AI infrastructure spending to hit $350 billion as power constraints tighten

The massive influx of capital into specialized AI facilities is creating a two-tier infrastructure market where power availability, rather than location, dictates capacity. For streaming providers, this necessitates a shift toward modular edge outposts to maintain low-latency delivery as traditional hubs like Northern Virginia reach electricity limits. As hyperscalers like Google and Oracle lock up gigawatt-scale power rights, smaller streaming platforms may face increased colocation costs or be forced into secondary markets. The industry must now transition from simple server hosting to managing vertically integrated energy microgrids to ensure uptime. Watch for the first phase of Novva Data Centers' 300 MW Arizona campus to go live in 2026 as a benchmark for new power-independent builds.

The race to secure power for AI workloads is intensifying across the data center sector, with multiple operators pursuing onsite generation and modular builds to bypass grid interconnection delays. In early 2025, CoreWeave raised $7.5 billion in its initial public offering to fund GPU cloud expansion, making it one of the largest infrastructure-focused listings in recent years and signaling investor appetite for power-hungry compute facilities. Meanwhile, Northern Data Group announced plans to expand its European GPU cluster capacity by 40% through 2026, targeting AI inference workloads that compete directly with streaming CDN operators for rack space and electricity in constrained markets. The company's Taiga Cloud division has been converting former cryptocurrency mining sites into AI-ready facilities, a pattern mirrored by Crusoe and other repurposing plays. On the capital and permitting side, institutional investors are pouring money into data center energy infrastructure as a distinct asset class. Brookfield Asset Management committed $10 billion to AI data center development in partnership with Microsoft in September 2024, a fund specifically targeting sites where power procurement is the binding constraint rather than land or fiber. That deal structure has become a template: Blue Owl Capital and DigitalBridge launched a joint venture in early 2025 focused on acquiring and developing data center sites with secured power entitlements, reflecting how power rights have become the primary valuation driver. Bloom Energy has benefited from this trend, with its solid oxide fuel cell systems deployed at multiple hyperscale sites as bridge power while grid connections are pending, offering operators a path to energize facilities 18 to 24 months faster than traditional utility interconnection timelines. Technical benchmarks from recent deployments illustrate the scale of the power challenge facing streaming and CDN operators sharing these markets. Vantage Data Centers announced a 1.2 GW campus in Texas in mid-2025, one of the largest single-site developments in North America, designed to serve both AI training and content delivery workloads with dedicated natural gas generation. EdgeCore Digital Infrastructure has taken a different approach, , a strategy that aligns with streaming platforms seeking low-latency points of presence outside congested hubs like Northern Virginia and Dallas. Oracle's cloud infrastructure arm has similarly , targeting deployment timelines under 12 months, a pace that traditional utility-fed builds cannot match and that may set the standard for node deployment. Recent debates further highlight the political risks facing these massive infrastructure projects, as continue to throttle cloud growth. As the expands, sustainability remains a critical hurdle for operators.