The ballot box just met the data center

Good morning. For two years the AI buildout has been a story about chips, capital and power. Last night it became a story about voters. In Texas, Ohio and Georgia, people went to the polls and chose candidates and ballot measures that would slow the data centers going up near their homes, and the AI trade woke up to a new kind of risk. Plus, Apple put rogue AI agents in its list of business risks, OpenAI says ChatGPT’s search traffic is now about a fifth of Google’s, and OpenAI brought in outside hackers to test its defenses. Here is what moved while you were away.

Axios reports that voters showed deep distrust of the AI race on Tuesday, approving dozens of data center restrictions and electing a wave of candidates who favor slowing the buildout. The biggest shock came in Texas, where Democrat Gina Hinojosa won the governor’s race after campaigning to keep a moratorium on new data centers until the state passes rules on water, electricity, air and property rights. Texas accounts for 41% of new U.S. data center capacity expected in 2028, according to SemiAnalysis, and Democrats also won Ohio, Georgia and Arizona, the next three states on that list. Of 41 data center measures on ballots in 10 states, about two dozen designed to restrict or block development appeared to pass, according to Ballotpedia. One advocate called it an AI earthquake.

Fortune reports that Ohio rejected the AI buildout almost everywhere it appeared on the ballot, with at least 14 municipalities passing bans or other limits on data centers. Sunbury appears to have banned data centers that use 25 megawatts or more, Trenton passed its ban 80% to 20%, and Ashville voted to block a campus that is already under construction. Ohio already has 125 local moratoriums, and a state poll last month found 55% of residents opposed new data centers. Ohio may be the clearest preview of how communities respond when the AI boom arrives next door.

CNBC reports that SpaceX and Tesla shares fell close to 4% in premarket trading on Wednesday after big losses for Republican candidates Elon Musk had backed, including Gov. Greg Abbott in Texas. The broader AI trade was mixed. Neoclouds CoreWeave and Iren slipped and Oracle dipped, while Nvidia, Alphabet, Meta, Amazon and Microsoft rose about 1% to 2% and Nasdaq-100 futures were flat. Wells Fargo estimates that 27% of planned U.S. data center gigawatts sit in Texas. Musk’s reply on X was that data centers in space were looking pretty appealing right now.

CNBC reports that Apple added new warnings about AI agents to its annual filing, saying agents may behave in unintended or unpredictable ways and take actions that were never directed or authorized. Apple said agents can access systems they should not, alter code, send messages and complete transactions, and that its ability to oversee agents built by other companies may be limited. Apple does not yet ship an agent of its own, though new CEO John Ternus said on Monday’s earnings call that Siri has real room to grow. When the most careful company in tech puts rogue agents in its risk factors, the rest of the industry should take note.

Big Technology reports that OpenAI now puts ChatGPT’s volume of search-like queries at roughly one-fifth of Google’s. Consumer product lead Adam Fry said at a dinner in New York that ChatGPT grew from 1 billion to 1.2 billion weekly users in about two months, that people create 1.5 billion images a week, and that 250 million people use its voice features weekly. Fry counts a search as any question that sends ChatGPT to the web, and the comparison is OpenAI’s own estimate. Google is still far bigger, but the gap is no longer measured in orders of magnitude.

TechRadar reports that OpenAI has partnered with Cathedral Systems, whose researchers will work alongside OpenAI’s own teams to probe its infrastructure, find weaknesses and test fixes using OpenAI’s models. The company is also raising its bug bounty, with payouts for critical findings set to roughly quadruple. OpenAI said traditional cyber defenses alone are no longer sufficient. The labs building the most capable models are now racing to use them on their own front doors first.

---

Deals Desk

The newest compute and power deals on our tracker:

• Nov 3 — Brookfield and Radiant × Drax: A partnership to develop and operate AI data center capacity on the Drax Power Station estate in North Yorkshire, England. Phase 1 targets up to 100 MW from early 2028, with an ambition to reach 500 MW by 2031 and more than 1 GW in time. The partners put total potential investment at more than £20 billion, with £2.5 billion for phase 1, which still needs planning approval and a tenant. Drax

• Nov 3 — Etihad GO Telecom × ezditek: The Saudi telecom company will lease dedicated data center space for AI services from ezditek for five years, for total fees of SAR 76.85 million, about $20.4 million, excluding VAT. Capacity was not disclosed. ETDatacenters

Full tracker → theaigentic.com/deals

---

Yesterday’s movers

Biggest winners

Biggest losers

---

Watch

Four picks this morning: Latent Space inside the software factory at Warp, Peter Yang letting an AI agent haggle with his cable company, The Koerner Office with a 60-year-old founder selling AI assistants to local businesses, and Alex Finn on squeezing more out of your agents.

Latent Space — The 2026 Software Factory Infra Stack, with Zach Lloyd of Warp

Latent Space host swyx sits down with Zach Lloyd, the founder and CEO of Warp, for a look at the system his team uses to turn tasks into pull requests.

The demo follows a request from Slack all the way to a finished PR, then shows how Warp shares agent traces, scores cost and throughput, and builds benchmarks from its own engineering work. Lloyd says tuning models and subagents cut Warp’s internal cost from about $80 per pull request to roughly $15 to $20.

Watch this if you run coding agents and want a real way to measure whether they are getting better or just getting busier.

Watch on YouTube →

Peter Yang — This New AI Feature Just Saved Me $500 a Year

Peter Yang hands a phone call to an AI agent and plays the whole thing back, from the phone tree to the small talk.

His agent called Comcast and negotiated his internet bill down from $84 to $45 a month, while fending off three upsells along the way. There is a twist, though, because the bot still needed him at one key moment, and Yang walks through which calls are safe to hand off and which are not.

Watch this if you have a stack of customer service calls you have been avoiding.

Watch on YouTube →

The Koerner Office — Forget Building Websites. Local Businesses Pay $4K a Month for This

Chris Koerner talks with Bob Brennan, a 60-year-old entrepreneur who built an AI concierge business bringing in $25,000 to $30,000 a month in about a year.

Brennan builds chatbots and voice assistants that help small businesses catch more leads, and he explains how he finds clients, offers free 30-day trials and prices his work on the value of the leads it brings in.

Watch this if you think the AI money is only in Silicon Valley. Some of it is in the local dentist’s office.

Watch on YouTube →

Alex Finn — How to Get Way More Out of Your AI Agents

Alex Finn has a practical tune-up for anyone running a crew of AI agents that has started to sprawl.

He audits which bots he actually uses, fixes the routines that waste effort, cleans out group chats, and shows how giving an agent the right tools, skills and plugins gets more done with less.

Watch this if your agent setup has grown faster than your ability to keep track of it.

Watch on YouTube →