What comes after the data center backlash?

Kann splits the opposition into two issues wearing the same jacket. One is a list of concrete, negotiable complaints, several well founded, answerable by siting differently, paying for the upgrade and structuring the tariff so the big load carries its own cost. The other is not about that data center at all and would survive a facility that drew half the power and never touched the water table. Janous traces the first partly to compression: what Quincy, Washington absorbed over two decades now arrives as an eighteen-month proposal, which invites people to multiply the water and power numbers, and some of that arithmetic extrapolates from older cooling designs rather than today’s closed loops. But correcting it does not land, because of the double whammy. A car factory has worse externalities and still gets a red carpet, since people like cars. Here the objection is to the building and to what happens inside it, so fixing the building fixes half the problem.

Why sentiment turned so fast is a puzzle he cannot solve. He reaches for an inverted Fukushima: nuclear’s collapse in public opinion had a meltdown behind it, and this had none, since the water and power concerns were always true. By his reading of Politico’s monthly polling, perception was steady across both parties and then fell off a cliff starting around the previous September. His hope is a catalyst in the other direction, something people actually want AI to do, but the politics run against it: through the midterms and a presidential race, candidates compete to brand each other the pro-data-center candidate, and because it works, others will copy it. Favorable local stories exist, but nobody with influence will carry them before an election, and neither speaker claims to know whether a positive catalyst would work even if one arrived.

On prices the two of them hold a distinction rather than collapsing it. Kann relays a colleague’s argument that a data center arriving in your territory today tends to lower your rates, since that is what the utility deal is designed to do, while the national buildout simultaneously inflates the price of transformers, turbines and gas, and therefore everyone’s electricity. Janous accepts both halves and reframes the second: producing more of something usually makes it cheaper, and prices did not rise across a century of grid expansion, but this expansion follows a long stretch of no load growth and has to rebuild supply chains that were allowed to wither. What he does not concede is the causal claim. Bills are higher, and whether that is because of data centers is “maybe partially,” with pressures that exist regardless doing the rest. His answer is not to stop growing but to take the short-term pain seriously and aim community benefits at the specific thing a specific community is hurting about, rather than writing generic checks that read as a bribe.

The exits mostly fail on one point. Bring-your-own-capacity policies in Texas and Pennsylvania are worded vaguely enough that nobody knows whether they mean on-site generation or an obligation to fund new supply. Janous’s own version is the latter, and Cloverleaf is doing it on a project with a small municipal utility in Oklahoma, sourcing the wind, solar and storage and handing the package to the utility to serve the data center. If the intent is behind-the-meter generation instead, he thinks it makes things worse, because the inflation comes from equipment cost and buying the same equipment for your own fence line adds to the same pressure, and because islanding forces you to overbuild. He starts to argue that pulling a large load off the grid also removes the denominator that spreads costs, then walks himself back: not negative, just not positive. Kann pushes hardest here, arguing that if grid connection hits a ceiling, off-grid is the small plausible step because the country has empty places. Janous does not buy it, because off-grid does not touch community acceptance; you still have to permit land, and sea and space will stay marginal. Edge computing is the one he has warmed to, on economics rather than politics: in a short market the marginal megawatt carries real value, so models that failed on break-even in an unconstrained world start to pencil, though he still expects most compute in facilities of 100 megawatts or more.