Where AI money worked in the midterms

Three AI-related super PAC efforts shaped the election. Leading the Future spent $4.8 million supporting Republican Darline Graham’s Senate bid, but she lost to Democrat Annie Andrews. Its other general-election spending helped Rep. Jared Moskowitz win, while supported Republican House candidates lost.

Public First Action spent $20 million. Its network backed winners Valerie Foushee, Brian Poindexter, Manny Rutinel, and Scott Wiener. The Republicans backed by Public First’s network lost, giving the group a mixed result rather than a complete victory.

Guardrails Alliance PAC opposed five candidates and spent more than $3.1 million. All five lost, producing a 5-0 record. Yet the biggest AI-related result was outside candidate races: about 40 local measures restricting data center development passed. Leading the Future had not spent resources fighting those measures.

A super PAC is an independent political committee that raises and spends money to influence elections. Unlike a traditional campaign committee, it can generally accept unlimited contributions from individuals, companies, unions, and other groups. Its defining limit is coordination: it cannot legally coordinate spending decisions or campaign strategy with candidates or their campaigns.

Super PACs influence voters through advertising and other independent communications. They can praise a candidate, attack an opponent, fund mailers, sponsor online messages, or support voter outreach. Spending is usually aimed at changing public opinion rather than directly giving money to a candidate.

The article shows this model in action. Leading the Future spent $4.8 million supporting Darline Graham, while Guardrails Alliance spent more than $3.1 million opposing five candidates. Their results varied, showing that large independent spending can shape campaigns but does not guarantee victory.

The article identifies several major spending figures, but they cover different scopes. Leading the Future spent $4.8 million supporting Darline Graham’s Senate campaign. Public First Action spent $20 million across the general election. Guardrails Alliance spent more than $3.1 million opposing five candidates.

Those numbers show that AI-related political spending was substantial but unevenly targeted. Leading the Future’s $4.8 million figure concerns one candidate, while Public First’s $20 million covered its general-election activity. Guardrails Alliance’s spending produced a clear 5-0 result because every candidate it opposed lost.

The most consequential count was not campaign spending. About 40 local ballot measures restricting data center development passed across the country on Tuesday. Those votes mattered because they directly addressed construction projects tied to the broader AI infrastructure debate, while the PACs mainly focused on candidates and national rules.

When voters approve restrictions on data center development, a proposed project may lose the local approval needed to proceed. The measure could block construction, limit where facilities may be built, impose conditions, or require a longer approval process. The exact consequence depends on the wording and legal authority of each local measure.

The article reports that about 40 such measures passed across the country. It does not specify what every measure did, so their individual effects cannot be stated precisely. It does show that the votes challenged plans by hyperscalers and became the election’s most salient AI-related outcome.

These measures can force companies to change projects or their public messaging. The article says hyperscalers were scrambling to find a way forward in certain jurisdictions. It also reports that Leading the Future was surprised and had put no resources into fighting the ballot measures, showing how local decisions can escape national campaign strategies.

“Hyperscalers” generally means very large technology companies that operate extensive computing, cloud, or internet infrastructure. The article does not name the companies. It says these firms were scrambling to find a way forward in jurisdictions where local voters approved restrictions on data center development.

New data centers are physical projects, not just software decisions. Companies need suitable land, electricity connections, water arrangements, construction approvals, and local government permission. They also need communities to tolerate the project’s effects and accept its economic or employment claims. A national AI strategy cannot automatically grant those local approvals.

That dependence explains the election’s tension. AI-focused PACs backed candidates and promoted a national regulatory framework, but about 40 local data center measures passed. Hyperscalers therefore had to rethink both their plans and their messaging. Winning national allies may help, but local governments and residents still control many practical conditions for construction.

Candidate races and national law operate on different tracks. Electing a preferred senator or representative may improve a group’s influence, but it does not guarantee agreement among lawmakers or passage of a specific regulatory framework. Legislation still requires congressional action, negotiation, and enough support to move through the process.

The article illustrates this gap. Public First Action backed winners including Valerie Foushee, Brian Poindexter, Manny Rutinel, and Scott Wiener. Guardrails Alliance also went 5-0 against its targets. Yet about 40 local ballot measures restricting data center development passed, despite the PACs’ candidate-focused spending.

Leading the Future said its investments aimed to build a durable coalition for a national framework. It also called the effort multi-cycle and pointed toward 2028. That language acknowledges that one election is only an early step. Candidate victories may create allies, but they cannot by themselves settle local conflicts or deliver national AI rules.

Modern AI systems rely on data centers for large-scale computing, storage, networking, and model development or operation. These facilities house powerful computers and supporting equipment. As AI use expands, companies seek more capacity, which means building or expanding physical sites in particular communities.

That construction can create conflicts over resources and land. Data centers may require major electricity connections, cooling water, large parcels, and industrial buildings. Their equipment can also create traffic, construction disruption, or ongoing noise. Residents and local officials may object to these effects, question who benefits, or demand a stronger say over development.

The article records the political result without detailing each project’s impacts. About 40 local measures restricting data center development passed, while hyperscalers revised their plans and messaging. This shows why AI infrastructure cannot be treated only as a national technology issue. Local approval and community acceptance can determine whether capacity gets built.