Power and Data Centre Limits Threaten to Slow Down Global AI Expansion

Power supply and data centre infrastructure may soon matter more for artificial intelligence than simply buying the newest computer chips, a Morgan Stanley research report warns.

Shortfall Estimates

US data-centre developers could face a 32-gigawatt power shortfall through 2028, even after using extra power generation methods. This gap makes up about 34 per cent of all projected power needs. Chip maker Nvidia confirmed this view, stating that power demand exceeds supply and more semiconductors are being built than there are places to put them. The total cumulative US power shortfall for data centres could even grow to 68 gigawatts by 2029.

Market Impacts

Access to electricity now favours large firms with strong finances and steady revenues. Major tech companies and cloud providers will likely make up about 94 per cent of global data-centre capacity additions in 2026, 64 per cent in 2027, and 57 per cent in 2028, potentially squeezing out smaller developers. If data centres cannot install ordered chips, customers might delay deliveries or cancel orders, posing risks to suppliers of memory chips and optical equipment. Still, Morgan Stanley expects Nvidia and Broadcom to meet their 2027 forecasts.

Global Shifts

This shortage might push tech companies to build computing infrastructure outside the US, including in Asian markets. Morgan Stanley's Asia data-centre team estimates that 14 gigawatts of demand from US hyperscalers could move to Asia through 2030, though international growth will not solve the US power problem.

Context

Advanced artificial intelligence systems require vast computational power, which depends on massive server farms that consume electricity on the scale of small cities. As technology companies race to build these capabilities, the physical limits of local power grids and land availability create major bottlenecks.