Global AI Infrastructure Spend Will Hit $31.6 Trillion by 2050 — And Power, Not Chips, Decides Who Gets It

PwC’s Global Data Centre Outlook 2026-50, built with Oxford Economics across 46 countries, puts cumulative global AI infrastructure investment at $31.6 trillion through 2050 — with annual data-centre capex climbing from roughly $800 billion in 2026 to $1.8 trillion by 2050. The US captures nearly half that spend, $15.1 trillion, tied to its position in the advanced-chip ecosystem; Asia Pacific, led by China and India, accounts for $8.2 trillion more. What makes the forecast unusual is its shape: PwC expects the buildout to keep climbing rather than taper the way past infrastructure booms did, because AI chips need replacing every few years — ICT equipment’s share of total investment grows from 70% today to 93% by 2050.

That trajectory tracks a semiconductor market already compounding faster than most planning cycles assume. Gartner’s separate forecast puts 2026 worldwide semiconductor revenue at $1.6 trillion, a 92% jump from 2025’s $809 billion, with memory revenue alone growing from 27% of total semiconductor revenue to 54% inside a single year as DRAM and NAND pricing cycles turn. Gartner’s read is that the AI data-centre ecosystem’s share of that revenue expands from 36.5% in 2026 to over 53% by 2030 — the same compute-replacement cycle PwC is modeling from the capital-spend side.

PwC names power availability, not chip access or connectivity, as the single biggest factor determining where that capital lands. Two stress tests matter for anyone advising on AI capital planning: tighter export controls could cut cumulative investment by nearly a fifth, to $25.5 trillion, while a stronger sovereignty push wouldn’t shrink total spending but would redistribute it toward countries building domestic capacity. Data centres, in other words, are no longer a one-time build — they’re a long-duration, geopolitically exposed asset class, and the chip forecast underneath them is moving on the same multi-year clock.