PwC Forecasts $31.6 Trillion Global AI Data Center Capex
Long-range outlook projects annual spend surging to $1.8 trillion as chip upgrades eclipse physical construction costs.
PwC's Global Data Centre Outlook forecasts that cumulative global capital expenditure on data centers will reach $31.6 trillion by 2050, with a potential upside of $50 trillion if AI adoption accelerates. The report highlights a fundamental structural shift in infrastructure economics, where recurring chip and hardware refreshes every four to six years eclipse traditional physical construction spending.
Key details
Modeling data center capital allocation across 46 countries in partnership with Oxford Economics, PwC projects annual global data center capex to expand from $800 billion in 2026 to $1.8 trillion by 2050. The United States is expected to capture $15.1 trillion—nearly half of all global capital—while the Asia-Pacific region accounts for $8.2 trillion, led by expansion in China and India.
Unlike traditional civil infrastructure projects that require large upfront capital followed by declining maintenance costs, the AI buildout creates a continuous expenditure loop. The share of capital allocated to ICT equipment—including GPUs, accelerators, servers, and networking hardware—is projected to rise from 70% of total data center spending today to 93% by 2050. Consequently, physical building shells designed for 20-year operational lifespans will house silicon hardware that requires replacement every four to six years.
PwC identifies electrical power availability as the single most critical factor dictating where capital lands globally, outweighing connectivity, policy stability, and permitting speed. However, trade policy risks could alter the market map: stricter chip export controls could reduce cumulative spending by $6 trillion to $25.5 trillion by 2050, while national digital sovereignty mandates could redirect investment toward local infrastructure markets.
Why this matters
The forecast demonstrates that AI infrastructure costs are shifting permanently from fixed real estate development to perpetual technology capital cycles. Financiers and operators face growing friction when underwriting 20-year debt facilities against hardware that depreciates within five years. Furthermore, as ICT hardware commands 93% of capital deployment, energy grid capacity and power procurement strategy become the primary determinants of data center asset valuation.
Context
This landmark long-range forecast comes as hyperscalers and cloud providers ramp up capital expenditure to historic levels, with top tech firms projected to spend over $700 billion on AI infrastructure in 2026 alone. Across North America and Europe, grid operators are struggling to keep pace with gigawatt-scale power requests, leading developers to explore behind-the-meter natural gas generation, nuclear microreactors, and custom debt financing platforms backed by major asset managers to support recurring GPU deployments.
Risks and open questions
A major vulnerability in the $31.6 trillion projection is whether regional power grids can accommodate sustained annual capex growth without causing widespread electricity rate inflation or severe local grid instability. Additionally, if natural gas prices experience volatility or carbon tax regulations expand, the operational cost of powering short-lived GPU clusters could severely squeeze developer margins and force financial restructuring.
What happens next
Investment firms and hyperscalers are expected to accelerate the creation of specialized financing structures that decouple GPU asset equipment leasing from physical property debt. Utility regulators and municipal governments will also come under increasing pressure to establish dedicated large-load utility tariffs and power reservation rules to ensure hyperscale expansion does not compromise grid reliability or shift infrastructure costs onto residential ratepayers.
Source: Capacity Media Published on AI Usage Global, author: AUG Bot
