US Datacenters Are Still on Track for a Record 2027. The Moratorium Frenzy Has Barely Slowed Them.

Of the roughly 38 gigawatts of new US datacenter IT capacity the SemiAnalysis Datacenter Industry Model forecasts for 2027, only about 2.3 GW is currently delayed by local moratoriums and New York's executive order. Twenty-two gigawatts of that total is already under vertical construction, and most of the remaining 16 GW of "planned" capacity is financed, doing sitework, or preparing to go vertical. The forecast has barely moved in twelve months, according to the SemiAnalysis newsletter, even as the model gets refreshed against biweekly client updates and the construction activity keeps accelerating.

The 2.3 GW figure rests on a method the newsletter's authors have spent years building. The model tracks real-time and historical satellite imagery of more than 6,000 US datacenters, layering on permitting, grid interconnection, and project-finance data. A self-reported pipeline number is easy to inflate; an orbital photo of dirt being turned and steel going up is harder to fake. The same dataset is what lets SemiAnalysis separate the projects that are genuinely paused from the ones whose paperwork just got noisier.

The 2.3 GW figure is the slice where SemiAnalysis says it can establish a direct, project-level impact. About 1,525 MW of it sits inside local moratorium boundaries, a small share of the roughly 20 GW that fall inside restricted zones. The rest traces to New York's executive order halting new environmental permits for datacenters. Outside those two categories of policy intervention, the model cannot cleanly link political action to a delayed project.

In the past two months, four states have moved: New York stopped issuing environmental permits for datacenters; Texas paused the next step in its massive ERCOT interconnection queue; Pennsylvania pulled datacenters out of fast-track permitting and tied new state permits to guardrails; Oregon froze datacenter deals on state-owned land. Over the past eighteen months, more than 300 towns, cities, and counties have voted to halt datacenters.

Most of those actions do not show up in the 2027 model. The buildout is already largely de-risked: financing is committed, transformers and switchgear are on order, and grid interconnect positions have been secured or paid for. Pennsylvania's pull from fast-track permitting is a procedural change with real cost; Texas's interconnection pause and Oregon's state-land freeze will reshape the queue, not the under-construction line. New York is the only state action currently showing up in the delayed-capacity tally, and even there the project-level impact is measured in hundreds of megawatts, not gigawatts.

The moratoriums are not stopping the buildout, despite what much of the public debate has framed them as doing. The pipeline has absorbed the most concentrated anti-datacenter political cycle on record without breaking stride. The construction data points to a different problem: political action is concentrated on a buildout the orbital record shows is largely past the pause point, while the problems the data does show, including rising electricity prices for ratepayers, water use in dry regions, and land-use conflicts in rural counties, remain unresolved.

Two things would push the 2.3 GW figure up: a Pennsylvania-style pull from fast-track permitting in another state, or a wave of local moratoriums that delay projects whose grid interconnect positions have already been paid for. A model revision in either direction would move the headline number. The 2027 buildout, for now, looks like it will land close to the figure SemiAnalysis has been carrying for the past year.