American voters just handed the AI industry a problem it cannot build its way out of — at least not at home. On November 3, communities across 10 states voted on 45 local ballot measures targeting AI data center construction. As of Wednesday morning, 33 had passed. If you build software on cloud infrastructure, that number matters more than any earnings report this week.
What Actually Happened on Tuesday
The measures varied in form but shared the same message. Some communities voted for outright bans — Alhambra, California became the second city in the state to prohibit data centers entirely. Others passed temporary moratoriums to pause permitting while local governments figure out what they are actually agreeing to. Ohio, which had the most measures on the ballot (21), passed restrictions by overwhelming margins: Urbana voted 76% in favor of a ban; Hubbard hit 77% against allowing new facilities. Grove City added language requiring voter approval for any project drawing more than 20 megawatts or consuming more than 500,000 gallons of water daily.
Thirty-three of 45 measures passed. Six more were still being counted. Bloomberg tracked 45 measures across 10 states; the Forbes full list covers every jurisdiction. This was not a close contest.
Why Voters Are Saying No
It is tempting to frame this as technophobia. That would be incorrect. The numbers behind the objections are real. A typical hyperscale data center draws 100 megawatts — the equivalent of 100,000 homes. Newer AI training facilities draw up to 2,000 megawatts. The Lawrence Berkeley National Laboratory projects data centers could account for 12% of all US electricity consumption by 2028. For communities already stretched on grid capacity, that is not an abstraction.
Water is the sharper concern. US AI data centers are projected to consume up to 32 billion gallons of water annually by 2028, almost all of it evaporated, not recycled. One project caused a noticeable drop in local water pressure. When a community votes to require public hearings before permitting a facility that might consume half a million gallons of water per day, they are making a reasonable demand for information — not staging a revolt against technology.
Exit polling found 56% of voters considered AI data centers a crucial or major factor in their congressional vote. A UMass Amherst poll found only 11% of Americans would support a data center being built in their community. These numbers have been building for months. Tuesday was the first formal expression.
The Supply Problem This Creates
US cloud infrastructure was already under pressure before Tuesday. Between 30% and 50% of data centers planned for 2026 are expected to be delayed or canceled — not because of voters, but because of transformer shortages, power grid interconnection queues, and construction cost inflation. Community opposition is now a third variable stacked on top of those two.
The practical effect is consolidation. Northern Virginia already handles 41 to 43% of all AWS traffic through us-east-1. That concentration was a known risk; it just became harder to dilute. The fallback sites — West Texas and Columbus, Ohio — were selected for access to cheap power, not proximity to users. That tradeoff shows up as latency. Azure has been capacity-constrained in several regions for months, with some AI model deployments on waitlists. The November 3 results are unlikely to improve that backlog on any near-term timeline.
Where Big Tech Is Actually Building
The response from large cloud providers is already visible in the investment announcements. Google announced $15 billion in Malaysia on November 4 — the day after the votes — covering new campuses in Selangor and Negeri Sembilan, plus more than a gigawatt of new clean energy capacity. Microsoft committed $5.5 billion to Singapore. Amazon pledged $6.2 billion to Malaysia. Oracle added $6.5 billion for a Malaysian public cloud region.
This is not coincidence. It is a direct routing-around of US community resistance and domestic power availability constraints. Southeast Asia offers land, government cooperation, and cleaner permitting timelines. For developers building global applications, more Asia-Pacific availability zones are coming in 2027 and 2028.
What Developers Should Take From This
The mental model of “cloud capacity is effectively infinite and US-based” needs updating. The practical implications are not dramatic today; they become a planning consideration over the next 24 months.
Concentration in Northern Virginia is a risk that is now harder to hedge. AI inference workloads that require low-latency US-East are pinned to an already-crowded region. New US availability zones in useful locations — dense population, existing network infrastructure — face permitting environments that are measurably more hostile than a year ago. The $10 trillion AI infrastructure buildout that analysts have been projecting assumes a permitting environment that no longer exists.
The industry will adapt — it is already adapting, as the Malaysia announcements show. But adaptation takes time, and the gap between demand and available domestic capacity is real. If you are designing systems that depend on specific cloud regions, the assumption that new capacity will appear in the US on the schedule you need is worth revisiting.
