’s $1.5 billion bet on Fluidstack — have already shown capital markets straining to fund the build-out through increasingly creative instruments. PwC’s forecast effectively asserts that the money will be found, year after year, for a quarter century.

That is the optimistic read. The pessimistic one is that $31.6 trillion represents demand if AI economics keep working — if inference revenue, agentic workloads, and enterprise adoption continue to compound fast enough to justify the next tranche of capex. The report’s own $50-trillion upside and the existence of downside scenarios acknowledge the range of outcomes. Historical precedent cuts both ways: railway manias built real infrastructure and still bankrupted their financiers.

What it means

For the industry, PwC’s Outlook functions as a formalization of what the last two years of announcements already implied. The question is no longer whether AI infrastructure is a big capex cycle, but whether it is the capex cycle — the defining allocation of global capital for a generation. On PwC’s numbers, it is: $31.6 trillion over 25 years averages $1.26 trillion per year, sustained, in a single asset class.

For everyone else, the report is a preview of the physical world the AI economy is building: continents dotted with gigawatt campuses, power markets reshaped around compute, and a chip industry that becomes, by revenue, plausibly the largest manufacturing sector on Earth.

The number is huge. The bet is explicit. And for the next 25 years, it is the yardstick against which every AI infrastructure announcement will be measured.