Hyperscaler capex spending reached $260 billion in 2025 as Amazon, Microsoft, Google, and Meta raced to secure GPU capacity, data centre land, and power contracts. You can track exactly where those billions went and what the spending trajectory means for AI infrastructure through 2027.
Total Hyperscaler Capex Spending in 2025
The four largest hyperscalers collectively spent $260 billion in capital expenditure during 2025, a 52% increase over 2024. Amazon led with $86 billion, followed by Microsoft at $80 billion, Google at $52 billion, and Meta at $42 billion. Each company confirmed these figures in quarterly earnings calls throughout the year, and every CEO signalled that 2026 budgets would grow further.
This level of spending is unprecedented. The entire global semiconductor industry generated $627 billion in revenue in 2024. Four companies alone now spend at a rate approaching half that figure on infrastructure. You are watching the largest private capital deployment cycle in history.
Where the Money Goes: GPUs, Data Centres, and Power
Roughly 60% of hyperscaler capex flows directly to GPU and AI accelerator purchases. NVIDIA captured the majority of that spend, with its H100 and Blackwell platforms dominating order books. Amazon and Google also invested heavily in custom silicon, with AWS deploying Trainium2 chips and Google scaling its TPU v5p clusters.
Data centre construction accounts for approximately 25% of total capex. Microsoft announced $9.7 billion in new campus deals across Sweden, Japan, and the United States. Amazon committed $20 billion to data centre builds in Virginia, Ohio, and international markets. The remaining 15% covers networking, cooling, and long-term power purchase agreements.
Power Procurement Is the New Bottleneck
Power contracts have become the single largest constraint on Microsoft AI investment and its competitors. A 500 MW campus requires enough electricity for 400,000 homes. Hyperscalers signed over 15 GW of new power agreements in 2025, spanning nuclear, gas, and renewables. Microsoft signed a 20-year deal with Constellation Energy to restart Three Mile Island Unit 1 specifically for data centre power.
2026-2027 AI Infrastructure Spending Forecast
Wall Street consensus projects combined hyperscaler capex reaching $350 billion to $390 billion in 2026. Microsoft guided for $80 billion in the first half of fiscal 2026 alone, putting it on pace for $150 billion annualised. Amazon indicated sustained investment at or above 2025 levels, suggesting $90 billion or more. Google and Meta both signalled double-digit capex growth.
The AI infrastructure spending forecast through 2027 depends on three variables: AI revenue growth rates, GPU supply from NVIDIA and AMD, and power grid availability. If AI cloud revenue grows 40% annually as current trends suggest, you should expect capex to reach $450 billion to $500 billion by 2027. If revenue growth decelerates below 25%, spending could plateau at $350 billion.
What Hyperscaler Spending Means for the AI Supply Chain
Every dollar of hyperscaler capex creates a multiplier across the supply chain. GPU makers, memory manufacturers, networking vendors, and power utilities all benefit. When you see Microsoft commit $80 billion, that translates to purchase orders at Foxconn, Celestica, Vertiv, and Eaton within 12 to 18 months.
FAQ
How much did hyperscalers spend on AI infrastructure in 2025?
The four largest hyperscalers spent approximately $260 billion in total capex in 2025. Amazon led at $86 billion, Microsoft at $80 billion, Google at $52 billion, and Meta at $42 billion. Over 60% went directly to AI infrastructure including GPUs, data centres, and power.
Which hyperscaler spends the most on AI capex?
Amazon spent the most at $86 billion in 2025, but Microsoft is projected to overtake it in fiscal 2026 with a run rate exceeding $150 billion annualised. Both companies allocate the majority of capex to AI-enabled data centre infrastructure.
Will hyperscaler capex spending continue to grow in 2026?
Yes. Analyst consensus projects combined capex reaching $350 billion to $390 billion in 2026, a 35% to 50% increase over 2025. Every major hyperscaler has signalled continued investment, driven by AI workload demand that outpaces available GPU and data centre capacity.