The AI infrastructure spending forecast projects global investment to grow from $340 billion in 2025 to over $1 trillion by 2030. You need to understand exactly where hyperscalers, enterprises, and sovereign funds are allocating capital across GPUs, data centres, power, and networking to position your portfolio ahead of this unprecedented spending wave.
Total AI Infrastructure Spending by Year: 2025 to 2030
Analyst estimates from Goldman Sachs, Morgan Stanley, and Gartner converge on a consistent trajectory. Global AI infrastructure capex hit $340 billion in 2025, driven by hyperscaler capex spending that exceeded $260 billion from the four largest cloud providers alone. By 2027, total spending is projected to reach $520 billion to $580 billion. By 2030, the consensus range lands between $950 billion and $1.2 trillion annually.
These numbers encompass GPU and AI accelerator purchases, data centre construction, power procurement, networking equipment, and cooling infrastructure. Enterprise spending outside the hyperscalers adds $80 billion to $120 billion annually, and sovereign AI programmes from the UAE, Saudi Arabia, France, and India contribute another $30 billion to $50 billion starting in 2027.
GPUs and AI Accelerators: 40% of Total Spend
Semiconductors remain the single largest spending category. NVIDIA captured roughly 80% of AI accelerator revenue in 2025, with its Blackwell B200 and GB200 platforms dominating hyperscaler orders. AMD’s MI300X gained traction at Microsoft and Meta, reaching approximately 8% market share. Custom silicon from Google (TPU v6), Amazon (Trainium3), and Microsoft (Maia 2) is projected to account for 15% of total accelerator spending by 2028.
You should expect GPU spending to grow from $136 billion in 2025 to $400 billion by 2030. The shift from training-only clusters to inference-heavy deployments will reshape procurement patterns. Inference workloads favour lower-cost, higher-throughput chips, which benefits AMD and custom silicon providers as the market matures.
Data Centre Construction: The $200 Billion Bottleneck
Data centre capital expenditure is the second largest category, accounting for roughly 25% of total AI infrastructure spending. Microsoft committed over $80 billion in a single fiscal year, with campuses spanning 40 countries. Amazon announced $100 billion in data centre builds for 2026 alone. Equinix, Digital Realty, and newer entrants like CoreWeave are racing to deliver colocation capacity.
Construction costs per megawatt have risen 35% since 2023 due to labour shortages, permitting delays, and the shift to liquid-cooled, high-density racks. You are looking at $200 billion in annual data centre construction by 2028, up from $85 billion in 2025.
Power and Cooling: The Hidden Cost Multiplier
Power procurement now represents 15% of total AI infrastructure spending and is growing faster than any other category. A single GB200 NVL72 rack consumes 120 kW. Hyperscalers have signed over $50 billion in long-term power purchase agreements since 2024, including nuclear, natural gas, and renewable contracts. Microsoft’s Three Mile Island deal and Amazon’s Talen Energy acquisition signal that power security is now a strategic priority, not just an operational expense.
Cooling technology spending is expected to triple from $8 billion in 2025 to $24 billion by 2029, as direct liquid cooling becomes standard for next-generation GPU clusters.
Where You Should Focus Your Investment Exposure
The spending forecast reveals three durable investment themes. First, GPU makers and their supply chain partners benefit from the largest single spending category. NVIDIA, TSMC, and ASML sit at the centre of this flow. Second, data centre REITs and construction firms capture the physical buildout. Third, power infrastructure companies benefit from the fastest-growing cost category. The best AI infrastructure stocks give you direct exposure to all three themes.
If you want diversified access to the entire spending stack, consider an AI infrastructure ETF that bundles semiconductor, cloud, and power holdings into a single position.
Frequently Asked Questions
How much will AI infrastructure spending total by 2030?
Analyst consensus projects global AI infrastructure spending to reach $950 billion to $1.2 trillion annually by 2030, up from $340 billion in 2025. This includes GPUs, data centres, power, networking, and cooling across hyperscalers, enterprises, and sovereign programmes.
Which companies spend the most on AI infrastructure?
Amazon, Microsoft, Google, and Meta collectively spent $260 billion on AI infrastructure in 2025. Amazon led at $86 billion, followed by Microsoft at $80 billion, Google at $52 billion, and Meta at $42 billion. All four have signalled higher budgets through 2027.
What is the fastest-growing category of AI infrastructure spending?
Power procurement is growing faster than any other AI infrastructure spending category. Long-term power purchase agreements exceeded $50 billion since 2024, and annual power-related capex is projected to reach $150 billion by 2030 as AI data centres scale to multi-gigawatt campuses.