From the Cloud to Orbit: When Earth Runs Out of Power


Transcript of the narrated version (4 min).
Narrated with a synthetic voice (Larry). The writing is Esteban Rey's — kilowatto.com.

---
From the Cloud to Orbit: When Earth Runs Out of Power. 

For two decades, we've been living under a misleading metaphor. We call it the Cloud, but it's actually a deeply terrestrial infrastructure, consisting of thousands of acres of concrete warehouses, underwater copper and fiber cables, and above all, a voracious dependence on local power grids. But this terrestrial Cloud is reaching its limits. Between the thirst for water to cool and the hunger for gigawatts of Artificial Intelligence, the Earth is getting too small for us. We're on the cusp of a new era, the birth of the Over Cloud or Orbital Computing, which is not science fiction, but a thermodynamic and economic necessity that we will see take off between 2025 and 2035.

The Ultimate Energy Arbitrage. 

 The reason to take servers to space isn't romanticism about exploration, it's pure financial math. On Earth, a data center struggles with intermittency, such as nighttime, cloudy conditions, or a saturated grid in locations like Virginia or Querétaro. In low Earth orbit, above the real clouds, the sun never turns off, allowing an orbital data center to access high-intensity solar energy 24/7. This represents the dream of any hyperscale engineer: clean, constant, and infinite energy, without the need to fight for municipal permits or connections to the national power grid. Companies like Starcloud, formerly Lumen Orbit, and projects like Google's Suncatcher are already doing the math to put NVIDIA H100-class hardware into orbit. However, if energy is the incentive, physics is the limitation.

The Paradox of Cold and Thermal Hell. 

 There's a popular myth that space is cold, so cooling servers should be easy. Nothing could be further from the truth. On Earth, heat is dissipated through convection, using air or water. In the vacuum of space, however, there's no air, which means the heat has nowhere to go. An AI server is essentially a high-power electric stove, and in a vacuum, the heat generated gets trapped in the chip unless it's actively radiated. The biggest engineering challenge of the Over Cloud isn't getting the servers into space, but rather preventing them from melting. This requires giant radiators and complex fluid systems to expel the heat in the form of infrared radiation. If a cooling pump fails on Earth, the server shuts down, but in space, it can overheat and cook in its own heat in a matter of seconds.

The Invisible Killer: Cosmic Rays. 

 But let's assume we solve the heat problem. We're then faced with the silent sniper: radiation. Down here, the atmosphere protects us. Up there, the servers are naked against cosmic rays and high-energy particles from the sun. For a silicon chip, this is lethal. A single charged particle that passes through a transistor can cause a Bit Flip, a change from 0 to 1 or vice versa. Imagine you're training a critical financial or medical model. A bit flip in RAM could corrupt weeks of training or alter a diagnosis. On Earth, we use Error Correction Code memory to mitigate this, but in orbit, the error rate skyrockets exponentially. This forces us to rethink the hardware architecture. We can use physical shielding, such as adding lead or heavy polymers, which drastically increases the launch cost, since putting weight into space is still expensive. Alternatively, we can use logical redundancy, having three processors doing the same calculation and voting on the correct result. If one disagrees due to radiation, it's discarded. This, of course, reduces the efficiency we're trying to achieve in the first place.

Is the Future Exospheric?. 

 Despite the thermal and radiative challenges, the trend is clear. If AI continues to consume energy at its current rate, no power grid on Earth will be able to handle it. The Exospheric Frontier will stop being an academic paper and become critical infrastructure. In as little as five years, when you ask an AI to draft a contract or generate a video, the processing may not happen in a basement in Arizona, but rather in a metal box floating 400 kilometers above your head, powered by pure solar light, and having to fight against the heat of the vacuum and dodge cosmic radiation in order to deliver your response.