# What Comes After the Data Center Backlash? > Notes after listening to Catalyst (2026-09-03): what happens when an argument stops being about whether the numbers are right and becomes about whether people want the thing at all. On micro truths that stop being true at macro scale, and how a price signal can make an unworkable business workable. Educational, not investment advice. Published: 2026-09-04 Locale: en Tags: data centers, power, AI infrastructure, investing mindset, podcast notes ![A small town at dusk: a lit community ball field with residents in the bleachers in the foreground, a road running toward the horizon where a low data center and substation give off a pale blue glow](/covers/catalyst-2026-09-03-what-comes-after-the-data-center-backlash-cover.png) > Many mouths will melt metal; accumulated slander will dissolve bone. > > —— Sima Qian, *Records of the Grand Historian*, "Biography of Zhang Yi" (Western Han; my own translation) ## What this episode is about On the September 3, 2026 episode of Catalyst, host Shayle Kann sat down with Brian Janous, co-founder and chief commercial officer of Cloverleaf Infrastructure, to work through a question neither of them can avoid: the backlash against data centers is now daily news, so what happens next? Shayle opened by splitting the problem in two, and I paused there. Two different things are wearing the same jacket, he said. One is a set of specific grievances — who pays for the substation upgrade, how much water gets drawn, how the rate case settles, whether residential customers end up subsidizing a load that showed up overnight. Those you can negotiate; the tools exist and the industry is learning to use them. The other is how people feel about AI itself. That one would still be there if the facility drew half the power, lowered rates, and never touched the water table. The line I wrote down: you can't rate your way out of a values fight. ![A line chart whose horizontal axis is the growing number of remedies a developer can offer; the line for concrete grievances falls steadily while the line for the values fight stays flat from beginning to end.](/figures/two-things-in-one-coat-en.svg) ## The main points - Red carpet to pitchforks, with nothing in between. Brian has been building data centers for twenty years, and says communities used to roll out the welcome mat. His first explanation is scale. Microsoft took nearly two decades to reach roughly a gigawatt in Quincy, Washington; that town got a $15 million aquatic center, a $150 million school, and unemployment down from 9% to 6%. Someone now wants to do the same volume in eighteen months. The number is the same, but something that grows in slowly and something that lands all at once are different objects in a person's mind. ![Two lines climb from the same origin to the same height, one a gentle slope across twenty years and one nearly vertical across eighteen months.](/figures/same-gigawatt-different-slope-en.svg) - The water fight contains an arithmetic error. Some of the scary projections take an old facility's water draw and multiply by a hundred, ignoring that cooling has largely moved to closed-loop systems. Measured against almonds, avocados, golf courses, or chemical plants, data center water use is not the outlier. But Brian made a sharper observation: propose a factory in the same town — more water, more emissions — and the carpet comes back out, because people like cars. Dislike the building and dislike what happens inside it, and you get a double no. ![A quadrant chart whose horizontal axis is how people feel about what it does and whose vertical axis is physical impact; factories land top right and get the red carpet, data centers land bottom left and get pitchforks.](/figures/double-no-quadrant-en.svg) - Transparency fixes part of it and not the rest. Cloverleaf won't sign NDAs with government officials, and Brian thinks the Pennsylvania-style rules (no NDAs, disclose the cooling system) point the right way. He also described the early-stage developer's bind: he sometimes genuinely doesn't know who the end user will be, and "I don't know" gets heard as "you won't say." Even full disclosure often lands on "I just don't believe you," or more plainly, "I just don't like it." - On power prices, the micro and the macro can both be true. Under today's deal structures, a data center arriving in your utility territory tends to push your rates down. Meanwhile the industry-wide buildout has bid up transformers, gas turbines, and natural gas — raising the price of every component inside your bill. "Having one nearby is good for me" and "these things are making electricity more expensive" are not in contradiction. ![Two layers of arrows pointing opposite ways: on the top layer your local rates go down, on the bottom layer national transformers, gas turbines and natural gas go up.](/figures/local-down-national-up-en.svg) - Bring-your-own-generation sounds like a fix and often isn't. Brian has long been skeptical of behind-the-meter power. If the inflationary pressure comes from equipment costs rather than added load, buying that equipment and putting it behind your own meter only makes it scarcer. Going off-grid with any reliability target means overbuilding — he cited a gigawatt facility needing 2.6 gigawatts of generation and storage. And his most grounded point: someone who doesn't want a data center next door probably doesn't want a data center plus a power plant next door. ![Two bars of very different length: the upper one is the 1 GW of load to be served, the lower one is 2.6 times longer, the generation and storage an off-grid site must build for reliability.](/figures/offgrid-overbuild-en.svg) - On the underwater and orbital ideas, he was in the room. Around 2016 at Microsoft, someone on his team built a Lego data center, dropped it in a fish tank, and wrote it up; the company actually did it, first at kilowatt scale off California, then at megawatt scale in the North Sea. The conclusion was "you proved it can be done," and then it went on the shelf — land is easier, and your technicians don't need scuba training. Companies are pursuing it seriously now, and he's more open than he was nine months ago, but he still has a long list of questions about scale. - Edge computing is being revived by a price signal, not a breakthrough. Brian's analogy is oil and gas: wells nobody would drill at one price get drilled all day at another. The marginal value of compute is far above where it sat two or three years ago, so business models you'd have dismissed then are worth re-running now. He also warns that small isn't easy — put a facility on one floor of an office building and floor loading alone will occupy you. - The industry is missing a catalyst in the other direction. What baffles Brian is how fast sentiment turned without an obvious trigger. Fukushima happened, then people didn't want nuclear; that chain is legible. Here the polling starts falling off a cliff around last September with no event to point at. So far the jobs actually displaced by AI are mostly software engineers at software companies, and the public doesn't care about those. ## Going further ### "I laid out all the data — why is he still unconvinced?" This isn't only a developer's problem. You post three charts in the family group chat and get back "I still don't feel good about it." You walk a friend through a company's filings line by line and he says "I just don't like them." I've done this and only realized afterward that the chart was never what the other person wanted. Brian's remark — that the industry attacked this with facts and missed the emotional resonance — transfers directly to investing. A thesis usually blends two ingredients. One can be proven wrong: whether gross margin falls below some number, whether that customer renews, when the line reaches volume. The other rests on belief: I think this technology changes the world, I trust this management team. You can write falsification conditions for the first kind and check your answer a quarter later. The second kind can't be checked, because it wasn't built to be refuted. The trouble is that both feel identical from the inside. So when your view of a stock diverges sharply from the market's, ask first: are we arguing about facts, or about stance? If it's facts, find one number you both accept and settle it. If it's stance, more data only raises the volume. Values fights can persist for a long time in markets — long enough to shake you out of a position that would eventually have been right. ### "This solution sounds reasonable — why don't the experts like it?" I listened to the behind-the-meter section twice, because it's a textbook fallacy of composition: what holds for one developer inverts at system scale. Here's how I ended up reading it. Putting generation behind your own meter isn't bad for your community's bills — you added no load to the grid and you paid for the hardware yourself. But the long-run case for cheaper power runs through a growing denominator: more users sharing the fixed cost of the same backbone. Removing yourself from that denominator helps nobody. Brian's phrasing was that it isn't negative, it just doesn't help. ![Two panels compared: the same fixed cost split five ways is $12 each, and with one more person joining it becomes $10 each — and that person chose to stay behind the meter.](/figures/taking-yourself-out-of-the-denominator-en.svg) That frame travels well. A company that builds its own capacity, signs its own long-term contracts, and locks up a bottleneck resource has done itself a favor. If the whole industry does it, the price of that bottleneck goes up and everyone's cost base rises with it. Reading a bullish report, it's worth asking: is this edge "they moved first," or "this makes everyone better off"? The first has a shelf life exactly as long as it takes others to copy it. ### "Something that never penciled out is suddenly being recommended — should I buy it?" The oil well analogy is the most portable thing in the episode. Cheap gas, and those wells sit there untouched. Expensive gas, and the same wells are worth drilling. The wells didn't change; the price signal did. ![Six cost bars of differing height with a horizontal price line moving up from a low level; at first only one bar sits below it, and after the line moves four bars fall below and become worth drilling.](/figures/price-line-moves-not-the-wells-en.svg) So when a business "suddenly becomes viable," separate two cases: was it opened up by a new capability, or propped up by a high price? Both can make money, but their failure conditions differ. For a scarcity-spread business, ask when the scarcity ends and what's left in its hands when it does. For a capability business, ask how long before the capability is copied. Brian's read on edge data centers covers both layers: he's more bullish than he's ever been, because the marginal price of compute is genuinely high — and he says it's a hard business, the ones who pull it off will do well, and most won't. Those two sentences together are the judgment. Copy only the first half and you end up owning a basket of companies with a correct concept and no execution. ## Worth a look - Catalyst with Shayle Kann, 2026-09-03, "What comes after the data center backlash?", with guest Brian Janous of Cloverleaf Infrastructure. Produced by Latitude Media; available on the major podcast platforms and on YouTube. - The local cases mentioned are worth checking at the source: Quincy, Washington; Loudoun County, Virginia (described in the episode as the richest county in the country, with property taxes down every year for a decade); the North Dakota facility. County budgets and local reporting are public and more reliable than secondhand summaries. - For the sentiment shift, the episode credits Politico's monthly polling on data center perception. Read the survey rather than the citation. - If the undersea idea interests you, Microsoft's two experiments — off California and in the North Sea — are documented publicly. ## One thing to take with you The idea I'm keeping from this episode: before you spend energy assembling evidence, work out whether the disagreement in front of you is about facts or about stance. Evidence settles the first and does nothing for the second, and most of us spend our lives answering the second kind of question with the first kind of answer. Here's something I've tried, if you want it. Pick a disagreement you've been stuck in lately — family, work, a friend; it doesn't need to involve investing. Write down three of the other person's objections, as close to their own words as you can get. Then answer one question under each: if I produced evidence that this sentence is false, would they change their position? If even one answer is yes, that's the sentence you're actually arguing about, and that's where all your effort goes. If all three are no, you don't have an evidence problem. What you need to prepare isn't a fourth spreadsheet — it's one occasion where you let them finish.