Pennsylvania voters move AI data centers to the top of the ballot
Two fresh polls, a Financial Times dispatch from the state, and a governor's course-correction all point at the same story: AI data centers have stopped being a zoning story and started being a ballot-box story in Pennsylvania, and Kimi K3 just topped a new domain-specific reasoning benchmark aimed at geoscience.
Two new Pennsylvania polls put data centers at the top of voter concerns, and the issue is now reshaping both the governor's race and the down-ballot legislative map. A Franklin & Marshall poll, reported on by the Erie Times-News, found 79% of Pennsylvania voters do not want a data center built in their community — the first time the F&M poll asked the question — and 84% want AI companies more closely regulated by the federal government, up from 79% in October. A separate New York Times/Philadelphia Inquirer/Siena University poll found 60% of likely voters oppose data center development statewide, with 83% of likely voters ages 18 to 29 against — a generational gap large enough to flip the next decade of state politics. The Financial Times dispatched a feature to the state on the same backlash, quoting residents worried about electricity bills, water use, and the noise of hyperscale campuses, and the New York Times' coverage names rising utility bills as the proximate trigger. Both Republican Stacy Garrity and Democratic Governor Josh Shapiro have now walked back earlier support for data center build-outs; Shapiro's August executive order restricting development and his November gubernatorial opponent now both call for a temporary pause on new projects. The same backlash is spreading to the legislature: a companion bill to restrict data centers' access to public-utility status passed the Pennsylvania House 201–1.
The political weight of this is bigger than the data centers themselves. Pennsylvania is the swing state that decided 2016 and 2024, and four congressional seats Democrats are targeting for the House are now running on a data-center record. The current buildout is six hyperscale campuses under construction, a 2-million-square-foot project near King of Prussia, and the Homer City repowering of a former coal plant. Each one was a back-room deal a year ago; each one now has an organized local opposition, an audience with the state legislature, and a polling number attached to it. The next phase of the AI buildout is no longer being negotiated only with utilities and county commissioners.
Kimi K3 has topped the new Groundtruth benchmark for AI models on geological reasoning — the first domain-specific eval aimed squarely at applied geoscience. The Groundtruth benchmark, published by Eigenform AI with infrastructure partner MatchPoint, was designed to stress the layered thinking stratigraphy, petrology, and structural geology require, and a single-domain leaderboard is the natural next step as the broad MMLU/Chatbot Arena race plateaus. Eigenform's press release puts Kimi K3 on top, with a breakdown of per-question costs that should matter to anyone running these models against seismic data, mineral-exploration workflows, or compliance reporting. The unverified bit matters: the scoreboard is press-release-coded today, and we don't yet have prompts, context lengths, or full leaderboard numbers. The interesting test for Kimi K3 isn't the score — it's whether Moonshot's model actually links to the GIS platforms, seismic software, and internal knowledge bases geoscience teams already run, which is the difference between a benchmark and a workflow.
The bigger market read is that narrow, demanding scientific domains are now where the next AI infrastructure dollar is going to go. Petroleum exploration, mining, and environmental assessment are trillion-dollar markets where specialized reasoning can shave real money off exploration risk and operational cost. Kimi K3 is the first major model launch framed around that thesis; it will not be the last.
Google's AI Mode is showing shoppers 21.6% higher prices than traditional search for the same products, according to a new Productrise data study. The Productrise team matched products across identical queries on the same day and found AI Mode lists the same item at a higher price more than two-thirds of the time, with a different seller on nearly half of matched products. The seller swap is the bit that should worry retailers most: AI Mode isn't just re-ranking the same offer — it's picking a different one on its own criteria, likely weighing feed data, availability, third-party citations, and reviews alongside the merchant's own feed. Prices agreed about 68% of the time even when the lead seller was different, but when they disagreed AI Mode was the expensive side two-to-one. Google now surfaces AI Mode as a prominent button directly on the standard search box and in the Chrome address bar, so most shoppers reach it without choosing to. The finding follows Productrise's July study showing AI Mode surfaces fewer shopping products overall.
If data-center backlash becomes the template for how US states handle AI infrastructure, what does the next decade of buildout look like — hyperscale retreat, or just relocating to less-organized states? Tell us in the comments.
Sources: Financial Times — Pennsylvania voters unite against data centres · Erie Times-News — What Pennsylvania voters think about AI and data centers · Philadelphia Inquirer — 6 in 10 Pennsylvania voters oppose data center development · Capital & Main — In Pennsylvania, the Data Center Wars Hit the Ballot Box · EIN Presswire — Kimi K3 Tops New Benchmark of AI Models for Geological Reasoning · Eigenform — Groundtruth benchmark · Productrise — Google AI Mode shows the same products 21.6% more expensive than traditional search
