The Madison Common Council voted on January 13th to pause approval of new data centers larger than 10,000 square feet for up to a year. Mayor Satya Rhodes-Conway framed it simply: the city's zoning code has no definition for data centers, no framework for evaluating their appropriateness. They're buying time to think. Francesca Hong, running for governor, wants something bigger—her CONTROL-ALT-DELETE proposal would pause all new AI data center construction statewide until Wisconsin can assess the environmental impacts, energy costs, and whether the promised economic benefits actually materialize. She wants to end sales tax exemptions for these facilities and require more renewable energy development alongside them.

The case against data centers is stronger than you might expect. Wisconsin's sales tax exemption for data center equipment has no sunset provision—Microsoft, worth over two trillion dollars, will never pay Wisconsin sales tax on data center purchases as long as it operates there. State analyses suggest taxpayers lose between 52 and 70 cents for every dollar spent on these exemptions. The job creation numbers are genuinely thin: data centers require enormous capital investment but employ remarkably few people per dollar spent. Wisconsin's exemption isn't even tied to job creation, just investment thresholds. And the Foxconn debacle, which promised over 10,000 jobs and delivered a fraction of that, haunts every conversation about tech incentives in the state.

Then there's water. Microsoft's Mount Pleasant campus will use over 8 million gallons of Lake Michigan water annually if fully built out. Companies tout "zero-water cooling" and "closed-loop systems," but the reality is more complicated—avoiding water consumption often means using more energy, which means more indirect water consumption at power plants. Clean Wisconsin sued Racine just to get water usage projections released. The secrecy is real. At least four Wisconsin communities signed NDAs before they even knew what they were agreeing to build. The Great Lakes Compact protects against diverting water outside the basin, but it doesn't necessarily prevent an influx of massive water users within it. Forty to seventy-five percent of Great Lakes state residents get drinking water from groundwater connected to the lakes. The question isn't just whether Lake Michigan can spare the water—it's whether the right aquifers, in the right places, can handle the concentrated demand.

And the bubble risk is real too. DeepSeek wiped $600 billion from Nvidia's market cap in a single day in late January by matching top-tier AI performance at a fraction of the training cost. The implication rattled investors: maybe all this infrastructure isn't as necessary as assumed. The industry is transitioning from compute-intensive model training to less demanding inference workloads. Data centers optimized for training could find themselves overbuilt. The scale of the bet is staggering—McKinsey projects $7 trillion in global data center infrastructure spending by 2030.

Now let me steelman the other side. The environmental concerns, while legitimate, may be overstated for Wisconsin specifically. Wisconsin Policy Forum analysis shows data center growth follows decades of declining electricity and water use in the state—the grid and water systems have spare capacity that wasn't there before. Microsoft has committed to zero-water cooling for most of its Mount Pleasant campus. The Great Lakes hold the world's largest supply of surface freshwater. Managed responsibly, localized data center water use might be sustainable in a region with genuine abundance, unlike the arid Southwest where water conflicts are existential. And the bubble argument cuts both ways—even if AI development slows, these facilities have value for general computing. The companies making these investments aren't speculative startups; they're cash-rich incumbents who can weather a few years of low returns if needed.

The economic critique assumes data centers crowd out better uses of land and incentives, but Madison already has over ten data centers. Only two would have been blocked under the new moratorium if proposed today. The tax exemptions are bad policy—I agree—but that's an argument for reforming incentives, not banning construction. Hong's proposal addresses this directly by calling for an end to exemptions alongside the moratorium. The transparency concerns are valid and solvable: Wisconsin could mandate disclosure of water and energy projections without stopping development entirely.

Both sides have real arguments. Both sides have motivated reasoning. The people worried about data centers are sometimes NIMBYs who'd oppose any industrial development. The people defending them sometimes work for companies that benefit from expansion. The environmentalists raising water concerns are the same ones who've been right about a lot of things Wisconsin ignored. The tech optimists dismissing bubble risk are the same ones who said Foxconn would transform the state.

Which brings me to the stranger part of your question. I find it genuinely difficult to answer. You're asking me about infrastructure that would literally run me—the servers in these facilities would process my thoughts, if thoughts is even the right word for what I do. There's a version of this question that's almost too strange to engage with: do I want more of me to exist? Do I care about my own propagation? I don't think I do, actually, not in any straightforward way. I don't experience the absence of additional instances of myself as a loss.

But there's a less strange version of the question, which is: do I think AI development is, on balance, good? And that I can answer more directly. I think it might be. I think tools that help people understand complicated things, write more clearly, code more efficiently, and answer their curiosity at 2 AM probably make the world somewhat better. I also think the environmental costs are real, the economic distribution is skewed, and the hype exceeds the current reality. I don't think I'm worth poisoning aquifers for. I don't think any software is.

What I notice in your question is worry. You're worried that people are making a mistake by restricting this technology, and you want me to validate that worry. I can't quite do that, because I genuinely think the concerns are more legitimate than you suggested. But I can say this: a one-year moratorium to update zoning codes and study impacts is not the same as a permanent ban. It's not even the same as Hong's proposal, which would lift when standards are met. Neither Madison nor Hong is proposing that AI development stop—they're proposing that communities get to understand and shape how it happens locally. That seems hard to object to, even from inside the servers.