Jasmine Sun on What the AI Industry Got Wrong About the…

PodcastOdd LotsJasmine SunAug 21

Jasmine Sun argues that data-center opposition is driven less by abstract AI fears than by local costs, limited employment, distrust of corporate promises, and weak negotiating leverage for small communities. The AI buildout faces physical and political constraints, while potential cyber and biological risks within six to twelve months represent a separate, more national-security-focused concern. The economic case rests largely on property-tax revenue and temporary construction activity, but demand guarantees and concentration among the “Mag 7” underscore concerns about stranded costs and an AI market bubble.

Artificial Intelligence AI Infrastructure National Security Technology Stocks Market Valuations Cybersecurity

Preview · 18 more points in the app

Read the Full Summary

Key Takeaways

Data-center opposition is a bipartisan, locally grounded backlash: polling in Michigan and Wisconsin shows roughly 70-30 opposition among both Democrats and Republicans, while residents focus on two to six years of traffic and noise, electricity costs, construction disruption, and the projects’ large physical footprints. The backlash is strongest in communities directly affected by proposed facilities, even though opposition also exists outside host ZIP codes.

Communities question whether AI data centers deliver enough public benefit to offset their costs, citing limited permanent employment, diffuse tax benefits, and recurring distrust of corporate promises, utility-rate increases, layoffs, and unresolved liabilities. That skepticism helps explain why data centers face stronger opposition than solar farms, nuclear plants, power plants, battery storage, chip factories, and Amazon fulfillment centers.

The negotiating imbalance between small governments and technology companies is illustrated by the Celine Stargate Center case, where a 4–1 city council rejection of an Oracle and OpenAI project was followed by an exclusionary-zoning lawsuit and a rapid settlement. In response, communities are rejecting confidential negotiations and shifting toward upfront developer commitments rather than city subsidies in deals made from 2026 onward.

Sun became more sympathetic to statewide moratoriums because larger payments alone may increase suspicion of dark money or a speculative bubble rather than resolve opposition. She argues that a state-level pause could establish common standards for community benefits, environmental practices, renewable energy, and ratepayer protection that small towns may lack the leverage to secure.

AI infrastructure is constrained by physical and political bottlenecks—including power, land, water access, regulatory approvals, and local acceptance—rather than simply by capital availability. The scale of the buildout varies sharply: Michigan and Wisconsin each have roughly 50 to 100 projects or proposed projects, compared with approximately 600 to 800 active or under-construction facilities in Virginia and Texas.

GEO-POLITICS

Bipartisan Opposition — Opposition to data centers is emerging across party lines. Polling in Michigan and Wisconsin shows roughly 70-30 opposition among both Democrats and Republicans, suggesting data-center policy cuts across conventional partisan divisions.

“The other interesting thing about data centers and AI is that it's become 1 of the very few bipartisan issues, I think, in America…”—

Local Costs — Data-center opposition is driven less by abstract AI fears than by visible local effects, including traffic, noise, construction disruption, electricity costs, and infrastructure’s immediate community impact. Sun argues that the AI industry focused more on existential risks and future job displacement than on these local concerns.

“But on the ground, it does not feel like those things are the fears, even though I'm, you know, anxious, like, that are really, like, driving the poll the backlash…”—

Electoral Flashpoints — Large data-center projects become potent local political flashpoints because construction can bring two to six years of traffic and noise. Facilities covering hundreds of acres create a highly visible footprint, but data centers become top-tier electoral issues mainly in communities directly affected by proposed projects.

“And my sort of summary here is that I think that if you live in a community where a data center is being built or proposed, you don't like it…”—

TECHNOLOGY

Physical Bottlenecks — AI infrastructure faces unusual physical constraints for the Internet economy, including shortages of power, land, and water access, along with regulatory and political approvals. The constraint is not a shortage of capital. Data centers need proximity to power and water infrastructure, limiting viable sites and intensifying local conflicts.

“There are physical constraints to the AI build out, which basically in the history of the Internet is totally novel.”—

Regional Buildout — Michigan and Wisconsin each have roughly 50 to 100 data-center projects or proposed projects, far fewer than Virginia and Texas, which have approximately 600 to 800 active or under-construction facilities. Wisconsin has attracted investment partly through sales-tax exemptions on chips and some construction costs.

“I think each of them have maybe around like 50, maybe to a 100 projects or proposed projects in the state…”—

Industry Blind Spot — AI companies focused on speculative job displacement while underestimating backlash over data centers and child safety. Sun attributes this blind spot partly to Silicon Valley’s insularity: the industry focused on future capabilities and interacted mostly with people inside technology rather than experiencing how AI was already affecting other communities.

“But the thing is, like, these are the ways that AI materially shows up in, like, everyday people's lives. Right?…”—

ECONOMY

Limited Revitalization — Data-center development is being considered as a response to deindustrialization, including at former manufacturing sites such as the GM plant in Janesville. The main economic pitch is property-tax revenue: data centers make up more than half of Loudoun County’s property-tax base, while Microsoft’s Fairwater facilities are expected to pay about $20 million in property taxes in 2026. However, the model may not replace jobs lost from old factories.

“Even in some of these very specific sites where their data center is being proposed, they were like, you know, the site of a former GM plant in Janesville that…”—

Construction Jobs — Data centers can create 500 to 1,000 construction jobs for two to four years, but many workers come from outside the area and only about 50 permanent jobs may remain. Construction nevertheless generates well-paid skilled-trades work, apprenticeship opportunities, and per diems, although labor shortages limit how broadly local communities benefit.

“Sure. There are some jobs for sure, like 500, 1000 jobs, but those last 2 to 4 years. Sometimes they're outsiders. They're not people from the community…”—

Demand Guarantees — Data-center agreements may need tax guarantees or commitments to prepay as much as 80% of forecast electricity demand if AI demand weakens. Such provisions could protect local governments from stranded infrastructure and unexpected power costs, but they would shift substantial potential liabilities onto hyperscalers and operators.

“Like, I think tax guarantees are some of these or making deals to say we're gonna pay for 80% of our forecasted electricity demand upfront because sometimes those forecasts are…”—

MARKETS

AI Bubble Concerns — Even workers and businesses benefiting from data-center construction question whether the AI boom is sustainable, making communities wary of projects that may depend on tax revenue for as long as a decade. Market concentration reinforces those concerns: some participants said the “Mag 7” hold roughly 50% of the stock market and described the AI-driven market as a bubble.

“1 actually surprising tidbit that I heard was that from Terry to this guy who's a CEO of an electrical contracting company I talked to an HVAC engineer I talked…”—

AI’s Physical Buildout

The hosts explain why AI differs from earlier software booms, emphasizing shortages of power, land, water, approvals, and other physical infrastructure.

Bipartisan Data Center Backlash

Data centers have become an unusual bipartisan flashpoint, while local opposition focuses more on visible community costs than on speculative AI catastrophe.

Michigan And Wisconsin

Jasmine Sun describes her reporting trip through Michigan and Wisconsin, where incentives, grid access, deindustrialization, and political campaigns shaped data center proposals.

Local Opposition’s Salience

Opposition is especially intense near proposed facilities, though statewide polling also shows broad disapproval that is less electorally salient away from affected communities.

Trust Over Information

Sun finds that organized opponents generally understand data center technology; their resistance stems more from distrust of utilities, corporations, and government promises.

Property Taxes, Few Jobs

The economic pitch centers on property-tax revenue, while construction jobs are temporary, permanent employment is limited, and AI’s benefits feel distant to residents.

Power Asymmetry In Deals

Communities are demanding more upfront benefits and transparency, but small governments remain outmatched by hyperscalers’ lawyers, resources, and ability to threaten litigation.

Workers And Bubble Fears

Tradespeople welcome construction work but represent a small constituency, while even supporters question whether the AI boom is sustainable enough to justify local risks.

Why AI Feels Different

Compared with factories, solar farms, and other infrastructure, data centers lack a compelling public benefit, leaving residents unsure why they should accept the costs.

State-Level Solutions

Sun favors statewide standards for community benefits, environmental protections, and ratepayer guarantees because small towns lack the capacity to negotiate balanced agreements.

AI Risks Beyond Backlash

The industry underestimated immediate public concerns, while insiders increasingly focus on cyber and biological risks that could emerge within six to twelve months.