U.S. Data Center Expansion Faces Multiple Headwinds; Goldman Sachs Expects Half of Planned Capacity to Be Delayed
nashnova research
Goldman Sachs estimates only 50%–60% of planned U.S. data center capacity will come online within the next two years, as grid bottlenecks, community opposition, and tightening regulation squeeze the buildout boom and force developers to rethink timelines and power strategies.
Half the pipeline delayed — what is actually blocking construction?
Goldman Sachs projected in May that just 50%–60% of planned capacity will arrive on schedule over the next two years. This means → nearly half of all projects already in the pipeline face delays or shelving.
Wood Mackenzie data confirms the slowdown: new development pipeline in Q1 2026 totaled 36 GW, down 19% from Q4 2025.
In plain terms = developers aren't losing interest — they're finding the path from blueprint to power-on increasingly impassable, and are pulling focus back to existing projects.
Microsoft, Google, Oracle — where exactly are the big projects stuck?
Microsoft's Michigan site faces permitting delays; its New Jersey site was fined $1 million for running gas generators without permits and given 45 days to obtain approval or shut down.
Google's Minnesota data center was ordered to halt construction by a court ruling requiring a full environmental review.
Oracle activated financial protections for its 2.5 GW "Project Jupiter" campus in New Mexico — the state regulator rejected the pipeline application needed to supply gas. This means → even projects that bypass the grid can be blocked by energy-infrastructure permitting alone.
Why are communities and politicians turning against data centers at the same time?
Geronimo Power CEO Blake Nixon said opposition is "unlike anything I've seen in over 20 years in the industry." Officials in Nobles County, Minnesota outright rejected the company's 400 MW data campus proposal.
Democratic governors in Pennsylvania, Virginia, and New Jersey have each moved to require data centers to bear more of their own generation and transmission costs.
Texas Governor Greg Abbott pushed through a blanket moratorium on data center grid interconnection. BloombergNEF estimates this threatens roughly 20% of the entire U.S. development pipeline.
Who pays the bill when power gets scarce?
In the PJM Interconnection region, the independent market monitor attributed 38% of costs in the most recent capacity auction to data center load growth.
This means → data centers are not just consuming large volumes of power — they are driving up electricity costs for every other customer in the area. That dynamic is the core trigger behind community opposition and political backlash.
Can "bring your own power" solve the problem?
Some developers are shifting to "bring your own capacity" arrangements — co-locating with new generation facilities or self-funding local power. Goldman projects 25% of data centers will use behind-the-meter self-generation by 2030.
OpenAI signed agreements this month to take on 8 GW under such arrangements.
But Oracle's Project Jupiter shows the limit: off-grid self-generation still requires regulatory approval for supporting energy infrastructure. Put simply = a different road, but no fewer checkpoints along the way.
Can data centers really deliver the share of U.S. power demand that forecasts project?
The Electric Power Research Institute (EPRI) forecast in February that data centers will account for 17% of U.S. power demand by 2030 and 20% by 2035.
This reflects a core tension: demand projections remain steep, but supply-side delays and cancellations are piling up.
This means → for utilities, whether the data center demand curve materializes on schedule will be the hardest variable to pin down across planning cycles for years to come.
市场有风险，内容仅供研究参考，不构成投资建议。