Artificial Limits

An open, collaborative argument

The Data Center Fix

The public argument about data centers has hardened into two positions, and neither one can be acted on.

One says the build-out is inevitable and that objecting to it is a luxury the region cannot afford. The other says the costs are unacceptable and the answer is no. Both are arguments about whether. Almost nobody is arguing about under what conditions — which is the only question a permit, a utility rule, or a water appropriation is actually able to answer.

That gap has consequences. A decision process needs two capacities and most have only one. It has to be able to stop a project that cannot protect the people nearest to it. It also has to be able to say yes in a way that binds — with conditions that are measured, reviewable, and still enforceable long after the announcement. A process that can only delay protects no one. It just makes the outcome depend on who can outlast the procedure.

So this project works on the narrower question, the one the book puts at the center of its chapters on infrastructure:

What must a project prove before other people can be required to carry its burden?

This project grows out of Artificial Limits, where these chapters began. It is not anti-development and it is not a booster campaign. It is an argument about evidence and permission, and it is open to anyone willing to make it better.

The argument

Four claims we will defend and revise

Each one is falsifiable. If you think a claim is wrong, the fastest way to change it is to show us why — that is what this site is for.

  1. 01

    Every number is a boundary

    Before a benefit or burden can be counted, someone chooses the unit, the period, the geography, and the population included. Those choices are not housekeeping — they decide whose reality becomes visible. If the beneficiary is counted globally and the burden bearer locally, the comparison has been tilted before any arithmetic begins.

  2. 02

    Efficiency is a unit result, not a system result

    Energy per task has fallen dramatically. Total demand rose anyway. Lower energy per task is not lower total energy once cheaper capability changes how often it is used and how ambitious the surrounding system becomes. A more efficient prompt answers one question. It does not answer the other.

  3. 03

    Disclosure without authority is an archive

    Transparency becomes leverage only when it is attached to authority. The chain has to close: set the condition before approval, measure against it, make performance reviewable, give affected people a real route into the proceeding, and name what happens when the condition fails. Without that last step, you have documented the harm rather than prevented it.

  4. 04

    Mitigation is work, not absolution

    Credible mitigation keeps five things visible: the actor, the mechanism, the measured outcome, the boundary, and the residual burden. Improvement can be real without becoming a receipt. What mitigation cannot buy is the right to make the original bearer invisible.

The proposal

Five decisions, not one switch

At the end, the public deserves more than approve or deny. Most of the damage done by permitting fights comes from forcing a complex judgment through a binary gate. These are the five outcomes an honest process should be able to reach — and each one publishes what evidence would change it.

  1. Proceed

    Protection and cumulative capacity are demonstrated, project-caused costs are internalized, and any heightened trigger has a proportionate justification. The record states the conditions and what happens if they fail.

  2. Modify

    Another design serves the same purpose with less total burden. The comparison must catch transfers — dry cooling can raise peak-grid stress; reclaimed water still needs treatment, energy, and residuals management.

  3. Stage or pause

    Material uncertainty, missing infrastructure, or cumulative planning blocks full approval, but the gap can be closed. Interim limits, a deadline, review, and exit criteria — so a pause cannot quietly become a ban.

  4. Relocate

    Another site can deliver the capability with materially lower total burden and acceptable service consequences. Compare both places on common boundaries, and do not export the harm to a weaker community.

  5. Refuse

    Protection is infeasible, enforcement cannot act in time, costs remain shifted, plausible harm is severe or irreversible, or the justification cannot carry the requested burden. Publish the reason — and the evidence that could reopen the decision.

These five outcomes are only as real as their afterlife. A staged approval needs a later decision that is not presumed. A refusal needs findings that distinguish protection from political convenience. Permission should be a living public judgment, not a ceremonial day at the beginning of construction.

The library

What we're reading

Research, reporting, and argument worth your time — including work that disagrees with us. Contributed by readers, checked before listing, and labeled with what each source can and cannot establish.

Reading list — open for submissions

If you know work that belongs here — a paper, an investigation, a dataset, a dissent — send it below. Every entry is labelled with what it can and cannot establish.

Contribute

Add to the argument

Share a link, pitch an essay, correct something we got wrong, or offer your own experience. Everything is read by a person. Nothing is auto-published.

The one rule: engage the idea, not the person. Disagreement is welcome and actively wanted — a claim that cannot survive its strongest objection should not survive. Contempt isn't, and won't be published.
What are you contributing?

Select what you are contributing

Enter your name

So we can follow up. Never published.

Enter an email address, like name@example.com

What it is, and why it matters to this argument. A couple of sentences is plenty for a link; write as much as you need for anything else.

Tell us about your contribution

Confirm we may publish this before sending