Pluvio
Next-generation radiators · for satellites
AI ran out of power before it ran out of chips.
The limit used to be silicon. Chip factories could not keep up. That is being fixed: fabs are being built and supply is catching up.
The limit now is electricity, and the grid is not catching up.
Global data centre power ran 49 gigawatts in 2023. It reaches 96 by 2026. Getting to the 200 gigawatts the industry expects by 2030 means building 26 gigawatts of new capacity every single year. The world has never managed more than 15.7.
That is not a gap you close with money. It is a gap you close by finding power somewhere else.
Everywhere on Earth, power has a floor.
You can buy cheap electricity. You cannot buy cheap electricity that is there when you need it.
Solar and wind are cheap, and they get cheaper: $16 to $19 per megawatt-hour by 2050. But they only work when the weather cooperates, and a data centre cannot pause overnight. To run around the clock you need firm power: nuclear, gas, geothermal. That costs $59 to $67.
The $40 gap between those two numbers is not a technology problem waiting on a breakthrough. It is the price of fighting the fact that the sun sets. The US National Renewable Energy Laboratory projects it every year out to 2050, across every scenario they model. It never closes.
Firmness is exactly what a data centre buys. On Earth, it always costs extra.
In orbit, the sun never sets.
Put the same solar panel in the right orbit and it produces power constantly. No night. No weather. No seasons. Roughly 40% more intense, because there is no atmosphere in the way.
Orbital compute does not buy firmness. It has it, structurally, for free.
This is not a thought experiment. Starcloud has raised $170 million at a $1.1 billion valuation to build it. SpaceX has filed with the FCC for one million orbital data centre satellites. The European Commission studied the idea and concluded it works.
Serious people, with serious money, are already building this.
But free power is not the same as cheap compute.
Orbit has to beat Earth's floor, not Earth's average. Right now it does not come close.
A gigawatt of compute in orbit costs $6.44 billion per year. The same gigawatt on Earth, power and infrastructure included, costs $1.70 billion.
Orbit loses by nearly 4x.
Four things decide that number.
Launch cost. Solar efficiency. Compute density. Heat rejection.
That is the entire equation. Change those four and you change the answer.
Three of them get better every year.
Rockets are getting cheaper. Falcon 9 delivers a kilogram to orbit for about $2,700 today. Starship targets $100.
Solar panels are getting lighter. Chips are doing more work per watt.
These three ride the same industrial learning curves that made AI possible on Earth in the first place. They have improved for decades. They will keep improving. These are the trends that convinced serious people to start building.
The fourth one has not moved at all.
A chip in orbit turns electricity into heat, exactly like a chip on Earth.
On Earth you blow air over it, or run water through it. In space there is no air and no water. There is nothing to carry heat away. The only way to shed it is to radiate it into the dark, and that takes enormous panels: heavy, fragile, and growing in lockstep with every watt of compute you add.
There is no learning curve underneath this. It is not an engineering problem that yields to iteration. It is a physical law, governing how much heat a surface can shed at a temperature a chip can survive.
Every watt you add is a watt you have to radiate. Permanently.
That one subsystem is 76% of the cost.
Not the rockets. Not the solar. Not the chips.
Everything else combined is the other quarter.
The radiator alone costs 2.9 times more than an entire terrestrial data centre.
And cheaper rockets make it worse.
Here is the part nobody expects.
At today's launch prices, heat rejection is a third of the cost. At Starship's target price it is three quarters. Make launch completely free tomorrow and it becomes 80%.
Every dollar taken out of launch raises the share of the remaining bill that is radiator.
The industry is racing to fix the part that is already working.
Pluvio moves the fourth line.
We build the thermal layer that does not scale in mass with power.
Same rocket. Same chips. Same solar. One subsystem replaced.
A gigawatt of orbital compute goes from $6.44 billion per year to $1.37 billion.
Earth's floor is $1.70 billion.
Orbit wins.
The market is already funded. The physics is what is missing.
Every argument on this page except the last one has already been made, and made well, by people who moved first and moved correctly. They were right about the demand. They were right about the power. They were right that three of the four constraints are falling.
They are stuck on the fourth, because inside the architecture everyone is using, it has no answer.
We are not betting on orbital compute happening. We are the reason it can.