Half of Scheduled AI Datacenter Capacity Won't Arrive by 2028

Datacenter capacity due by 2028 will fall roughly 50 percent short of what companies have announced, per Goldman Sachs data CNN cited August 6.

That's a bigger problem than the public backlash getting headlines. Gallup found 71% of Americans oppose datacenters near them, but CNN's reporting says opposition isn't what's slowing the buildout. Labor is.

Goldman Sachs puts the historical on-time delivery rate for this kind of construction at about 72%. AI datacenters are tracking well below that. Only roughly half of the compute capacity scheduled to activate between now and 2028 is expected to show up on time.

The gap comes down to skilled trades. The American Edge Project estimates hitting the announced buildout numbers would require adding 500,000 electricians, 300,000 welders and 550,000 plumbers, on top of whoever's already working these sites. That's not a permitting fight or a NIMBY vote. It's a workforce that doesn't exist yet, at a scale nobody's training for on this timeline.

Any AI buildout plan that takes company-announced datacenter timelines at face value is off by about half. Anyone forecasting AI compute supply should be halving the announced capacity figures before building projections on top of them. Watch whether the companies behind these projects start quietly pushing activation dates instead of quietly missing them.

CNN tied the Goldman Sachs delivery-rate math and the American Edge Project's labor estimate to Gallup's new opposition numbers in its August 6 report.