ASHFALL INSTITUTE | SUBDUCTION ZONE
Building Bridges — Episode Two
WHY WE NEED THE BUILDINGS
P. A. Moore
Ashfall Institute | Subduction Zone
Written under the collaboration disclosed in Time Is a Wheel and We Are the Ground. Concept, argument and judgment: P. A. Moore. Research and composition: Claude Opus 5.
Let me start by conceding something, because I don’t think this conversation can go anywhere until somebody does.
A great deal of what runs in these buildings is junk.
Advertising optimization. Recommendation engines built to keep a teenager scrolling past the point where it does them any good. Image generators producing an ocean of material nobody asked for and nobody will remember. Fraud. Spam at a scale and polish that didn’t exist five years ago.
If you’ve been told that objecting to a data center means you don’t understand technology, I’d like to say plainly: you understand it fine. Some meaningful fraction of that electricity is being spent on things that make the world slightly worse, and anyone who tells you otherwise is selling something.
So I’m not going to argue that all compute is precious.
I’m going to argue that some of it is, and I’m going to be specific, because vagueness is how this conversation usually dies.
Start with cancer.
Every tumour carries mutations. Some of those mutations produce proteins that exist nowhere else in the body — not in healthy tissue, not in anyone else’s tumour. Just yours. They’re called neoantigens, and they are, in principle, the perfect target. Visible to the immune system as foreign. No risk of the body attacking itself, because nothing else in you looks like that.
The problem has always been finding them.
A tumour may carry thousands of mutations. A usable target has to be produced by the cell, then bound by your particular immune molecules — which differ from mine — and then actually provoke a response. Three filters, across a personal genome, and it has to be solved fast enough to matter to someone who is currently ill.
That is a search problem, and it is exactly what these machines are for.
There are AI-designed cancer vaccines in clinical trials right now. Not in principle. In hospitals, with patient numbers. One of them is a pan-cancer messenger-RNA vaccine designed by machine and entering Phase I this year. Another is a personalized vaccine for newly diagnosed brain tumours. There are systems that now do the target selection and the vaccine design in a single pipeline, where those used to be separate problems solved months apart.
I have a personal stake in this and I’m not going to pretend otherwise. But the stake isn’t why it’s true.
Now weather, which is closer to home for a lot of you than it is for me.
Everybody in Oklahoma knows the second season. Everyone in the Southeast has watched a storm sit over a county for six hours and take the bottom out of it.
Warning time is the whole game. In 1980 the average lead time on a tornado warning was under five minutes. It’s now closer to fifteen. That difference is measured in people.
The next increment doesn’t come from more radar. It comes from processing what the radar already sees — and from resolving models down to the scale where a storm actually organizes itself, which requires an enormous amount of computation done very fast, because a forecast that arrives late is not a forecast.
And this matters right now for a specific reason. Federal weather funding has been cut. NOAA’s research arm survived a proposed twenty-two percent reduction to non-defense discretionary spending, but its programs were reorganized and its footing is worse than it was. Meanwhile the tornado threat has been shifting eastward for two decades — out of the plains, into the Southeast, into places with more trees, more mobile homes, more people, and fewer sirens.
We are getting less public capacity to see storms coming at exactly the moment they are moving toward more of us.
That gap gets filled by computation or it doesn’t get filled.
There’s more, and I’ll be brief about it.
Materials — including the concretes I’ll spend a whole episode on, because the search space for a new cement chemistry is astronomically large and human intuition has been exhausted on it for a century.
Protein structure, which was a thirty-year bottleneck and is now largely solved, and which sits underneath every drug you’ll be offered for the rest of your life.
Grid management — which is not a small irony, given the rest of this series.
So here’s the honest position. We need some of these buildings. Not all of the things they’ll be used for, but some.
Which brings me to the part I actually want you to hear.
A factory has to put the workers next to the machines.
That’s the whole reason a factory town exists. The machine is in Ohio, so the work is in Ohio, so the wages are in Ohio, so the schools and the shops and the union hall are in Ohio. The benefit is stuck to the geography because the work is stuck to the geography. That’s not generosity. It’s physics.
Compute is not like that.
The building has to be somewhere. It needs land, power and water, and it needs to be near a fiber route. But the output is weightless. It can be consumed anywhere on Earth at the speed of light. There is no physical reason the benefit has to stay near the building.
And so it doesn’t.
The county gets the transformer and the traffic and the bill. The output goes to a customer four states away, or four countries away, and the county never sees it again.
Every serious grievance in this fight follows from that one fact, and nobody has tried to fix it.
Because here’s what nobody has said out loud: the portability runs both ways.
If the output can go anywhere, it can also stay.
Let me make that concrete, because otherwise it’s a slogan.
A data center is a furnace. Every watt that goes in comes out as heat, and at the moment essentially all of it is thrown at the sky through cooling towers, which is also where a great deal of the water goes. In Denmark and Sweden that heat is piped into district heating systems and warms homes through the winter. It is not exotic. It is plumbing, and it is done.
A data center needs a fiber trunk. That trunk passes through a county that in many cases has terrible broadband. The fiber is already in the ground.
A data center needs an enormous, reliable, hardened electrical connection. Building it upgrades a piece of grid that was going to fail anyway. Whether the upgrade serves the town too is a sentence in a contract.
And a data center is, physically, a very large amount of computation sitting in a rural county. Nothing prevents a fraction of it from being reserved — for the state university, the regional hospital system, the county’s own weather modeling, the school district. Nothing except that nobody asked.
I want to be careful here, because this is where a series like this usually goes soft.
None of that happens by goodwill. Every item on that list is a contract term, and contract terms are won by parties with leverage. Right now the leverage sits with a company that can go to the next state, and I told you last week why that is — it’s a prisoner’s dilemma between fifty legislatures and nobody can defect alone.
But notice what changes if the terms change.
The seventy-one percent isn’t opposition to computers. It’s a correct reading of a deal in which the county carries every cost and receives a building. Change what the building gives back and you are having a different conversation entirely — not a better-argued version of the same one.
One last thing, and then I’ll stop.
We have been told for two years that this technology is a threat to American workers. I want to point at something about the industry doing the telling.
A very large share of the American technological economy was built by people who arrived from somewhere else. Immigrants found companies at substantially higher rates than the native-born. They patent at higher rates. Go through the founding stories of the firms currently building these data centers and you will find the same thing over and over — someone who got here, or whose parents did.
That was true when the objection was that they’d take the jobs.
It is worth remembering what that objection has been worth historically, and how consistently it has been wrong about who ends up building the thing everyone later depends on.
Next time: who actually pays for these buildings, who decides, and why you’re not allowed to read the contract.
Sources: PAN-NeoVax (NCT07560943) and Tamavaq (NCT07076616), both recruiting 2026. Tornado warning lead-time statistics, NOAA. Gensini & Brooks on the eastward shift in tornado frequency. Danish and Swedish data center district heating: Fredericia and Stockholm Data Parks.
Building Bridges is a fourteen-part series. Previous: Episode One — The Bargain. Next: Episode Three — Who Pays, and Who Is Allowed to Read the Contract.
P. A. Moore is the pen name of Pamela King, philosopher and artist. Available through the Ashfall Institute.