Although U.S. data center construction for 2027 is essentially locked in thanks to a "time gap," the unexpected upheaval from the midterm elections is exposing 62.2GW of planned capacity for 2028 directly to enormous risks of permitting delays and regulatory rewrites.
On November 3, 2026, the day of the U.S. midterm elections, prominent semiconductor and infrastructure research firm SemiAnalysis published its latest edition of the SemiAnalysis Weekly.
The firm's managing director overseeing data center and energy infrastructure research, Jeremie Eliahou Ontiveros, along with analysts Maya Barkin and Wayne Ma, held an in-depth discussion on how the U.S. election results would affect data center capacity expansion.
During the program, Maya offered what at the time appeared to be the safest judgment:
All the polls show Abbott will win.
She was referring to Republican Greg Abbott, who has served as Texas governor since 2015, previously held the office of state attorney general, and was this time seeking a fourth term.
Contrary to market expectations and the firm's earlier "base case," Democrat Gina Hinojosa won the Texas governorship, ending the Democratic Party's 36-year losing streak in gubernatorial elections there; Democrats also retook both the Senate and the House, with Ohio's Amy Acton defeating Vivek Ramaswamy and Georgia's Keisha Lance Bottoms defeating Rick Jackson, who had self-funded more than $200 million.
The program was recorded before the results came out. The three participants divided U.S. data center capacity into two segments: the 2027 portion is already under construction and cannot be changed by the election; the 2028 portion is still about 90% on the drawing board and happens to fall in the states that voted this time.
Three numbers support this dividing line. The top 20 states plan to deliver 33.8GW before 2027, with about three-quarters already under construction; U.S. net additions for 2028 total 62.2GW, with 90% not yet started; and Texas alone accounts for nearly 30% of 2028's new additions.
2027 Capacity "Locked In": Ballots Cannot Change Foundations Already Poured
Regarding the short-term earnings guidance and order fulfillment that the market cares about most, SemiAnalysis offered a clear reassurance: policy changes cannot catch up with construction progress, and capex and delivery schedules for this year and next are basically secure.
When breaking down the capacity model, analyst Maya pointed out clearly that 2027 capacity has already entered the construction phase, and election results cannot pose a substantive obstacle to it. She said:
In the top 20 states for 2027, about three-quarters of capacity has already broken ground. So today's election results can't stop it, which is also why our judgment from a few weeks ago that global annual additions would be more than 50GW still holds.
The core logic behind this lies in the "time gap" of physical construction. A gigawatt-scale campus takes about two years from groundbreaking to operation, and site selection, grid connection, air permits, and equipment scheduling have long been completed. As Jeremie put it:
To affect 2027 construction, you would need to find something that the projects slated for 2027 delivery are missing.
At present, the vast majority of projects slated for 2027 delivery have already obtained construction, stormwater, and air pollution permits. A new governor taking office can only change surrounding rules, not transformers that have already been ordered.
2028's 62.2GW Test: 90% of Capacity "Exposed to Risk"
In stark contrast to 2027's certainty, 2028 has become a "dammed lake" where policy risks are concentrated and ready to burst.
SemiAnalysis's model shows that the U.S. plans net additions of about 62.2GW of data center capacity in 2028. But Maya warned:
Most of it has not yet started construction, which is the opposite of 2027. These projects mostly fall in states electing governors this year, and they are all waiting on permits, power, and construction schedules.
Data shows that among the 43.5GW of planned capacity across 13 key states, about 93% has not yet broken ground, and 10 of those 13 states saw gubernatorial turnover this year. As candidates advocating permitting pauses or tighter rules won, the original 30.3GW of capacity moved directly into the risk zone.
Regarding the nature of this risk, Maya offered a precise definition:
Of course, that doesn't mean all of this capacity will be delayed, only that this capacity is exposed to risk—the capacity those states plan to bring online in 2028.
Because campus construction cycles are rigid, every extra month of a permitting freeze means one less month in the 2028 delivery window. The market risk is that delivery timetables will be forcibly pushed back.
Texas Becomes the Biggest Downside Risk: "Half-Built Shells With No Power to Use"
Among all states facing permitting risk, Texas, which accounts for nearly 30% (about 18GW) of 2028 U.S. new additions, has become the biggest variable shaking the nationwide data center growth guidance.
Previously, a large number of developers used Texas's relatively short permitting cycle to raise billions or even tens of billions of dollars and race ahead on construction schedules.
But with Democrat Gina Hinojosa's surprise election as governor, Texas's original permitting pause policy will most likely continue. Because the legislature cannot pass protective provisions until its regular session from January to May 2027, this means projects must come to a hard stop for 8 to 12 months.
Faced with such a long wait, all the buffer time originally reserved for projects has been completely exhausted.
Jeremie listed Texas as the biggest downside risk and bluntly described the shock brought by this mismatch:
Without final approval, you cannot build power plants. Data centers can be built, and will continue to be built, but power plants cannot. So you may see a bunch of shells half-built or basically complete, yet with no power available.
Cross-Party Policy Rewrite: Farewell to Tax Incentives, Developers Bear the Costs
Setting aside partisan disputes, making data center developers bear more infrastructure costs has become a bipartisan consensus in U.S. local politics.
Analyst Wayne pointed out that among the 20 gubernatorial nominees surveyed, 15 advocated requiring data centers to pay their own grid costs, and 12 advocated eliminating tax incentives. Wayne said:
Whether you're a Republican or a Democrat, if you're an ordinary resident and a data center is built right next to you, it will make you unhappy.
Policy tightening falls into two categories: "pauses (essentially bans)" and "added conditions." Maya explained:
If new conditions are imposed, developers can pay to upgrade the grid, redesign cooling... it may be slower and more expensive, but there is a path to approval. A pause is different. For projects that have not yet obtained all permits, the work simply cannot move forward.
This means the era of local governments attracting investment by "paying out of pocket" with incentives is completely over. Take Ohio as an example: to attract data centers, the tax incentives offered locally soared from $136 million to $1.6 billion in just two years, a nearly 12-fold increase in three years.
In summary, Maya said:
Computing demand is now so enormous that the supply-demand relationship has completely reversed. Local governments no longer need to humbly beg companies with tax incentives to build data centers. This kind of "government-subsidized" investment promotion concession policy will probably disappear very soon.
In the future, for tech giants, permits, the power grid, and water resources will become hard constraints, and changes in the external environment will directly reshape the investment return models of data center projects.
Comments