Topics Pioneers Insight Method Author
Back

Ep. 043 - The Midterms and the 2028 Datacenter Buildout (Datacenters)

·SemiAnalysis

Rendering the long image…

Topics AI Data Center Power Bottlenecks
Pioneers Maya Barkin

Summary

  • The panel remains comfortable with 2027 construction while treating 2028 approvals as the larger election risk. Maya says roughly three-quarters of 2027 capacity in the top 20 US states is already under construction, supporting the team’s 50-plus GW global forecast. By contrast, most of the modeled 62 GW of US additions in 2028 has not started construction: 2027 is largely “locked in,” while 2028 is not.
  • Texas is the key downside risk because its permitting freeze now reaches on-site power. The state represents close to 30% of planned 2028 US additions, and bringing your own generation no longer bypasses the hold for projects needing permits. An Abbott victory could allow a December restart; his Democratic opponent’s legislative conditions could add five to nine months, potentially leaving some 2027 campuses with shells but no electricity.
  • The polling-based scenario exposes 5.7 GW of planned 2028 capacity, versus 30.3 GW in the adverse election scenario. Across 20 candidates in 10 governors’ races, nine favor pauses and eleven favor conditions, with a roughly even party split. These figures describe “the capacity that is exposed,” not predicted delays: conditions raise costs while preserving an approval path; moratoriums stop projects lacking permits.
  • Communities have gained bargaining power as data center demand and the cost of incentives have grown. Wayne’s Ohio example runs from an anticipated $136 million in annual incentives to $555 million two years ago and $1.6 billion last year. Jeremie argues that a $50 billion, 1 GW project can contribute materially more locally: “capitalism is going to solve part of the problem.”
  • State directives and local approvals are more consequential for immediate delivery than a potential Democratic sweep of Congress. Maya expects Trump’s veto and the need for Republican Senate votes to constrain federal changes. The bipartisan direction is toward developers paying grid costs upfront or posting more collateral, which could delay projects but “doesn’t prohibit development”; Georgia remains a closely contested pause-versus-conditions race.
  • Resolving opposition requires operating improvements and visible community benefits alongside financial contributions. Jeremie cites a Virginia incident where simultaneous UPS responses removed 1.5 GW of grid load, illustrating a legitimate reliability concern. Wayne also challenges the idea that $10,000 payments would settle opposition: “some people don’t want to be paid off,” particularly when construction jobs go to workers brought from elsewhere.

Deep dive

1. Construction status separates 2027 from 2028

  • Maya Barkin’s distinction is construction status: roughly three-quarters of 2027 capacity in the top 20 US states is already being built. That supports the team’s “fifty-plus” GW global year. The election could change construction rules, but this pipeline has “a big head start already.”

  • For 2028, Maya’s model has about 62 GW of US additions, most not under construction and largely in states electing governors. Permits, power and schedules remain unsettled. This differs from September’s local restrictions: of roughly 20 GW inside those restrictions, only about 1.5 GW was estimated to be delayed, and that finding still holds.

  • Across 20 candidates in ten governors’ races, nine want pauses and eleven want conditions, with a roughly even party split. Maya’s dividing line: “conditions make a project more expensive”; developers can fund grid upgrades, redesign cooling or sign community agreements and retain a route to approval. A pause stops permit processing for projects without all required approvals.

  • The adverse election scenario puts 30.3 GW of planned 2028 capacity in exposed states; the conditions-favoring scenario drops to 5.7 GW, also the polling-based estimate. Maya stresses this is “the capacity that’s exposed,” not predicted delayed capacity. The analysis takes campaign proposals “at face value,” even though some might deserve “a grain of salt.”

2. Scale gives communities leverage and creates real operating risks

  • Jeremie Eliahou Ontiveros sees bargaining power reversing: counties once competed for projects, but the 2027 US build of 40 GW is roughly 10–20 times the annual scale five years earlier. GPUs make campuses highly valuable relative to their footprint; a 1 GW facility represents $50 billion. Developers can contribute more locally: “capitalism is going to solve part of the problem.”

  • Wayne Ma’s Ohio example makes the fiscal shift concrete: incentives expected to cost $136 million annually reached $555 million two years ago and $1.6 billion last year. Jeremie’s pushback is worth keeping: tax breaks are normal for enormous industrial investments. Supply constraints now let communities demand materially larger contributions; the earlier arrangements were not unusual simply because they offered incentives.

  • Jeremie distinguishes legitimate operating concerns from generalized backlash. Older open-loop cooling could be energy efficient but water inefficient; noise and competing projects change what communities can absorb. In Virginia, data centers reacting to a voltage dip disconnected simultaneously, removing 1.5 GW of load. Such a fast disappearance can trigger a blackout; developers need equipment to be a “good grid citizen.”

  • His conditional optimism: “data centers are going to become much better neighbors,” and that should hopefully improve the buildout as designs adapt to scale.

3. Texas timing can determine whether completed shells get power

  • Texas alone represents close to 30% of modeled 2028 US additions. Maya says Abbott’s late-September TCEQ permit freeze, layered over an ERCOT interconnection audit begun in August, also captures on-site power. September’s escape route was bringing your own power; “it isn’t anymore.” Even a few lost months in this market could move the national total.

  • At recording, Abbott was favored for a fourth term. Maya expects restrictions to ease: ERCOT’s report is targeted for early December, after which permitting could reopen with stricter requirements. Most projects could absorb that interruption if the schedule holds, and his directives can expire without reversal. Abbott also wants 2027 legislation ending incentives, protecting ratepayers, adding electricity supply and tightening water rules.

  • His Democratic opponent would retain the moratorium until legislators pass protections. With the regular session running January–May 2027, Maya estimates an eight-month pause if approvals move through that session, or nearly a year if rules start in late summer: five to nine additional months versus December. Proposed conditions include household electricity and water-bill protection, ending subsidies, banning NDAs and union labor.

  • Jeremie identifies the 2027 exception: some campuses already under construction and funded with billions of dollars lack final air-pollution permits because Texas historically approved them quickly. An extension into mid-2027 could leave partly or mostly completed shells without electricity: their power plants cannot be built without those permits. He calls this “entirely possible,” although it is not the team’s base case.

4. Governors can halt permits, but local conditions still matter

  • Jordan Nanos challenges translating governors’ races into project outcomes: approvals also involve councils, county boards and utilities. Maya agrees local decisions remain project specific, but governors can quickly direct agencies to stop permits while legislatures move more slowly. Jeremie’s example favors local discretion: Indiana spans PJM and MISO, with different rules; a blanket bring-your-own-power requirement can ignore electricity already available locally.

  • Georgia is Wayne’s other watch point: Keisha Lance Bottoms wants a temporary pause; Rick Jackson favors conditions, and polling is neck and neck. Jordan flags another uncertainty: neither has stated a position on ending tax incentives, potentially a major project cost. Wayne sees Democrats leaning toward union labor, transparency and bans on NDAs, and Republicans toward local control, despite broad agreement on energy costs and incentives.

  • On a Democratic sweep of Congress, Maya expects little immediate change: Trump retains a veto, and a narrow Democratic Senate would still need Republican votes for stricter legislation. Bipartisan agreement centers on large users paying their grid costs. Upfront payments or more collateral could delay projects, but “it doesn’t prohibit development”; permitting reform could actually accelerate needed infrastructure.

5. Community benefits can help where individual payouts fall short

  • Wayne sees Amazon and Google dropping NDAs as a start: secrecy creates the impression of “something secret going on” or politicians serving corporate interests. Maya calls the backlash a “2026 phenomenon” and connects opposition to uncertainty about jobs and children’s futures. One representative she interviewed said: “if data center developers were hanging up Christmas lights, people wouldn’t be opposed to them.”

  • Jordan’s construction-site visit provides a counterweight: workers described high-paying jobs important to their families, with some commuting 1½–2 hours. He acknowledges the selection bias and says $50 billion for a few thousand jobs is “probably a bad trade,” while emphasizing that local benefits still exist.

  • Wayne rejects the idea that $10,000 per resident would settle opposition: “some people don’t want to be paid off.” An Iowa board-of-supervisors candidate told him many construction workers come from other states or countries, weakening the local-jobs argument. Maya favors “a more comprehensive process”: transparency, education about newer data centers’ water use and environmental impact, and contributions such as schools or roads, rather than individual payouts.