Cloud Infrastructure Spending Hit $106.9B in Q4 2025 as AI Workloads Force Capacity Limits

Capacity Constraints Now Limit Cloud Growth

Global cloud infrastructure revenue hit $106.9 billion in Q4 2025, the first time quarterly spending exceeded $100 billion. The $7.5 billion quarter-over-quarter increase represents the largest sequential jump on record and brings full-year 2025 spending above a $400 billion run rate. The driver: enterprise AI workloads shifting budgets from traditional storage and virtual machines to high-compute inference for chatbots and enterprise tooling.

The growth is creating a new problem for enterprise buyers. Microsoft CEO Satya Nadella noted that Azure growth is now capacity-constrained, not demand-constrained. This marks a fundamental shift in cloud economics. For the first time in the hyperscale era, the bottleneck is physical infrastructure — specifically GPU availability and power — rather than sales or adoption. Buyers who need AI compute now face pressure to sign multi-year commitments to secure access, increasing lock-in risk at a time when workload portability matters more than ever.

AWS, Azure, and Google Cement Market Dominance

The Big Three — AWS, Microsoft Azure, and Google Cloud — now control 63% of the market, up from 61-62% two years ago. AWS holds 29% share with approximately $31 billion in quarterly revenue and grew 20% year-over-year, its first time hitting that growth rate since Q4 2022. Azure captured 20% share at roughly $21 billion, growing 39%. Google Cloud holds 13% at around $14 billion, accelerating 34% and gaining two points quarter-over-quarter.

Capital expenditure tells the capacity story. Microsoft spent $35 billion in the quarter, AWS $34 billion, and Google $24 billion. These figures dwarf spending by neoclouds like CoreWeave, Crusoe, Nebius, and Lambda, which are gaining traction by specializing in AI workloads but lack the capital base to compete at hyperscale. Oracle is inching higher through AI-specific contracts, while Alibaba and Salesforce are losing share due to slower growth. IBM remains static as it pivots strategy.

AWS CEO Matt Garman projected the company will reach $600 billion in annual revenue by 2036, double prior internal estimates, with infrastructure spending potentially exceeding $200 billion total. The forecast assumes AI inference becomes embedded in enterprise operations rather than remaining a burst workload for model training. That changes the buying calculus: sustained inference requires different capacity planning than episodic training runs.

Multi-Year Commitments Replace Spot Pricing

The shift to capacity-constrained cloud infrastructure changes how enterprises negotiate. Buyers who once optimized for cost and geographic distribution now select vendors primarily on AI capacity availability. Multi-year GPU commitments are becoming standard, reducing flexibility and increasing switching costs. This favors buyers willing to commit early but penalizes those who need workload portability or want to evaluate multiple vendors.

The hyperscalers are building custom silicon to reduce reliance on Nvidia. AWS uses its own Trainium and Inferentia chips, Microsoft develops Maia, and Google deploys TPUs. This vertical integration benefits the Big Three against neoclouds like CoreWeave, which depend entirely on Nvidia GPUs and face allocation constraints. For enterprise buyers, vendor lock-in now extends beyond APIs and data gravity to the chip architecture running their models.

Security Budgets Rise Alongside AI Infrastructure

Upwind Security raised $250 million to expand its cloud-native security platform, reflecting heightened focus on protecting AI infrastructure. The funding positions Upwind against incumbents like Palo Alto Networks Prisma Cloud, CrowdStrike Falcon, and Sysdig. With 89% of enterprises using multi-cloud environments and 80% running hybrid setups, security complexity is compounding as AI workloads add high-density GPU clusters and new attack surfaces.

Gartner forecasts 90% hybrid cloud adoption by 2027, driving security budgets higher. Buyers are prioritizing vendors with integrated security to avoid outages or breaches in environments where a single compromised GPU rack can cost six figures in lost compute time. The economics of AI infrastructure make downtime far more expensive than in traditional cloud workloads, shifting security from a compliance checkbox to a core operational requirement.

What to Watch

Capacity constraints will force more enterprises into long-term contracts in 2026, reducing leverage in vendor negotiations. Watch whether neoclouds can secure enough GPU supply to remain competitive or if capital intensity consolidates the market further around the Big Three. AWS's $600 billion revenue target assumes inference workloads grow faster than training, which depends on whether enterprises move AI from experimentation to production. If buyers delay production deployments, the capacity crunch eases and pricing pressure returns. Security spending will track infrastructure growth — expect consolidation among cloud security vendors as enterprises seek platforms that span multi-cloud and on-premises environments without adding operational complexity.