AI executives are starting to ask: should we just mail a check to every nearby household?
Data centers face 70-30 bipartisan opposition — not because water usage is misreported, but because AI has no constituency. Solar has one. Auto plants have one. Data centers have only utilities and an already-unpopular AI industry.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The NDA is what manufactures the rumors
Developers ask city council members to sign NDAs — they can't disclose that the project is a data center, whether the customer is OpenAI or Anthropic, or how much power it will draw. But as one building trades union vice president put it, the developer finds the general contractor, the GC finds the subcontractors, the subcontractors find the workers, and every link in the chain signs an NDA — and the word still gets out. Bound by the NDA, council members lose the ability to get ahead of the social media narrative — a line local government officials repeated again and again. Microsoft has already stopped using NDAs because the backlash got too big, and other people running compute at AI labs are considering following suit. Even people who support building data centers regret signing the NDA.
— Jasmine SunWater is the sticky symbol; electricity is the real problem
Nearly every data center built today uses a closed-loop cooling system, and its water use is a small fraction of a golf course's — there is a golf course right next to the Janesville data center site. But water became the sticky symbol of resource consumption, and residents near the Great Lakes ask: if you weren't trying to drain the lake, why build here? Electricity consumption is the real issue: chips and clusters need low-latency interconnection, which requires new generators and power plants, most likely natural gas in the near term. A data center running normally produces little air and water pollution, but the power plant built for it is another matter.
— Jasmine SunEvery supporting argument runs into ‘I don't believe you’
The company says it will treat the water chemically and keep it out of the lake; residents say I don't believe you. The company says it will cover the cost of grid expansion and guarantee Michigan residents' electricity rates won't rise; residents say DTE has raised rates almost every year since 2022, so why not this year. The company says it will create 500 jobs; residents still say I don't believe you. The data center lands in an environment of extremely low trust in government and corporations, and the proponents' arguments barely land. Jasmine says that even if all the misinformation about water usage disappeared, people would not be any less angry.
— Jasmine SunAI has no constituency; solar does
Solar farms face almost the same playbook of opposition as data centers: Facebook groups organizing, packed town halls, talk of farmland and community vision, zoning lawsuits. But solar has a clear pro side — people who care about the environment and want renewable energy, who are willing to ‘take one for the team.’ Auto plants have one too: the 7,000 workers at the old GM plant in Janesville and their families are a constituency. Data centers, apart from utilities and an already-unpopular AI industry, have no persuasive constituency. Charles Franklin of the Marquette poll says an issue with strong pro and strong anti sides is usually 50-50, whereas data centers are 70-30 bipartisan opposition, and it doesn't matter whether you live nearby — because there is no strong pro argument at all.
— Jasmine SunSued, a small town can only settle
The Saline Township board voted down the rezoning for a data center site — a rare case in the US. The developer promptly sued the town of a few thousand people, on the grounds of exclusionary zoning: you can't ban industrial use across the whole town. The town chose to settle: take a few million for the fire department and the schools, and stop fighting the case. For locals, this is a serious assault on democracy with a lowercase d: a company with money goes to the politicians, doesn't tell anyone else, and the politicians can't tell residents either — and the deal they reach is fundamentally asymmetric because the two sides' resources are so unequal.
— Jasmine SunThey are willing to flip a 51-49 coin
The author sorts practitioners' self-justifications into three types: technological determinism (superintelligence will inevitably be built, so I want to be involved, to make money, to make it happen in the least bad way); ‘it might kill everyone, or it might bring superabundance, immortality, cures for all disease, and drive the cost of housing, energy and every consumer good to near zero’; and people who are simply fascinated by the technical problem and not very worried about the consequences. She sums up the second type with the SBF thought experiment: a 51% chance of doubling all human welfare and a 49% chance everyone dies — he would flip that coin.
— Karen HaoNo one agrees on what AGI means, so the finish line keeps moving
The author says one of the first things she did after reporting deeply on AI was figure out what AGI actually means, and found no agreement: from AI that can build itself, to AI that can do all human work, to AI that produces some economic value figure — everyone has their own milestone. That means the race has different finish lines, and moving ones. Model performance is extremely jagged: it can be excellent at math and terrible at cards, good at breaking into cybersecurity problems yet unable to make anything in the physical world. So she argues the technology is not as general as people say, and the finish line is harder to define — because with no way to adjudicate, every player will always feel they are behind on some dimension.
— Karen HaoWhether you believe in the race determines the whole AI policy stack
The author says whether you believe ‘we are in a race with China, and whoever reaches recursive self-improvement first is permanently and decisively ahead’ is the core dividing line in AI policy. Believe it, and what happens over the next one to three years is definitional; don't believe it, and the argument that ‘let's run faster toward a bad place’ is less persuasive, and policies demanding we slow down and give more voices and considerations a hearing are not crazy. From this she notices a strange convergence: safety people who genuinely believe in the race but think winning it might end humanity, and AI populists who don't want to hand this much power to tech billionaires, are arriving at similar positions from completely different mechanisms — Ron DeSantis and Max Tegmark holding a roundtable, Bernie Sanders and Eliezer Yudkowsky making a video.
— Karen HaoIt's not a marketing problem, it's a product problem
The author says that in 2026 AI executives and researchers began taking the public backlash more seriously, but the self-diagnosis she heard was ‘can we do better marketing,’ ‘why is Waymo so unpopular,’ ‘should we just mail a check to every nearby household.’ Her response: if you believe the things you keep saying, and the things you tell me privately, then you don't have a marketing problem, you have a problematic technology, a product problem — because people will not want that future. She also distinguishes among the companies: they have long marketed only to potential employees and potential investors, winning vibes on AI Twitter in San Francisco, as if they never realized everyone else could hear them.
— Karen HaoSilicon Valley's worldview is being shattered by data centers
Ezra Klein says the difference between the worldviews of DC and Silicon Valley is this: Silicon Valley people see ‘impossible problems proven solvable,’ DC people see ‘possible problems proven unsolvable.’ Now these two worldviews are colliding inside the AI industry — these people really did build AI, which is impossible becoming possible; but then they discover they can't even build a data center. He judges this will impose on the AI industry a realism it has always lacked: many problems can't be solved, not because you haven't thought of a way, but because you have to face the messy reality of society, politics, values and logistics.
— Ezra KleinIn their own words · checked verbatim
there's an old Irish saying that the only way to keep a secret between three people is to kill two of them
Jasmine Sun10:16
the pro arguments almost do not land because people just aren't interested in anything an outside tech company is going to tell them
Jasmine Sun15:28
if you are so worried about the inequality, why are you building the agents to do it?
Jasmine Sun34:06
We're building the thing we're telling you to be afraid of, and we need to build it as fast as possible, even though we sort of admit that it'd be better if the whole thing was slowed down.
Karen Hao39:24
if you believe the things you have been saying and, in fact, the things you have told me personally, you don't have a marketing problem. You have a problematic technology.
Karen Hao1:07:48
a lot of people in these companies overrate how much of the bottleneck in scientific and human progress is raw intelligence.
Karen Hao1:14:00
in Silicon Valley, people's worldview is formed by seeing impossible problems prove possible to solve. And in DC, people's worldview is formed by seeing possible problems prove impossible to solve.
Ezra Klein1:17:10
it's actually really hard to buy an election. It's actually really hard to buy out the whole White House at once.
Ezra Klein1:18:12
Figures
| Share of voters opposed to a data center near their home (last August) | about 40% | 2:04 |
| Share of voters opposed to a data center near their home (this May) | 70% | 2:04 |
| Number of local or state-level moratorium proposals nationwide | more than 100 | 3:05 |
| Property tax Microsoft will pay in Mount Pleasant in 2026 | $19.6 million | 14:26 |
| Odds in the SBF coin-flip thought experiment | 51% chance human welfare doubles / 49% chance everyone dies | 42:28 |
| Odds some people describe for the superintelligence bet | 90 to 10 or 80 to 20 | 44:34 |
| The year the author is in | 2026 | 1:05:42 |
| Years the old GM site in Janesville has been listed for sale | five years | 54:57 |
Glossary
- NDA
- An agreement developers require city council members to sign barring disclosure of project details — which instead leaves council members unable to respond to public opinion.
- jagged frontier
- Model capabilities are extremely uneven: excellent at math yet terrible at cards, which makes generality very hard to define.
- AGI
- A widely used goal no one agrees on defining, from AI that can build itself to AI that can do all human work — everyone has their own milestone.
- recursive self-improvement
- The idea that AI can improve itself and thereby accelerate iteration — the core of the argument that whoever gets there first leads permanently.
How to listen
Founders, investors and policy researchers watching AI infrastructure land, local politics and tech regulation.
The book recommendations at 01:20:22 can be skipped.