Phones (a few watts) already run small language models locally; a gaming PC runs serious open models offline, with zero water use and total privacy. Language apps, story archives and personal records can live here, on hardware the household already owns.
P4A proposal, react, revise, refuse
The proposals / sovereign compute
Possession needs hardware.
The data sovereignty principles in the record room end on a physical question: OCAP®'s fourth letter is possession, and you cannot possess what only exists in someone else's cloud. This room proposes a gradient (from the phone already in your hand to a single rack in a community building) on which a nation's data, models and digital memory can physically live, on Country, and keep working when the cables are cut.
Last research run
2026-07-10
Hardware figures checked against manufacturer specifications and project histories on this date; they age fast. The gradient framing itself comes from the P4A decentralised-compute papers, rapid iterations, not engineering guarantees.
The analogy we think with
Compute is water management.
On the driest inhabited continent, everyone understands water. You would not pipe water from a corporate dam a thousand kilometres away while rain falls on your own roof, yet that is exactly how most communities treat their data and their intelligence: shipped offshore for processing, rented back by subscription, cut off when the link drops.
So the gradient reads like a catchment: personal rainwater tanks, a neighbourhood dam, a town reservoir, and, far away and mostly optional, the corporate mega-dam. For remote communities and for any people whose relationship with the state or the market is complicated, the difference between owning the tank and renting the dam is not technical. It is sovereignty.
The pattern
Four levels, most of them already in your community.
L0L1L2L3A compact AI workstation drawing about as much power as an old light bulb (60–240 W class hardware, stamped 2026-07-10) can hold a nation's language model, maps and archives in the community hall, solar-powered, mesh-networked, and still working in a cyclone when the mainland tether fails.
One modern liquid-cooled rack (~120 kW class) is a municipal-scale supercomputer: digital twins of Country, land-management simulation, archive digitisation, owned by the nation, its waste heat warming the building, its potable-water use near zero in a closed loop.
Gigawatt campuses that train frontier models have their uses, but everyday community intelligence does not need them, and dependence on them recreates the extraction pattern this whole site exists to name. The proposal is not to abolish L3. It is to make it a choice.
Read against the record room: this is CARE's Authority to Control and OCAP®'s Possession with a power lead attached. A nation whose language model runs on its own L1 kiosk does not need to ask a platform's permission to speak its own language.
Established record, cited
A gold panel inside a purple room
Shared idle compute already made history once.
The idea that ordinary devices can do extraordinary work together is not speculative. In 2007, Stanford's Folding@home project, running on idle PlayStation 3 consoles in living rooms, became the first distributed computing network in history to cross a petaFLOP, recognised at the time as the most powerful computing network on Earth. Millions of people lent their sleeping machines to protein-folding research for medicine.
The precedent matters here for one reason: the world's most powerful computer was, briefly, a volunteer community. Nobody owned it but everybody ran it. That is the existence proof under this whole room.
Why a nation might care
Offline-first is remote-first is sovereignty-first.
A community-held model fine-tuned on a nation's own language archive, under the nation's governance, with elders deciding what it may and may not be taught, instead of a platform scraping the archive and speaking it back without consent.
Fire, water, fisheries and habitat simulation on community hardware, with the data classified under the nation's protocols, sacred sites simply absent from any layer that leaves Country.
When cyclone, fire or cable-cut severs the link, an L1 node sheds its optional loads and becomes the local beacon: offline maps, shelter directories, message relay across the mesh. Remote communities are last to be reconnected; a sovereign node never disconnected.
Every query answered on Country is a subscription not paid offshore, a dataset not surrendered, a skill grown locally. The intelligence dividend lands where the intelligence lives.
The seams in this proposal
What we don't know, said out loud.
Hardware prices and capabilities shift monthly; everything above dates. Community hardware needs maintainers, and training local maintainers is the real project. Small models are genuinely good now but not miracle workers; some tasks will still want L3, and pretending otherwise would be its own kind of dishonesty. And a rack in a community building is a governance question before it is an engineering one, who holds the keys is a question only each nation can answer.
pattern not productspecs dated 2026-07-10maintainers are the real infrastructurerefusal welcome