TrendForce Report: AI Infrastructure Bottleneck Shifts from Chip Shortages to Power Constraints
A report by TrendForce reveals that the primary bottleneck limiting AI infrastructure expansion is shifting from semiconductor availability to electrical power supply and grid capacity. Global data center electricity demand capacity is projected to reach 161 GW in 2026, a 31% year-over-year increase driven by AI server deployment. This shift indicates that future AI cloud scaling will depend heavily on power infrastructure, grid interconnects, and transformer equipment rather than just chip manufacturing. If grid capacity fails to keep up, data center deployments and AI model scaling could face severe delays by 2028 and beyond. AI servers are expected to consume over 40% of total data center power demand by 2027, up from 25% in 2025. TrendForce estimates a supply-demand gap of roughly 268 GW by 2030, with data center demand reaching 490.7 GW against an available grid supply of only 222.6 GW.
## BACKGROUND
The rapid growth of generative AI initially led to severe shortages of hardware accelerators like GPUs and Google's custom Tensor Processing Units (TPUs). However, as server density increases with modern high-power AI chips, facility owners are encountering physical limits in regional power distribution, electrical transformers, and cooling systems.