What If the Building Paid for Its Own Resurrection? V2

I think I explained this idea backward the first time. The point was never to figure out how to squeeze another data center into an abandoned building. The whole reason I started thinking about the distributed-compute project was because I have been questioning why AI compute has to keep becoming synonymous with enormous centralized data centers in the first place.

What interested me was almost the opposite: distributed compute. Instead of one gigantic campus pulling an enormous new load from the grid, what happens if serious compute is broken into small, productive nodes and distributed through buildings that already exist? A compute center doesn't necessarily have to look like a data center. It could be a back room on the third floor of an otherwise completely normal building.

One of the models I have been working through is roughly a $250,000 compute stack capable, under the utilization and revenue assumptions I was modeling, of producing around $35,000 a month. Those numbers are a model, not a guarantee, and real revenue would depend on hardware, demand, utilization, pricing, operating costs and a pile of variables that can change quickly. But the possibility is what grabbed me: a relatively small amount of physical space could potentially create enough economic activity to affect an entire building.

Then I started working backward from the energy problem. I don't want the answer to AI demand to simply be, “Fine, pull more electricity from the grid.” The version I am interested in pairs the compute with solar generation and battery storage, with the goal of operating without creating an additional net burden on the grid. In the scenario I was working through, the solar footprint was under roughly 5,000 square feet, which is a radically different physical proposition from the giant campuses people picture when they hear the words AI infrastructure.

That is where the old-building idea came in. Imagine an underused three-story building where the compute itself occupies only a tiny piece of the property. The machines aren't the purpose of the whole building; they're the engine hidden inside it. If that engine can generate meaningful recurring revenue, maybe it helps finance rehabilitation of the roof, electrical systems, apartments, offices, storefronts, community space or whatever the rest of that particular building ought to become.

That is a completely different idea from turning old buildings into data centers. I don't want a beautiful old structure gutted so racks can consume every square foot behind locked doors. I want to know whether something closer to sixteen square feet of productive compute infrastructure in a back room can help make the thousands of square feet around it useful to human beings again.

The distinction matters because I have serious concerns about the direction of giant centralized data-center development. Communities are being asked to accommodate extraordinary concentrations of electricity demand, land use, infrastructure and other impacts so computation can happen somewhere far away from the people ultimately creating the demand. I keep wondering whether centralization is being treated as inevitable when it may simply be the model we scaled first.

Distributed compute asks a different question. What if thousands of smaller nodes could exist inside infrastructure we already have? What if they could generate more of their own electricity locally, store it locally and avoid demanding another massive block of capacity from the grid? What if the economics created by those machines stayed attached to actual buildings and actual neighborhoods instead of disappearing into another fenced-off campus?

And I don't want the benefit to stop at rehabilitating the property either. If the model works, part of the value should deliberately leave the building. I have been thinking about ways the revenue could feed community programs, including the Blok work, schools, surrounding neighborhoods or other locally useful projects instead of treating “community benefit” as a paragraph somebody adds to a proposal after the economics are already finished.

That's the part that makes this much more interesting to me than owning GPUs. The compute becomes a little economic engine. The building gets rehabilitated because the engine exists, the engine is designed around locally generated energy and storage rather than simply demanding more from the grid, and some of what the engine produces gets pushed outward into the community around it.

Then repeat it.

One building doesn't matter much in the scale of global computation. But a distributed network of small compute centers begins to become something else entirely. Instead of asking where America should build the next enormous data center, maybe sometimes the better question is how much useful compute can quietly live inside the physical infrastructure we already built.

There are obviously things that have to survive reality before I know how far this can go. Solar production changes by location and season, batteries cost money and degrade, hardware ages, compute pricing moves, utilization isn't guaranteed, buildings have different electrical and structural constraints, and the economics of a $250,000 stack can change quickly. I don't want to pretend a spreadsheet makes any of those problems disappear.

But I think the experiment is worth chasing because the underlying question is bigger than any one set of numbers. We are going to need an enormous amount of computation. That doesn't automatically mean we need to concentrate all of it into enormous facilities that require communities to build infrastructure around them.

What if the future of compute is smaller than the data center?

Put a serious machine in a forgotten room. Power it locally as much as the site can realistically support. Let it generate value, use that value to rehabilitate the building around it, return some of the benefit to the surrounding community and connect thousands of these little engines into something much larger.

Then the building really can help pay for its own resurrection.

Not because we turned it into a data center.

Because we stopped assuming a data center had to be a place at all.