AI super PACs: What the election results reveal
AI super PACs spent $79 million across 68 races, but key bets failed. See what the US results reveal about voter attitudes, regulation and campaign strategy.
AI Super PACs Spent Millions. What Did Voters Actually Reject?
The AI industry put serious money into the US election cycle, betting that campaign spending could shape who wins - and, eventually, how artificial intelligence gets regulated. One of the biggest pushes came in South Carolina. Groups tied to Leading the Future spent $4.8 million supporting Republican Darline Graham. She lost to Democrat Annie Andrews, who had dubbed Graham "Data Center Darline."
That loss fit a broader pattern. In the closest races, most Republicans backed by AI-aligned super PACs lost, according to Fortune's tally. Voters also rejected AI data centers in 40 of 41 ballot measures. And across 68 House and Senate seats, AI political groups spent a reported $79 million. The results don't add up to a simple "AI referendum," though. They do point to something more useful: political money can help in some places, it can backfire in others, and voters often care more about local impacts than industry arguments.
What happened in South Carolina and why it mattered
The South Carolina Senate race was the AI industry's biggest general-election bet in a competitive matchup. Two super PACs tied to Leading the Future spent $4.8 million backing Republican Darline Graham. Democrat Annie Andrews won. And the "Data Center Darline" nickname gave the race a very local hook: it connected Graham to an industry whose infrastructure plans can land in real communities - fast.
That's the crux. Campaign spending doesn't play in a vacuum. Voters might hear a candidate's AI position in an ad, but they usually judge it through issues they already understand: electricity prices, water use, jobs, land, privacy, and who's influencing government. Big claims about global competition or innovation fall flat when they don't answer those immediate questions.
This loss also made headlines because it challenged a tempting assumption - namely, that heavy spending can always protect a candidate from attacks tied to the AI industry. But one race can't prove cause and effect. Different candidates, different records, different opponents, different districts, different campaign conditions - there's no single explanation that fits neatly. What you can say is simpler: the spending didn't secure the outcome those donors wanted here. And it doesn't tell you what would have happened without it.
The wider cycle adds context. Tech Influence Watch, run by writer Molly White, counted $79 million in AI super PAC spending across 68 House and Senate seats. In Fortune's tally of the closest races, most Republicans backed by the AI spending fight lost. Separately, voters rejected AI data centers in 40 of 41 ballot measures, based on the reporting in the source.
Taken together, these outcomes put pressure on a narrow strategy: spend big, back candidates, and expect that investment to translate into friendlier policy. They don't show that AI policy is settled, or that every voter is against AI. But they do suggest money can't substitute for public trust, credible local engagement, and a clear explanation of what a proposed technology or facility would actually mean.
Why the race became a symbol
The campaign became shorthand for a bigger fight over who gets to set rules for AI and data infrastructure. That kind of symbolism is useful for spotting political risk, but it can also blur the details. A Senate race includes many issues, and voters don't reduce their decisions to one label or one PAC's spending.
How the AI political spending networks differ
The spending described in the source came from two main networks, each tied to a different approach to AI policy. Leading the Future pushes for lighter regulation and a single national AI rulebook that would override state-level laws. Its major donors include Andreessen Horowitz co-founders Marc Andreessen and Ben Horowitz, who each gave $25 million, according to filings reviewed by Fortune. OpenAI president Greg Brockman and his wife, Anna, were also among its largest donors.
Public First, backed by Anthropic, argues for stronger safeguards. Both networks run political operations that back candidates from both major parties. That matters, because this isn't just a straight party-versus-party fight. The spending groups are trying to shape the policy environment across party lines, even while their candidates compete in individual districts and states.
Other groups are in the mix too. Alliance for a Better Future, described as a conservative AI group, backed candidates including Mike Rogers in Michigan and Susan Collins in Maine. Guardrails Alliance, a tech worker group pushing for stricter rules, spent against candidates including Aaron Flint and California Representative Jimmy Gomez. These groups differ in ideology, policy priorities, and tactics, so you shouldn't treat their spending as one coordinated industry campaign.
It also helps to remember that voters don't experience labels as labels. Terms like "light regulation," "safeguards," and "innovation" are broad. The arguments get real when they address practical questions: Can a state regulate a data center's water use? What do developers have to disclose about risks? Would a national standard prevent a patchwork of conflicting rules?
Leading the Future launched in August 2025 with a reported war chest of about $100 million. WIRED reported that it ended the cycle with just over $20 million remaining. That kind of money gives a network room to keep investing - but a big balance isn't proof of political influence. The harder test is whether spending changes candidate behavior, voter understanding, or legislation.
The networks' different approaches also create a measurement problem. A candidate might win despite a PAC's spending, lose for reasons unrelated to AI, or adopt a position only partly aligned with the group backing them. Campaign finance is one input into politics, not a direct readout of policy outcomes.
Why national rules are part of the dispute
A federal framework could make compliance more predictable for companies working across state lines. It could also limit states from responding quickly to local risks - or when federal rules lag. Stronger safeguards might close those gaps, but they could also raise costs, add uncertainty, or create inconsistent requirements if they're poorly designed. The real policy question isn't "regulation or innovation." It's which risks get managed, at what level of government, and how accountability works.
Where the money went and what it achieved
The biggest spending clash in the cycle came earlier in the Democratic primary for New York's 12th District. Leading the Future spent more than $8.1 million opposing Assemblymember Alex Bores, a co-author of New York's RAISE Act. Public First spent about $15 million supporting him. Bores finished second to Micah Lasher, who also campaigned on AI safety.
That result is a reminder that PAC spending can target a candidate without locking in the final policy direction. Bores lost - but the winner also ran on AI safety. In other words, a political group can meet a narrow goal - defeating one candidate - without winning a broader fight for its preferred policy agenda.
In the general election, Leading the Future spent about $10 million, mostly on safe seats, WIRED reported. Its Republican arm, American Mission, spent $590,428 backing Aaron Flint in Montana's 1st District and $509,067 backing Laurie Buckhout in North Carolina's 1st District; both lost. Its Democratic arm, Think Big, backed Florida Representative Jared Moskowitz, who won.
Public First's Republican arm had a tough night. Defending Our Values spent $1.49 million supporting Carlos De La Cruz in Texas's 35th District and $974,908 backing Rhett Marques in Alabama's 2nd District; both lost. It also backed Scott Singer against Moskowitz with $857,426. Public First's Democratic arm, Jobs and Democracy, backed two winners in Ohio and Pennsylvania.
Other groups had mixed - or worse - results too. Alliance for a Better Future spent just over $1 million backing Mike Rogers in Michigan and $810,000 backing Susan Collins in Maine; both lost. Guardrails Alliance spent more than $1 million each against Flint and Gomez. Gomez lost to Angela Gonzales-Torres, but that outcome by itself doesn't prove Guardrails Alliance's spending caused the result.
The key point isn't that every AI-backed candidate lost. Moskowitz and some candidates in safe seats won, and some spending went toward races that weren't close to begin with. The real question is whether those wins translated into durable support for a policy position - or simply added money to races that were already likely to succeed. If you skip that distinction, you can make a political operation look more effective than it is.
Spending is easier to count than influence
Dollar totals are straightforward. Persuasion is not. If you want to judge a political investment, you have to ask whether it shifted an election margin, changed a candidate's position, shaped a legislative vote, or improved public understanding. Those outcomes don't always move together, and an operation can "succeed" on one measure while failing on the others.
What the results do and do not say about voters
These repeated losses are a warning for AI-backed campaigns, not a clean measurement of public sentiment about AI in general. Voters have a lot of reasons for their choices. Election results bundle together the economy, party identity, incumbency, local issues, and campaign quality. A candidate losing doesn't prove voters rejected every part of that candidate's AI platform.
Ballot measures give a more direct look at attitudes toward data center development - but they still require careful interpretation. Voters rejected measures tied to AI data centers in 40 of 41 cases, as reported in the source. That points to strong resistance in those specific local situations. It doesn't automatically mean people oppose all data centers, all AI applications, or every proposed facility. It also doesn't tell you whether the concerns were mostly about energy, water, land use, tax incentives, or trust in developers.
Still, local opposition should matter to companies and policymakers. A data center isn't "just software." It's physical infrastructure - power demand, water systems, construction impacts, municipal planning. When residents think they'll carry the costs while the benefits go elsewhere, a national message about technological leadership usually won't settle the dispute.
The campaign messaging itself may also have limits. A Republican lobbyist, Greta Joynes of Brownstein Hyatt Farber Schreck, told Politico that AI was unpopular and that "beat China" didn't land where it was supposed to. That doesn't prove one single cause for the losses, but it does point to a practical communications problem: arguing about international competition may not answer voters' questions about local costs or safeguards.
Public First and Guardrails Alliance both framed the results as a rejection of industry influence and anti-regulatory efforts. Brad Carson of Public First Action called it "a crushing blow" to attempts to buy a Congress that would look the other way. Alex Bores said aligning with parts of the AI industry had become a political liability. Those are arguments - not neutral analysis. They should be judged against the same evidence supporters and opponents point to.
The most defensible takeaway is narrower: visible political spending didn't consistently deliver the electoral outcomes its sponsors wanted, and opposition to specific data center proposals showed up across the reported ballot results. Whether that becomes lasting support for stricter AI rules will depend on what candidates, companies, and lawmakers do next.
The difference between AI and AI infrastructure
Public debate often throws AI models, applications, and data centers into one bucket. They're connected, but they raise different concerns. Risks tied to model testing and transparency don't play the same role as zoning, grid capacity, and water use. When political strategy treats everything as one "pro-AI" or "anti-AI" label, it gets harder to respond to what people are actually objecting to.
What AI companies should learn before the next election
The first lesson: campaign spending can't fix a weak relationship with the people who live with a project or policy. In real-world rollouts, trust gets built long before voters feel urgency. If companies announce infrastructure plans late, float vague economic benefits, or treat local concerns as misinformation, opponents get room to define the issue.
Second, separate policy advocacy from candidate promotion. A company or PAC might want a national standard, a specific safety requirement, or a predictable permitting process. Those are policy goals. Supporting a candidate is a different move, with different reputational risks. When voters connect a candidate to a powerful industry, that association can become part of the campaign story - even if the candidate's actual position is more nuanced.
Third, national messages still need local evidence. "Innovation," "jobs," and "beating China" don't answer the questions a community asks when it evaluates a data center. People want credible numbers on power demand, water use, tax arrangements, construction impacts, and expected employment. Those claims need to be concrete enough to check later. If estimates shift, say why. A promise you can't verify is more like a short-term communications tactic than a foundation for long-term support.
Fourth, policy design has to deal with the trade-off between uniformity and local control. A federal framework can reduce conflicting state requirements, but it shouldn't assume every local concern is an obstacle to progress. State and local rules can respond to specific conditions, but fragmentation can also raise compliance costs and make planning harder. The right question is which decisions should be handled with consistent national standards - and which need local authority.
Finally, measure outcomes beyond election night. Leading the Future said before the vote that primaries were a better use of its money because they could build candidate name recognition and help set AI agendas. The group also said it would keep spending in 2028. That plan should be judged by more than who wins. Did candidates adopt durable positions? Did lawmakers write clearer rules? Did public confidence improve? Did local projects earn informed consent?
Bores is preparing for the next cycle with a new nonprofit, Who Decides, which Fortune reported aims to raise $30 million before 2028. That's a signal the political fight is still ongoing. The next round is likely to bring more organized spending - and more careful efforts to make AI policy visible to voters.
A practical test for political investment
Before spending on a candidate or policy campaign, an AI organization should be able to answer four questions:
What specific outcome are we trying to change? Separate an election result from a legislative or public-understanding goal.
Who bears the practical consequences? Include local communities, workers, users and public services - not only companies and investors.
Can we support the message with verifiable evidence? Use concrete claims that can withstand scrutiny after the campaign.
What is the downside if the association backfires? Consider whether a candidate may become more vulnerable by being visibly aligned with the industry.
If those answers are unclear, raising the budget is unlikely to fix the underlying problem.
Key takeaways
The $4.8 million South Carolina bet failed, but one race doesn't explain the full political effect of AI spending.
AI-aligned groups spent a reported $79 million across 68 races, with mixed results across candidates and parties.
Rejections of data center measures point to local concerns about infrastructure, not necessarily opposition to all AI.
Money isn't a substitute for credible policy, clear local commitments, or public trust.
Campaign outcomes should be judged against the real goal: votes, legislation, candidate positions, or informed public consent.
Conclusion
The South Carolina loss is a visible setback for the AI industry's election strategy, but the overall picture is more complicated than a simple win-or-lose tally. AI political groups spent heavily. Candidates across the policy divide saw mixed results. And local voters repeatedly pushed back on data center proposals. That doesn't settle the debate over AI regulation. What it does show is that political influence doesn't buy results as reliably as media exposure.
For companies and advocates, the practical takeaway isn't just "spend more" before the next cycle. It's to define the policy outcome you want, understand who will be affected, and make claims that can be tested against what happens in practice. The next election will reward organizations that can explain not only why AI matters, but also how its costs, risks, and benefits will be managed.