Capacity Frenzy: Key Considerations in Procurement Agreements

Compute Capacity Crunch

The rapid proliferation of artificial intelligence applications, particularly generative AI and high-performance computing, has dramatically increased the need for massive compute resources. This demand outstrips the ability of traditional cloud providers and emerging specialized infrastructure players to supply capacity quickly enough, leading to what industry observers describe as a 'capacity frenzy' in procurement agreements.

Major tech companies are committing unprecedented capital—Amazon, Microsoft, Google, and Meta collectively exceeded $200 billion in capex in 2024, with projections for even higher spending in 2025 and beyond—to build out data centers and secure compute resources. Yet supply remains tight: cloud contracts increasingly feature ambiguities around guaranteed capacity, geographic availability, and the potential for providers to reallocate resources, leaving buyers vulnerable to shortfalls that could derail AI project timelines.

The stakes extend beyond individual deals. New entrants flood the market but often require creative financing structures to deploy capacity, while established providers grapple with how to allocate limited resources amid competing demands. This environment heightens risks around service levels, payment models, and risk allocation in contracts, as buyers seek protections against non-delivery in a constrained market.

Broader pressures compound the issue. Data center expansion faces parallel bottlenecks in power supply, with AI facilities driving electricity demand growth that strains grids and prompts alternative procurement strategies like direct power deals or on-site generation. These dynamics create trade-offs: pursuing dedicated capacity may offer reliability but at higher upfront costs and complexity, while relying on shared cloud resources risks availability in peak demand periods.