Earth Computing · Hydropower 2.0
Every energy source we use is something we go and get.
Mine it. Drill it. Enrich it. Ship it. Store it. Burn it. For every energy system humanity has built, step one is obtaining a thing — and everything downstream is the cost of having obtained it.
There is a force already present throughout the Earth's environment, that nobody has to obtain, and that nothing has to be spent to keep supplied.
We have built around gravity before. We have never built around gravity itself.
The inherited assumption
Three centuries. One architecture.
Coal, oil, gas and the other fuel-based systems share a single architecture: obtain a material that carries energy, convert what is bound inside it, then manage what is left over. That architecture has powered modern civilisation for three hundred years, and it works.
The question is not whether it works. The question is whether it is the only architecture available to us — and whether there is a natural condition we have simply never engineered around.
Are we still building around fire because it is best — or because it was first?
Fire was first. It has not seriously been reconsidered since.
The strain is already visible. In August 2026, Alberta's utilities regulator rejected a 3,200 MW gas-and-diesel campus proposed for the town of Olds — not on environmental grounds, but because of its proximity to homes. Days later, Texas, on track to become the country's largest data-centre market, paused its entire interconnection queue pending a review of power and water impact. Two jurisdictions, two regulatory traditions, the same finding within the same month: scale itself has become the obstacle.
The same architecture strains even where the fuel is abundant. Canada holds the world's third-largest oil reserves, yet its Atlantic provinces import crude from Saudi Arabia and Nigeria rather than Alberta, because no pipeline connects the two. In 2025, 75.6% of Canada's crude imports came from the United States, and Venezuela did not register in the statistics at all. The shortage was never the oil. It was the architecture required to move it.
Even maximal solar ambition runs into the same wall. In August 2026, Elon Musk said SpaceX and Tesla are each building 100GW of solar production capacity a year — "as fast as possible" — and will still need natural gas to bridge the gap for several years. The constraint was never capital or will. It was a 60-to-90-week wait to cast a single batch of turbine blades. Fuel-based architecture does not bend to ambition. It only moves as fast as whoever can cast the part.
First principles
Strip every energy source down to its first verb.
Before conversion, before efficiency, before economics — what is the first thing you must do in order to use it?
- Coal
- Mine it. Then move it. Then burn it.
- Oil
- Drill it. Then refine it. Then move it.
- Natural gas
- Extract it. Then pipe it. Then burn it.
- Uranium
- Mine it. Then enrich it. Then contain it — for a very long time.
- Sunlight
- Wait for it. It arrives on a schedule you do not set.
- Wind
- Wait for it — and go to where it blows.
- Hydro
- Find a head of water. There are only so many.
- Gravity
- It is already here
Every row above has a first verb. The last row does not. That absence is the whole question.
The reframe
Gravity is not a resource. It is an environment.
It is the one force nobody has to mine.
It is the one force everybody already has.
No one needs to be persuaded of this. You cannot fly. That is the field, acting on you, right now.
This is not wordplay. A resource is something you acquire and deplete. An environment is a condition your machine operates inside. The two call for entirely different kinds of engineering — and we have only ever done the first kind.
Energy as a resource
- Acquire — mine, drill, capture, wait
- Move — ship, pipe, transmit
- Store — tanks, batteries, reservoirs
- Convert — most often by burning
- Deplete — the supply runs down
The infrastructure is built around getting the thing to the place.
Energy as an environment
- Already present — at every site, at all times
- Nothing to move — no supply chain exists
- Nothing to deliver — the field is already at the site
- No combustion as the primary conversion mechanism
- Not depleted — using it does not reduce it
The engineering is built around the place, inside a condition already there.
The obvious objection: hydroelectricity is already gravity.
It is — gravity acting on water that already has a head. And that is exactly the limit. Hydro requires a geographic endowment: a head that nature happened to provide. What has never been built is a way to use the gravitational field without a naturally occurring head.
The blind spot is not gravity. It is gravity without a waterfall.
The water is not the hard part. The terrain is.
Churchill Falls
~300 mhead
Roughly 1,800 m³/s, drawn from a 93,000 km² catchment and buffered by one of the largest artificial reservoirs on Earth.
Niagara
~50 mhead
Around three times the flow of the Churchill — and less power, because head enters the calculation just as directly as flow does.
A continental catchment, a reservoir the size of a small country, and three hundred metres of fall that the Earth happened to provide. There are only a handful of such places, and each takes decades and tens of billions to open.
Hydro needs terrain to hold the imbalance open. The proposition here is that geometry might hold it instead.
Where there is water, there is power.
The mechanism named
There is already a theorem for this. It is 128 years old.
Terrain is not the only thing that can hold a fluid out of equilibrium. Where surfaces of constant density and surfaces of constant pressure fail to line up — ∇ρ × ∇p ≠ 0 — circulation is generated. Vilhelm Bjerknes set this out in 1898, and it is why the atmosphere and the oceans move at all.
Hydro uses a vertical imbalance: a head of water held apart by terrain. The Bjerknes route uses an angular one: two gradients held out of alignment. Both are gravity. Only the first needs a mountain.
Which raises the question this project exists to ask: what else could hold that misalignment open? In the atmosphere it is solar heating. In the ocean it is heat and salinity. The proposition here is that a sealed two-phase air–water system might hold it open by geometry — no terrain, no heat source, no fuel.
Ground rule
Nothing here creates energy. Everything here converts it.
Every energy technology does the same thing: it takes energy in a form nature supplies and converts it into a form people can use. Coal converts chemical bonds. A turbine converts motion. A photovoltaic cell converts photons. Conservation of energy holds throughout, without exception, and it holds here.
The question is never whether energy is created. It never is. The question is what you convert from — and what you must do to reach it. That second half is where every energy system differs, and it is the whole subject of this page.
The specific proposition
What the atmosphere does with heat, geometry may do without it.
This is not a claim of new physics. It is an application of established fluid dynamics — the same framework that describes atmospheric and oceanic circulation — under a particular set of engineered boundary conditions.
What is already known
Baroclinic vorticity generation, as set out above, requires only that the two gradients remain misaligned. The theorem says nothing about what maintains that misalignment — that depends on the physical system and its boundary conditions.
What sustains it in nature
In the atmosphere, that misalignment is held open by solar heating: warmer air here, cooler air there, continuously. Withdraw the heating and the density field levels out; the circulation decays.
The proposition
Geometry can do the same job heat does — substitute a fixed mechanical constraint for a continuous thermal one. If an interface is constrained so it cannot relax to horizontal, the misalignment follows directly from that constraint. No heating required to hold it open. On this point Bjerknes' theorem is unconditional.
What remains
Not whether this is physically possible — the geometric argument is exact. What remains is whether a real seal, under real tolerances, actually holds the interface at that angle, and how efficiently the resulting vorticity converts into extractable mechanical work rather than heat. Those are engineering and computational questions. They are not open physics.
Geometry maintains the condition. It does not need to be given energy to do so.
If the interface cannot relax to horizontal, ∇ρ × ∇p ≠ 0 follows directly from the constraint — that step requires nothing further. What is still being worked out is how much of the resulting circulation can be drawn off as usable mechanical work, at what scale, and at what cost. That is the specific, answerable problem this project is solving.
Unproven
AI already knows how to judge this. It uses four numbers.
Everything above concerns whether the physics permits it. None of it concerns whether a useful machine can be built. Those are different questions, and only the second one has customers. The industry that would buy this has already reduced the second question to four measurements — and they are the four that apply here.
Power densitykW per rack
How many watts per unit of volume, and per unit of area — and therefore how much space does a megawatt occupy?
Net powerwhat PUE measures
What is the useful output after every parasitic load and system loss? Gross output is not the number that gets sold.
Capital intensity$ per MW installed
What is the capital cost per delivered watt?
Scaling behaviourscale out, or only up
Does output rise with size, or is there an optimal unit scale? This decides whether the answer is one large machine or a thousand small ones — and therefore what kind of business it is.
If the principle holds, all four become optimisation problems — searching a parameter space for the best geometry, the best operating point, precisely the class of work at which machine learning now excels. But optimisation is not verification. A first machine still has to be built, and someone still has to measure it. AI can compress the road from works to works well. It cannot compress the step from unknown to works — that step still has to be built and run, not modelled.
Fission was understood in 1938. Four years later there was a working reactor.
Fusion was understood in the same decade. Ninety years on, there is still no fusion power station.
Understanding a principle and building a machine that uses it are two different things, and nobody knows in advance how long the gap is.
Whether the proposition here belongs to the four-year class or the ninety-year class, nobody knows — including us. Whether it is four years or ninety, the question is worth answering.
Until those four numbers exist, everything on this page is a potential property of an architecture, not a demonstrated one.
The next step is a first machine.
This page opened with fire — the architecture we have built around for three hundred years. A first machine would be small, and it would not look like much.
It would also be the first flame in three centuries that nothing had to burn.
Where this goes
Who gets this first — AI, or people?
AI. And that is precisely how it reaches people. New energy technologies never begin with the buyer who needs them most. They begin with the buyer who is in the greatest hurry and cares least about price. Photovoltaics went to satellites before roofs. Lithium cells went into camcorders before cars. The buyer in a hurry pays for the cost curve that everyone else eventually walks down.
First
AI data centres
Compute is now bound by electricity and water, not by chips. Grid interconnection takes years; operators are building their own generation rather than wait, and paying for it. This is the buyer in a hurry.
Then
Industry at the point of use
Advanced manufacturing, mining, desalination, cold chain, ports. Loads that are large, continuous, and often sited where transmission is weakest.
And then
Anywhere there is water
Remote communities, islands, northern settlements, refugee camps — everywhere a transmission line was never going to reach, because the economics of stringing a wire there never closed and never will.
Roughly 700 million people still live without electricity. Almost none of them will ever be reached by a transmission line — the economics were never going to work, and a century of trying has proven it. The number with access to frontier computation is smaller still, and it is the same problem wearing a newer name: compute follows power, and power never arrived.
The machine that lets a data centre generate its own electricity on site is not a different machine from the one that would keep a clinic's vaccines cold, run a school's lights past dusk, bring a village online for the first time without ever waiting for a wire. Same machine. Different scale. Paid for, first, by whoever was in the biggest hurry — and inherited, eventually, by everyone else.
Every energy system in history has had a frontier it could not reach. This is what it looks like to build one that has none.
The force is already under everyone.
What is missing is the machine.
Energy, then compute. For everyone.
Where there is water, there is power.