2035: The town's computer is a rack
Escherbridge
Speculation. Not a roadmap, not a delivery date.
This essay is tied to OPQPWR, the physical layer in the PrimusNeo stack. The first half is current and sourced. The second half is speculation about 2035. It is not a product roadmap, a deployment claim, or a description of hardware I have not opened.
What is true now
The International Energy Agency's Energy and AI assessment puts data-centre electricity use at about 415 terawatt-hours in 2024, around 1.5 percent of global consumption, and expects that figure to more than double to around 945 TWh by 2030. In the IEA Base Case, consumption reaches almost 1,200 TWh by 2035. Around 20 percent of planned data centre projects, the agency reported, already face possible delays because of long waits for grid connections; if grid and supply risks are not addressed, 20 percent of expected new capacity from 2025 to 2030 could be delayed. New transmission in advanced economies currently takes four to eight years to build.
Those are numbers about large halls in particular places, waiting on wires. They are also a reason to look at smaller, movable power.
Idaho National Laboratory describes a portable system it calls Microgrid in a Box: a collection of mobile trailers and containers that is "a fully functional microgrid." The published description includes 320 kilowatt-hours of battery storage. The system can tie into a modern grid and coordinate electricity for a small village, a military base, or, in a disaster, a hospital or other critical building. INL says it can manage about 250 kilowatts, can work with both U.S. and U.K. electrical standards, and can coordinate generation from several sources, including solar, wind, nuclear, or a diesel generator. An early version, the laboratory wrote, has already been deployed for military use in Kuwait. Researchers there argue that because the system can be disassembled and moved, it "sidesteps the need for construction." I am repeating INL's own account, not certifying the hardware.
This is a crowded field, not a unique invention. National laboratories, vendors, and utilities have been putting generation, storage, and controls into containers for years. What matters for this essay is the pattern, which is public and repeatable: a unit of power that can arrive without a new substation, serve a defined load, and leave.
OPQPWR itself is thinner in the public record. PrimusNeo materials describe it as a partnership that is starting, and as field-deployable racks, and they state that no on-premises deployment is running yet. I found no public GitHub repository under that name, and I did not reach an official OPQPWR website. I am therefore not repeating secondary product claims, and I am not treating a partnership announcement as a specification. What can be said is that the physical layer is intended, not demonstrated. Any future tense has to start there.
What follows is speculation
Suppose, in 2035, the IEA's direction is roughly right: more computation, more electricity, and grids that do not appear on demand. Suppose also that portable microgrids have continued down the path INL is already on — not as a curiosity for a test range, but as a normal way to put a few hundred kilowatts and a buffer of storage in a place that cannot wait for a transmission line.
Then a small community's interesting computer may not be an account in someone else's region. It may be a rack. Power on one side, computation on the other, sitting on land the community already has, sized for the work of that place: a model of a site, a records system that should survive a cut cable, the quiet jobs that do not need a training cluster. Self-sufficiency is a slogan. Locality is a constraint. If the large halls are delayed, the work that cannot be delayed will have to run closer to a generator someone can see.
I am not claiming that OPQPWR will be that rack, or that it can now do any of this. A partnership starting, with no public repository and no running on-premises deployment, is not a product you can write a 2035 brochure from. The speculation is about the job, not the vendor. The job is: a unit a town, a co-op, or a small municipality could own, move, and turn off; power and compute specified together, because a computer without a feeder is a paperweight, and a feeder without a workload is a very expensive shed.
There are obvious failure modes, and they should stay in the picture. A rack can be poorly maintained. A container can become another object nobody on site knows how to run. Fuel, if it is in the mix, has its own logistics; INL's own description includes diesel as one of the sources the microgrid can coordinate, alongside cleaner ones. Local compute can be just as extractive as remote compute if the data still leaves and the keys do not stay. The 2035 version that is worth wanting is the one where ownership is literal: the hardware is on a property, the accounts are the community's, and the system can be inspected without calling a vendor.
If that picture arrives, OPQPWR is the layer it would attach to — the physical side of a stack that otherwise lives in software. Until there is a public artifact to read, and a deployment to look at, that attachment is a hope about fit, not a report about progress. The facts we have are the IEA's queues, INL's portable microgrid, and a partnership that has not yet put a rack on a site. The rest is a guess, written down and not dated.