Goldman Sachs: 86GW of Projects Under Construction Underpin Data Center Build-Out Through 2030
nashnova research
Two Goldman reports show 86GW of U.S. data-center capacity already under construction — roughly $300 billion in total spend, six times the 2025 annual run-rate — enough backlog to sustain current spending for about six years even if no new project is ever announced.
86GW under construction — how much money is that?
As of late October, about 86GW of U.S. data-center capacity is under construction. At roughly $3.5 billion per GW, that implies about $300 billion in total build cost.
This means → the backlog is six times the ~$50 billion the U.S. will spend on data-center construction in all of 2025. Over $200 billion of it has not yet been deployed.
In plain terms = even if not a single new project is announced from today, the work already on-site can keep the industry spending at the current rate for roughly six years.
How big is the pipeline beyond what's already on-site?
Adding "high-probability" announced-but-not-started projects lifts the pipeline to roughly $900 billion.
Including medium- and low-probability projects pushes the theoretical opportunity to about $1.9 trillion.
Goldman explicitly flags the latter as "a measure of total pipeline value, not a final spending forecast." This means → $900 billion is the more reliable upper reference; $1.9 trillion is a ceiling number, not a projection.
Does the capacity-side math confirm the spending-side math?
Goldman IT-hardware analyst Michael Ng built an independent capacity model: the U.S. currently has 57GW of live data-center capacity, with 9GW under construction, expected to lift the total to 65GW by end-2026.
Projects already under construction will add a further 33 / 37 / 8 / 0GW in 2027–2030, pushing total U.S. capacity past 143GW — more than double today's level.
This means → two analysts, using two different lenses — dollars spent and gigawatts built — reach highly consistent conclusions. The projects already on-site alone are enough to cover Goldman's base-case forecast.
Can post-election regulation actually change this picture?
Converting at a PUE (power-usage effectiveness — total facility power divided by IT-equipment power) of 1.5, the under-construction pipeline delivers 22 / 24 / 5GW of net new IT capacity in 2027/28/29, closely matching Goldman's own forecast of 20 / 24 / 23GW.
Ng says this "gives us confidence in our recently revised forecasts and suggests they are conservative enough."
In plain terms = projects already under construction are the hardest part of the pipeline for regulators to stop. Whether post-election risk can materially change the picture depends on newly elected officials' enforcement vigor and federal legislative progress — but unwinding 86GW of active construction is an enormously difficult ask.
市场有风险，内容仅供研究参考，不构成投资建议。