Third-party data centers officially overtake on-premises enterprise IT workloads

The shift signals a fundamental decline in the viability of the private enterprise server room for high-performance video and AI workloads. As streaming platforms integrate more generative AI for encoding, recommendation, and metadata generation, the dependency on hyperscale infrastructure will deepen, potentially creating a tiered market where only those with cloud-scale partnerships can afford the necessary compute. This move to external providers also shifts the regulatory and environmental burden of carbon reporting from the enterprise to the colocation provider. Watch for shifts in cloud pricing models as providers look to pass through the 6.9% year-over-year rise in electricity costs reported in late 2025.

The transition to third-party infrastructure aligns with broader projections of a massive surge in power demand. Per Goldman Sachs Research (May 2026), U.S. data center power demand is forecast to jump from 31 GW in 2025 to 66 GW by 2027. This rapid expansion is primarily driven by the 'Age of Electricity,' where AI-optimized hyperscale facilities consume as much energy as 100,000 households annually. Consequently, data centers' share of U.S. peak summer power demand is expected to more than double, reaching 8.5% by 2027.

Investment activity is scaling to meet these infrastructure bottlenecks. Per reports from September 2024 and January 2026, Microsoft and BlackRock launched a $30 billion Global AI Infrastructure Investment Partnership, which aims to mobilize up to $100 billion including debt. The fund recently completed a $40 billion acquisition of Aligned Data Centers in July 2026, marking one of the largest infrastructure deals in the sector's history. These capital injections are necessary as technology companies' capex is expected to jump by another 75% in 2026, according to the IEA.

Resource constraints extend beyond power to cooling and water usage. The International Energy Agency's Electricity 2026 report indicates that global data center consumption could reach 1,000 TWh by 2026. This energy intensity is matched by water needs; per the Berkeley Lab (2024), hyperscale and colocation facilities already account for 84% of the 17 billion gallons of water consumed by U.S. data centers annually. With AI queries requiring roughly 10 times the electricity of conventional searches, infrastructure providers are pivoting toward nuclear power agreements and automated Kubernetes cost optimization to maintain operational efficiency, while AI infrastructure spending continues to climb to address these bottlenecks. Some firms are even looking to slash enterprise AI infrastructure costs by moving to in-house models.