Within a day of the November 3 midterms, estimates of how much US AI data center capacity is at risk ranged from 10 percent to "more than half." The numbers sound similar, but they measure different things. Goldman Sachs says states likely to tighten rules hold about 45 percent of planned capacity. Cleanview reportedly puts 159 GW, or 41 percent of the planned pipeline, in states where new governors back a pause. Data Center Dynamics reported that more than half of US hyperscale projects face "regulatory friction." Analyst Dan Ives talked about slowing a planned 800 to 1,000 data centers.
None of these says that capacity will be cancelled. Each says it is exposed. This post sets out what each source claims, what we could verify, and what builders should do with it. For the political backdrop, see our report on what the midterms change for state preemption, safety bills and data center moratoria.
TL;DR: claimed versus verified
| Claim | Source | What it measures | Status |
|---|---|---|---|
| About 45% of planned US capacity in states likely to tighten rules | Goldman Sachs note, as reported by investingLive | Exposure to state pauses and rules | Reported secondhand; note not public |
| 159 GW, 41% of pipeline, in states with new governors backing a pause | Cleanview newsletter analysis | Capacity in pause-friendly states | Published by Cleanview; method not fully read |
| More than half of hyperscale projects face "regulatory friction" | Data Center Dynamics report | Projects facing greater regulatory hurdles | Headline seen; DCD page blocked for us (HTTP 403) |
| 13 GW of Texas power could stall | Gordon Johnson, GLJ Research, via CBS | One analyst's Texas estimate | Method not explained |
| 800 to 1,000 planned data centers may slow | Dan Ives, Yorkville Ives & Co., via CBS | Analyst view | Opinion, no method given |
| Ballot measures blocked only about 1.6 to 1.9 GW directly | One analysis in our search | Directly blocked, sized projects | Single secondary source |
What each estimate actually says
Goldman Sachs: 45 percent, but "likely short-lived"
According to investingLive's account of the note, Goldman estimates that states likely to tighten data center rules hold about 45 percent of planned US capacity, with Texas alone roughly a quarter. It expects Democrats in Texas, Georgia and Michigan to impose short-term pauses while lawmakers consider tighter rules on electricity use, water consumption and tax incentives.
The same account says Goldman expects the curbs to be short-lived because data centers are not a top voter priority. Strategist Rich Privorotsky suggested some companies could pull spending into 2027 to start projects before new rules take effect. The stated risk is that pauses last longer than expected, which would turn a timing delay into a capacity loss. The article reports the note secondhand, "according to reports of the note," so we could not read Goldman's own text.
Cleanview: 41 percent in pause-friendly states
Cleanview's newsletter puts roughly 158.6 GW of planned data centers, about 159 GW, in nine states where a governor or governor-elect has promised or imposed a pause. That is 41 percent of the national pipeline. A further 69.2 GW, or 18 percent, is in states where the winner wants to change tax breaks or rules without backing a pause. Per search summaries of the newsletter, the pledges target new construction, so the 56.7 GW already under construction is less exposed, and Cleanview itself calls the at-risk figure "paper capacity." We did not read Cleanview's full method.
Data Center Dynamics: more than half of hyperscale projects
The headline on the DCD report says more than half of US hyperscale data center projects face "regulatory friction" after the midterms. The page returned a 403 error for us, so we could not check the definition of "friction" or how many projects were counted. "Face friction" is a much weaker claim than "will be cancelled."
Analysts quoted by CBS News
CBS News quotes Gordon Johnson of GLJ Research, who says Governor-elect Gina Hinojosa's win could stall 13 GW of data center power in Texas. CBS says the article does not explain how he calculated this. Dan Ives said the answer to whether the midterms will slow the planned 800 to 1,000 US data centers is "a clear YES" in his view, and noted Texas accounts for 25 to 30 percent of capacity expected online by 2030.
A rounded plot of land with a grid and one green square, representing planned data center sites facing state rules
Why the numbers should not be added
The Goldman and Cleanview figures overlap in scope. Both count Texas, Georgia and other states, and each uses its own definition of "planned" and "exposed." Adding 45 and 41 gives a number with no meaning. The DCD figure is about projects, not gigawatts. Ives and Johnson give opinions. The honest summary is this: several analysts think a large share of the pipeline sits in states about to get stricter rules, and none of them has shown that the share will be lost.
Also note the difference between a pause and a ban. Texas governor-elect Hinojosa said she plans a moratorium on all data center approvals until the legislature passes protections, per The Verge as cited in our earlier coverage. CBS reports Abbott had already suspended approvals, and Johnson says Hinojosa would only need to keep that freeze. Our post on what Hinojosa's plan means for projects already permitted separates new approvals from existing permits. In Georgia, CBS reports governor-elect Keisha Lance Bottoms said she would "sign an executive order for that pause" and plans to review state tax incentives.
What the economists say
CBS News also quotes more cautious voices. Stephen Dover of Franklin Templeton says the AI trade may have been overheating, so "unbridled expansion will be tempered a bit, but probably not a lot." Oxford Economics says a more restrictive stance would "stretch the timelines of future projects," and that slower construction "wouldn't necessarily be a bad outcome for the macroeconomy." Economists broadly argue that moratoriums may delay construction without seriously curbing AI capital spending or growth.
Voters have also acted at the local level. Redbud Consulting, cited by CBS, counts 38 ballot measures restricting data centers, with Harwood, North Dakota, the only place to approve one. We covered the Ohio side in our report on voters passing data center bans in 14 municipalities. Local bans matter, but a separate analysis in our search suggests they directly blocked only a small amount of sized capacity, so they do not explain the headline percentages.
Background: the pause idea is not new in Texas
The Texas freeze did not start with the election. CBS reports that Abbott had already suspended data center approvals before Hinojosa won. Earlier in the year, our coverage of OpenAI's letter to Governor Abbott showed that the largest AI labs were already lobbying on the question of a pause. That history matters for the estimates: a number built on "a new governor will pause projects" assumes a change, while in Texas part of the pause may simply continue.
It also explains why analysts disagree about duration. If a freeze is mostly a continuation, the lost time may be small. If a new governor adds fresh rules on power, water and tax breaks, the wait for each project could be longer. Neither Goldman nor Cleanview, as relayed in the reports we read, publishes a project-by-project schedule.
What this means for people who build with AI
This is not a markets story for us. The practical question is whether the compute you depend on will arrive on time and at what price.
- Expect schedule risk, not a cliff. Even the source that gives the biggest number, Goldman, expects short pauses. Plan for delays of capacity in Texas and Georgia, not for it to vanish.
- Ask your provider where capacity sits. If your cloud or model vendor leans on one state, a pause there matters more than a national percentage. Our look at how OpenAI is reportedly weighing a reset of 2027 capacity shows how labs are already reacting.
- Watch the instruments, not the estimates. Executive orders, bills and permit decisions are verifiable. Think-tank percentages are not.
- Build in portability. Keep your stack able to move between regions and providers, and cache or batch work that does not need low latency.
- Watch the market reaction with caution. Our note on the November 5 AI stock sell-off covers prices, which move faster than facts.
How to check these claims yourself
- Find the primary document: Cleanview's report, the Goldman note, the DCD study. If you cannot, treat the number as an opinion.
- Ask what the denominator is: gigawatts, projects, or dollars.
- Ask what "at risk" means: paused, delayed, taxed or cancelled.
- Compare with the executive orders and bills as they appear, using sources such as the governors' own announcements.
- Re-check in a month, when real pauses either exist or do not.
Bottom line
The best-supported reading is modest: a large share of planned capacity sits in states whose new leaders want more rules, and some pauses are likely. The data does not support "half the buildout is gone." Hold estimates loosely and watch the orders that actually get signed.
Related reading
- Midterms: state preemption, safety bills and data center moratoria
- Hinojosa and permitted Texas data centers
- Data centers on election night
- Ohio voters pass data center bans
- OpenAI weighs resetting 2027 data center capacity
- The November 5 AI stock sell-off
Figures are as reported on November 4 and 5, 2026, mostly secondhand, and may change as analysts publish full reports.
