AI Power Constraints Deepen as Grid Capacity Lags

AI power constraints are becoming a bigger obstacle to growth, with electricity supply, grid access and data-centre capacity increasingly limiting expansion alongside chip availability. UBS previously tracked 276 GW of data-centre capacity under construction or planned, but estimated that 2027 demand would require 57 GW against just 38 GW of sites with confirmed power. It highlighted the mismatch in timelines: data centres can take 12–24 months to build, while new power capacity may take five to 10 years. US grid interconnection queues total about 2.3 TW, with a median wait of roughly six years; UBS estimated $511 billion in new US power-generation investment would be needed by 2030. Transformer delivery can take up to 210 weeks. UBS also named Core Scientific (CORZ) and Applied Digital (APLD) among colocation operators it favours, projecting about 80% annual EBITDA growth from 2027 to 2029, and said power supply could reach around 415 GW by 2030 if planned projects proceed. The later Delphi Digital analysis adds signs of mounting grid stress. ERCOT’s large-load queue rose from 63 GW at the end of 2024 to 474 GW by June 2026, with data centres accounting for about 90%. PJM’s last three capacity auctions reached the price cap and remained 6.6–6.8 GW below reliability targets. Goldman Sachs forecasts data-centre power demand will rise 170% from 2025 to 2030, with $7.6 trillion in related capital spending through 2031, and expects hyperscaler debt issuance to increase by $420 billion in 2027. NERC projects a 224 GW rise in North American summer peak demand over the next decade and warns that 13 of 23 regions could face shortfalls by 2030. For traders, the reports reinforce that AI power constraints—and the ability to secure grid connections, equipment and project funding—are key indicators of whether planned data-centre capacity can come online. The trend may benefit power generation, grid equipment and interconnection providers, while creating execution and financing risks for AI infrastructure projects. UBS’s positive view on CORZ and APLD is company-specific; neither report identifies a direct price catalyst for cryptocurrencies.