The scale and speed of AI-driven electricity demand is beginning to outpace the physical limits of power systems.
By 2035, Deloitte estimates that electricity demand from AI data centers in the United States alone could grow more than thirtyfold, reaching 123 GW, up from just 4 GW in 2024. That level of demand rivals the output of entire national grids.
AI data centers are fundamentally different from traditional facilities. A five-acre data center that shifts from CPU-based workloads to GPU-intensive AI processing can see power demand jump from 5 MW to 50 MW almost overnight.
The largest AI infrastructure developers — known as hyperscalers — are building facilities at unprecedented scale:
These facilities create dense, 24/7 demand clusters, which stress grid stability. In fast-growing data center regions, utilities are already reporting:
Meanwhile, AI infrastructure is decentralizing, as models are deployed closer to users to reduce latency — spreading grid stress across more states.
Compounding the issue, some grid interconnection requests now face seven-year wait times.
AI investment has become a systemic macro risk.
Yet an MIT study found that 95% of organisations deploying GenAI achieved zero ROI, despite spending $30–40 billion across 300 initiatives.
AI is quietly becoming a systemic energy, credit and transition risk. The mismatch between AI growth rates and grid expansion timelines creates exposure not just for utilities, but for lenders, investors and sovereign balance sheets.
At ESG.AI, we see AI infrastructure as the next frontier of climate risk disclosure, where energy intensity, water use and capital leverage converge.
