Hyperscalers plan to bring just under 18 gigawatts of North American data center capacity online in 2029. Jefferies thinks they’ll get 5.86 GW. That’s the widest gap in a 2026-2030 buildout forecast from the investment bank, which argues the AI construction boom will be paced by crews, cooling gear and grid connections rather than by capital budgets.
The forecast was prepared by Jefferies managing director Jonathan Petersen with historical data from research firm Aterio and released Oct. 9. The Next Platform walked through its numbers Oct. 16. The model covers installed IT capacity only. It doesn’t price construction, equipment or power, and it weights each project by its odds of finishing instead of taking operator schedules at face value.
Planned vs. expected data center capacity
The two lines match this year at 6.21 GW, which isn’t much of a forecast with the year nearly over. Then they split.
- 2027: 8.37 GW planned, 7.53 GW expected
- 2028: 10.13 GW planned, 6.54 GW expected
- 2029: just under 18 GW planned, 5.86 GW expected
- 2030: 13.82 GW planned, 4.46 GW expected
Across all North American operators, not just hyperscalers, Jefferies tracks about 65 GW under construction and about 361 GW announced. The announced figure is roughly 7.6 times the installed base. The bank expects 12.9 GW of new capacity in 2026 and 18 GW in 2027.
What limits the data center buildout
Jefferies ranks the bottlenecks by year. Labor binds first, in 2026. Cooling equipment takes over from 2027 through 2029. Available power becomes the main limit in 2030. All three apply the whole time. Only the order changes.
Its bear case adds politics: growing pressure to make data centers pay their share of new generation, and doubts in host communities about what they actually get from these campuses. Even the bull case assumes supply chains ease only partway, with delays and some cancellations.
What the Jefferies forecast means for contractors
This isn’t a call that demand is fading. It says the backlog is bigger than the industry can build, which is a different problem. In aggregate, Jefferies expects about a third of 2029’s planned hyperscaler capacity to arrive on time. That supports pricing for electrical and mechanical contractors who have crews. It argues against treating every announced campus as a real job.
It lines up with what builders already report. Electrical contractors on these campuses are working 12- to 18-month schedules with little slack. Bain recently put the full AI buildout at $5 trillion to $6.5 trillion by 2030, and the two views can coexist. The money may be there. The chillers and electricians may not be.
The test comes in 2027. If cooling equipment deliveries don’t improve by then, Jefferies’ 2028 haircut is likely to look conservative.