Voters reject AI data centers in 38 midterm ballot measures

How the US midterm backlash against AI data centers, led by Texas, hits 2028 buildouts, plus OpenAI's 948 Linux kernel bug claim and Pierre deal.

GENERAL Newsletter
H
Hugenode

US voters turned against AI data centers in Tuesday’s midterms. Restrictive measures passed in most ballot fights on the issue. Texas Governor Greg Abbott lost to a challenger who campaigned against his data center record, and Democrats took control of Congress. Anyone planning AI capacity for 2028 should expect slower permits and higher costs. Separately, OpenAI says an unreleased model validated and generated patches for 948 Linux kernel privilege-escalation bugs in one run. It is also buying the startup behind several of Codex’s developer tools.

US midterms: voters back data center bans and anti-buildout candidates

The counts differ by source but point the same way. 404 Media counts data center proposals in more than 40 races across 13 states, with voters siding against data centers in 38 so far. Six cities banned them outright, and 12 capped how many megawatts they can draw. Polling firm Redbud Consulting, cited by Ars Technica, tracked 44 restrictive measures across 10 states and found 38 passed. Only one measure allowing a data center to proceed passed. Even that one makes developers in a North Dakota city pay for independent reviews and hold public hearings.

The biggest result came in Texas. Democrat Gina Hinojosa beat Abbott by more than 700,000 votes, according to 404 Media, after making his pro-data center record central to her campaign. Texas was expected to account for more than 41 percent of added US data center capacity in 2028, Ars reports, citing Axios. Bloomberg says about 13 gigawatts of capacity is now exposed. Hinojosa plans to hold projects until laws addressing water, electricity, air and property rights pass. GLJ Research’s Gordon Johnson, in a note cited by Bloomberg, expects projects that already have power, permits and steel to keep going. Projects with power but no state environmental permits, he says, will slip 12 to 18 months.

Democrats also won governor races in Arizona, Georgia and Ohio. Axios projects that nearly two-thirds of 2028 capacity could face delays like those threatened in Texas. Ohio is where OpenAI has eyed a $500 billion data center, and TechCrunch reports that the county where Meta had planned a 700,000-square-foot facility has just passed a ban. Morgan Stanley analysts see governor-elect Amy Acton as a major risk to Ohio’s buildout. Goldman Sachs predicts construction delays in Georgia and Texas.

Money did not decide these races. TechCrunch notes that pro-AI super PACs spent millions, yet candidates opposing the rapid buildout won in Texas, Maine, Michigan, Nevada, Georgia and South Carolina. Maia Woluchem of the research nonprofit Data & Society calls frustration with tech power a bipartisan issue.

People disagree about what the vote means. TechCrunch cites an NBC exit poll in which almost 80 percent of respondents oppose data centers in their neighborhoods and almost 60 percent called data centers a crucial or major factor in their vote. A civil rights advocate quoted by Ars reads the NBC polling as 3 in 4 voters opposed. Voters also ranked cost of living and government corruption above AI and data centers, and the industry is leaning on that. Chamber of Progress CEO Adam Kovacevich says Democratic gubernatorial candidates who opposed bans won at a higher rate than those who supported them. His summary: “Let’s not confuse an anti-Trump wave with an anti-tech mandate.”

Democratic control of Congress could also improve the odds for federal bills, TechCrunch reports. One is the Sanders and Ocasio-Cortez bill, which would impose a nationwide moratorium on new data centers until Congress adds guardrails. Another is Sen. Martin Heinrich’s GRID Savings Act, which would stop grid infrastructure costs from being passed on to communities. Trump can veto either bill, and Time, cited by Ars, reports that Democrats would likely need substantial Republican support to pass laws over a veto.

What it means: analysts cited by Ars say the new ballot measures largely spare data centers already under construction or approved through 2027. The impact falls mainly on projects planned for 2028. If you are siting capacity or signing multi-year compute contracts, plan for longer permitting, fewer tax incentives and pressure to pay for your own grid upgrades. TechCrunch reports that officials’ NDAs and quiet approvals under old zoning laws fueled the anger, so our read is that disclosure and community consent are becoming part of the cost of building. Natixis economist Gary Ng told the South China Morning Post that constrained US capacity could keep US AI token prices high. That is a forecast, not a finding, but it belongs in your cost planning.

OpenAI says its autonomous pipeline found 948 Linux kernel privilege-escalation bugs

OpenAI says its security researchers built an autonomous pipeline for finding, exploiting and patching vulnerabilities, according to Latent Space’s roundup of the announcement. In a single run with an upcoming, unreleased cyber model, the pipeline validated and generated patches for 948 local privilege escalation bugs in the Linux kernel. OpenAI puts the total API cost at $1.2 million. It also says the agents used at least 10 exploitation techniques that its team and consulted experts consider novel.

OpenAI’s Eric Wallace says researchers typically find about 20 Linux privilege-escalation bugs a year. Greg Brockman puts the cost at roughly $1,000 per vulnerability. A comparison cited in the roundup has GPT-5.6 Sol finding 166 such bugs for about $500,000 in July. OpenAI says it is working with Linux maintainers and Patch the Planet partners on reports and patches, and is putting “a significant amount” of frontier compute toward pro bono open-source patching.

All of these figures come from OpenAI and have not been independently verified. The roundup does not say how many of the patches maintainers have reviewed or accepted. Our read: if even part of this holds, security and platform teams should prepare for a wave of kernel fixes and advisories, and should check how quickly they can roll out kernel updates. The same capability raises misuse concerns. In the same roundup, Cognition says it is keeping the weights of its new SWE-2 Flash coding model private, citing cyber-misuse risk, and plans a secure weight-sharing program for partners and researchers.

OpenAI buys Pierre Computer Company to scale Git tooling for Codex agents

OpenAI is acquiring Pierre Computer Company, the startup behind the Code Storage service and several open-source libraries that already power Codex. Trees handles fast file navigation, and Diffs shows code changes and lets users fix bugs directly in the diff view. Highlights does real-time syntax highlighting as a developer types or an agent generates code. OpenAI says the aim is Git workflows that keep up as developers and agents write code side by side, with agents handling more tasks in parallel. It already runs Pando, its own version-control backend built for both. Pierre’s projects will stay open source, and the deal still needs regulatory approval.

This changes nothing today for most teams. It does show where the friction is: Git tooling built around human pace is straining under agent volume. The same Latent Space roundup reports that Meta is acquiring Conductor and Factory is acquiring HumanLayer. Agent-tooling startups are being bought quickly, so if you depend on one, check its roadmap and license.

What to watch

  • How long Hinojosa’s Texas pause lasts, and whether Arizona, Georgia and Ohio follow with moratoriums or grid-cost rules.
  • Whether the Sanders and Ocasio-Cortez moratorium or the GRID Savings Act advances in the new Congress, and whether Trump vetoes either.
  • Whether Texas projects move to Louisiana or Georgia, as Ars reports some may, given that Georgia also elected a Democratic governor.
  • How many of OpenAI’s 948 kernel patches Linux maintainers accept, and when OpenAI releases the cyber model behind them.
  • Regulatory approval of the Pierre deal, and whether Pierre’s libraries stay actively maintained as open source under OpenAI.

Sources