Data Center Reality Check
My Prompt 2026-09-24; ChatGPT Sol 5.6:
What's the reality of negative vs positive impacts of new data centers and how much are US paid campaigns and foreign campaigns influencing opinions of the segment of the general public that doesn't regularly follow the details. What are AI labs offering to mitigate local concerns.
ChatGPT Results:
Yes. There’s a fairly clear picture now, and it’s more nuanced than either “data centers are an economic bonanza” or “AI is draining America dry.”
### 1. The physical impacts are real — but enormously site-dependent
Electricity is the biggest legitimate concern. Berkeley Lab’s 2026 update estimates U.S. data centers could consume about 11.8% of all U.S. electricity in 2030, with a plausible range of 9.5%–15.3%, up from about 4.4% in 2023. That is not a rounding error; it represents hundreds of terawatt-hours of new demand. ([Data Center Energy][1])
Whether that raises your electric bill is a different question. Research is still developing and doesn't all agree. A June 2026 MIT/Duke/NCSU study estimated that data-center entry historically raised average retail electricity prices about 2.7%, including 2.1% residentially. A newer CEPR paper finds larger effects from hyperscale facilities. But Rutgers researchers examining utility territories found no statistically significant rate increase in their period of study. Virginia's nonpartisan legislative auditors reached a particularly useful middle-ground conclusion: data centers were paying their allocated current costs, but the huge amount of new generation and transmission required could raise everybody's future costs unless rate design changes. ([CEEPR][2])
So a statement such as “data centers always make residential electricity expensive” is too strong. But “poorly structured utility arrangements can shift significant costs onto households” is well supported.
Water is similar. The scary national numbers tend to obscure the real issue. Compared with agriculture or total U.S. water consumption, data-center water use is not dominant. But water isn't a national bathtub; it's local. A large evaporatively cooled facility in a stressed aquifer is quite different from an air-cooled or closed-loop facility in a water-rich region. Virginia's JLARC investigation found most centers there used water comparable to or below a large office building, while a minority used vastly more; it specifically warned that some local areas could face constraints even though Virginia overall is water-rich. ([JLARC][3])
That's why I would distrust both “they're draining our water supply” as a blanket claim and “data centers use hardly any water” as a blanket rebuttal.
### 2. The economic benefit is real, but the word jobs gets abused
Construction is genuinely a big economic event. Virginia's audit found a typical large center can put around 1,500 people on site at peak construction, and the state's industry supports extensive construction, supply-chain and service activity.
But after the building is finished, these are unusually capital-intensive operations. JLARC's representative 250,000-square-foot facility had only about 50 full-time workers, roughly half contractors. The audit explicitly concluded that most of the economic impact occurs during construction rather than ongoing operation. ([JLARC][3])
Tax revenue can be excellent or disappointing depending almost entirely on the deal. In mature Virginia markets, data centers contributed anywhere from less than 1% to 31% of total local government revenue. That's substantial. Yet Virginia's state sales-tax exemption cost roughly $928 million in FY2023. ([JLARC][3])
Georgia provides the other side of the subsidy problem. Its official 2025 evaluation concluded that about 70% of data-center construction likely would have happened even without the exemption—meaning much of the subsidy may not actually have been decisive in attracting the projects. ([Georgia Audits][4])
So the economic reality is roughly:
A data center can be a fantastic taxpayer and a mediocre employer at the same time.
And there's another distinction I think gets lost: the national benefit and local benefit are not the same thing. AI compute, cloud services, banking, medical records, communications and other digital infrastructure produce broad benefits. The host county, meanwhile, cares about its own electric bill, roads, aquifer, noise and property-tax base. Both viewpoints can simultaneously be rational.
---
## 3. Public opinion really has moved — dramatically
This isn't just anecdotal.
Pew surveyed more than 10,000 Americans this summer. From January to August 2026, the percentage saying data centers are mostly bad for:
the environment rose *39% → 54%**
home energy costs rose *38% → 50%**
nearby quality of life rose *30% → 49%**
local jobs rose *15% → 24%**
local tax revenue rose *12% → 21%**
And 60% said they would be uncomfortable with a data center operating in their area. Importantly, 18%–32% were still unsure about individual impacts, and 12% hadn't heard of data centers at all. ([Pew Research Center][5])
That last group—and the much larger group that has “heard something about them” without following utility regulation, cooling designs, tax abatements, etc.—is precisely the population you're asking about.
There is experimental evidence that their views are quite responsive to concrete information. A recent Nature Communications study found that a hypothetical project causing a 20% electricity-price increase lost roughly 20 percentage points of support; high water strain cost about 13 points. Even 250 promised local jobs recovered only around 10 points. ([Nature][6])
That suggests people aren't merely reacting to the words AI or data center. Electricity prices and water are potent cues.
---
# 4. How much are U.S. paid campaigns moving that less-engaged public?
There's a lot of money flying around now.
NPR's analysis of AdImpact records found that candidates and outside political groups had spent more than $45 million in 2026 ads mentioning data centers, with both Democratic and Republican campaigns frequently attacking tax breaks and electricity-cost impacts. ([Northwest Public Broadcasting][7])
On the pro-development side, the group Build American AI entered the fight this summer with about $50 million cash on hand, planning paid advertising, grassroots work, research and public education beginning in states including Kansas, Ohio and Wisconsin. ([Axios][8])
Environmental organizations and newly created anti-data-center coalitions are organizing nationally as well; Food & Water Watch, for example, is explicitly running a nationwide “Stop Data Centers” campaign and coordinating local organizing and legislative efforts. ([Food & Water Watch][9])
But here's the important caveat:
We do not have good evidence allowing someone to say “X% of Americans oppose data centers because of paid campaigning.”
Campaign advertising research generally finds that a single ad produces surprisingly small persuasive effects. One large series of randomized experiments testing 49 presidential ads found an average vote-choice effect of only about 0.7 percentage points, and decades of field experiments find that persuasion often decays quickly. Even a huge eight-month digital campaign targeting two million moderate and low-information voters in 2020 found no measurable average turnout effect. ([PubMed Central (PMC)][10])
What advertising can do more readily is make an issue salient, provide a simple mental model and repeatedly supply the same explanation:
> “My electricity bill is up → data center.”
or
> “China is beating us in AI → we need data centers.”
Once somebody has that framework, subsequent news gets interpreted through it.
So my evidence-based interpretation is that domestic campaigns are probably amplifying and organizing opinions more than manufacturing them from nothing. The striking thing about the polling is that opposition has risen among Republicans, Democrats, rural residents, suburban residents and urban residents simultaneously. ([Pew Research Center][5])
That's hard to explain as one partisan campaign successfully programming everyone.
---
# 5. Foreign campaigns: real activity, much less evidence of real influence
This part has become politically charged, so the distinction between existence and effectiveness matters a lot.
There is now documented Chinese-origin influence activity around data centers.
OpenAI identified and shut down a likely PRC-origin operation it called “Data Center Bandwagon.” Operators posed as Americans and generated posts and cartoons about data centers, electricity prices and grid strain. But OpenAI found “no evidence of meaningful breakout beyond its own activity.” In other words: an influence operation existed, but it apparently failed to get meaningful organic traction. ([OpenAI][11])
Separately, X reported finding roughly 200 accounts associated with a suspected Chinese bot network posting material intended to manipulate the American AI/data-center debate. ([Axios][12])
There are congressional investigations into allegations that foreign money may also be reaching U.S. advocacy networks. Those are investigations and allegations, not proof that American grassroots opposition is foreign-directed. A review published yesterday by PolitiFact found evidence of Chinese attempts to amplify the issue but concluded that characterizing the broader U.S. backlash as essentially a Chinese PR operation exaggerates the evidence. ([PolitiFact][13])
And the broader academic literature should make us cautious about attributing magical powers to foreign bots. One of the best studies of Russia's much larger 2016 Twitter operation found that exposure was highly concentrated, dwarfed by American political/news content, and showed no measurable meaningful relationship with changes in political attitudes or voting behavior. ([Nature][14])
So, as of September 24, 2026, I'd separate the propositions this way:
Foreign actors are trying to exploit the data-center controversy: documented.
They are inserting misleading or selectively framed material into the debate: documented.
They caused the broad American turn against data centers: presently unsupported.
That last distinction is pretty important.
---
# 6. What AI companies are now offering communities
This may actually be the biggest development of 2026. Companies have realized that generic “jobs and innovation” talking points aren't sufficient, and their commitments have become much more specific.
OpenAI now says each Stargate site will have an individually negotiated community plan. Its commitments include paying its own incremental energy costs, supporting special large-load utility tariffs, favoring closed-loop or low-water cooling, disclosing water/electricity information, workforce apprenticeships, and holding projects accountable for promised incentives and benefits. ([OpenAI][15])
Anthropic has gone unusually far on electricity: it says it will cover electricity-price increases attributable to its data centers, pay 100% of required grid interconnection upgrades, bring additional generation online, and use curtailment/grid-optimization measures during tight conditions. ([Anthropic][16])
Microsoft's five-point 2026 Community-First commitment is very concrete: prevent its centers from increasing residential electric rates; replenish more water than it consumes; create local jobs; expand the tax base; and invest in local training/nonprofits. ([Microsoft Blog][17])
Google says it will pay 100% of electricity and infrastructure costs directly caused by its growth. On water, it now promises to replenish more than it consumes by 2030, avoid water cooling in high-risk watersheds, publicly report data-center water use, expand reclaimed-water use, and has committed more than $500 million to local water/wastewater/reuse infrastructure. ([blog.google][18])
Meta says it pays its full energy and water infrastructure costs, has invested more than $550 million in community water/wastewater infrastructure, and funds water-restoration projects. ([About Facebook][19])
And Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI have all signed the federal Ratepayer Protection Pledge, under which they agree to obtain new generation, pay for associated delivery/grid upgrades, and use separate “take-or-pay”-style arrangements so ratepayers are not left with stranded infrastructure if anticipated data-center load doesn't materialize. ([The White House][20])
The important question with all of these isn't merely “Did they make a pledge?” It is whether the promise is written into the utility tariff, zoning approval, development agreement or community-benefits agreement in a form that survives corporate leadership changes and a 20-year operating life.
That's the difference between PR and risk transfer.
### So where does that leave the overall picture?
I would characterize the current evidence this way:
The backlash has a real factual foundation, but some claims are exaggerated. The benefits are also real, but the industry has historically oversold permanent employment and sometimes the necessity of tax subsidies. Domestic political and advocacy money is now pouring into both sides and undoubtedly affects what casual observers hear, but we cannot credibly assign the large opinion shift to advertising. Foreign influence attempts are real and worth policing, but the evidence so far says they are piggybacking on an already-existing American argument rather than creating it.
And perhaps the most interesting part is that the market seems to be responding: the industry's 2026 offers—**“we pay for the grid, we guarantee the load, we limit/replenish water, and we put community benefits in writing”**—are materially different from the much vaguer economic-development pitches communities were hearing two or three years ago. ([The White House][21])
If you want, James, I can take this one step further and build a “bullshit detector” for proposed data centers—say 8–10 questions that let you look at any particular project and tell whether it is likely a net local asset or whether the developer is quietly shifting the costs to everybody else.
[1]: https://datacenters.lbl.gov/publications/united-states-data-center-energy-2025?utm_source=chatgpt.com "United States Data Center Energy Usage Report: 2025 Update | Center of Expertise for Data Center Energy"
[2]: https://ceepr.mit.edu/workingpaper/who-pays-for-growth-evidence-from-datacenters-and-the-grid/?utm_source=chatgpt.com "Who Pays for Growth? Evidence from Data Centers and the Grid - CEEPR"
[3]: https://jlarc.virginia.gov/landing-2024-data-centers-in-virginia.asp?utm_source=chatgpt.com "JLARC | Data Centers in Virginia"
[4]: https://www.audits.ga.gov/ReportSearch/download/33298?utm_source=chatgpt.com "NET CHANGE IN ECONOMIC ACTIVITY"
[5]: https://www.pewresearch.org/short-reads/2026/09/22/americans-views-of-data-centers-have-turned-more-negative/ "More Americans now say data centers are negative for the environment, energy costs and more | Pew Research Center"
[6]: https://www.nature.com/articles/s41467-026-76500-9?utm_source=chatgpt.com "How sovereign control, decarbonization and energy costs shape public support for data centers | Nature Communications"
[7]: https://www.nwpb.org/npr-top-stories/2026-09-05/voters-are-fed-up-with-data-centers-both-parties-are-trying-to-cash-in-for-midterms?utm_source=chatgpt.com "Voters are fed up with data centers. Both parties are trying to cash in for midterms"
[8]: https://www.axios.com/2026/08/31/ai-advocacy-group-data-center-battleground-states?utm_source=chatgpt.com "Exclusive: AI advocacy group launches data center push in battleground states"
[9]: https://www.foodandwaterwatch.org/2026/01/22/stop-data-centers-now-campaign-launch/?utm_source=chatgpt.com "Breaking News: Our New Campaign to Stop Data Centers!"
[10]: https://pmc.ncbi.nlm.nih.gov/articles/PMC7467695/?utm_source=chatgpt.com "The small effects of political advertising are small regardless of context, message, sender, or receiver: Evidence from 59 real-time randomized experiments - PMC"
[11]: https://openai.com/index/disrupting-malicious-uses-of-ai-data-center-bandwagon/?utm_source=chatgpt.com "“Data Center Bandwagon” Campaign: US-targeted influence activity | OpenAI"
[12]: https://www.axios.com/2026/08/28/china-ai-data-center-backlash-bots?utm_source=chatgpt.com "America's data center revolt is getting a boost from China-linked bots"
[13]: https://politifact.com/article/2026/sep/23/china-influence-data-center-opposition/?utm_source=chatgpt.com "Is US data center opposition a ‘China PR thing,’ as Trump says? Experts say that’s exaggerated – PolitiFact"
[14]: https://www.nature.com/articles/s41467-022-35576-9?utm_source=chatgpt.com "Exposure to the Russian Internet Research Agency foreign influence campaign on Twitter in the 2016 US election and its relationship to attitudes and voting behavior | Nature Communications"
[15]: https://openai.com/index/public-policy-agenda/?utm_source=chatgpt.com "OpenAI public policy agenda | OpenAI"
[16]: https://www.anthropic.com/news/covering-electricity-price-increases?utm_source=chatgpt.com "Covering electricity price increases \ Anthropic"
[17]: https://blogs.microsoft.com/on-the-issues/2026/01/13/community-first-ai-infrastructure/ "Building Community-First AI Infrastructure - Microsoft On the Issues"
[18]: https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/affordability-pledge-responsible-energy-growth/?utm_source=chatgpt.com "Google’s approach for responsible energy growth"
[19]: https://about.fb.com/news/2026/03/meta-data-centers-support-energy-jobs-environment-local-communities/?utm_source=chatgpt.com "How Meta’s Data Centers Support American Energy, Jobs, the Environment, and Local Communities"
[20]: https://www.whitehouse.gov/fact-sheets/2026/03/fact-sheet-president-donald-j-trump-advances-energy-affordability-with-the-ratepayer-protection-pledge/?utm_source=chatgpt.com "Fact Sheet: President Donald J. Trump Advances Energy Affordability with the Ratepayer Protection Pledge – The White House"
[21]: https://www.whitehouse.gov/ratepayer-protection-pledge/?utm_source=chatgpt.com "Ratepayer Protection Pledge – The White House"