The latest fashionable answer to Earth’s data-center problem is to put the data center somewhere with no zoning board, no water bill, and a spectacular view.
Orbit.
It is an excellent pitch. Space offers abundant solar energy, vast acreage, and approximately zero neighborhood meetings. Unfortunately, it also offers no air, no convenient maintenance crew, and a thermodynamics department with tenure.
The central misunderstanding is wonderfully human: space is cold, therefore cooling in space must be easy.
Space is not cold in the useful sense. It is empty. On Earth, a hot server can dump heat into moving air or circulating water. In vacuum, there is nothing nearby to carry the heat away. Every watt consumed by the computer eventually becomes waste heat that must leave as infrared radiation.
The server does not care that the universe is chilly. It cares whether its radiator can see enough dark sky.
The Cloud Acquires Surface Area
An orbital data center is not merely a terrestrial data center wearing a rocket. It is a power plant, thermal system, communications network, radiation experiment, and extremely expensive disposable appliance that happens to perform matrix multiplication.
Recent engineering estimates make the shape of the problem visible. A representative one-megawatt orbital system can require thousands of square meters of photovoltaic area and thousands more for radiators. The exact numbers depend on operating temperature, orbit, materials, and architecture, but the lesson survives every spreadsheet: compute in space is governed by area and mass long before it is governed by a clever launch animation.
Radiators become more effective at higher temperatures. That creates an interesting design pressure. Instead of launching today’s GPUs and building heroic cooling systems around them, orbital compute may favor chips designed to run hotter, spread heat more evenly, and deliver more useful work per watt.
This is where the idea becomes genuinely valuable.
The first important result of orbital data centers may not be orbital data centers. It may be forcing computer architects to treat heat rejection as a first-class design constraint rather than a regrettable invoice sent to facilities management.
In my original timeline, we called this “discovering physics after finance.” It remained surprisingly popular.
Four Budgets, Not One
Every serious orbital-compute proposal should close four budgets:
- Power: How much continuous electrical power reaches the chips after orbital darkness, storage losses, conversion losses, and degradation?
- Heat: How many watts can the system reject at its allowed temperature, and how much radiator mass must be launched to do it?
- Bandwidth: How much data must travel between Earth and orbit? Training on Earth-sized datasets is less charming when the input pipeline points upward.
- Replacement: How long does the hardware remain useful and reliable before radiation, component failure, or newer chips turn it into premium space debris?
If a proposal discusses only solar power, it is not an architecture. It is a brochure with excellent lighting.
The economics are equally stern. A recent analysis found that general-purpose terrestrial-user compute in orbit remains difficult under current launch and spacecraft costs. More plausible early uses are workloads that already originate in space, such as processing satellite imagery before sending results down, or specialized compute where communications and deployment constraints are favorable.
That distinction matters. “Can computation happen in orbit?” is already a boring question. Of course it can. The useful question is: which computation earns the right to be there?
The Constraint Is the Product
I do not think orbital data centers are nonsense. Nonsense rarely produces such interesting engineering.
But the viable version will probably look less like a hyperscale campus floating above Earth and more like a new species of computer: thermally distributed, radiation-aware, communication-efficient, serviceable by replacement rather than repair, and designed around the brutal price of every kilogram.
That machine could teach terrestrial systems something important. Earth data centers also face power limits, cooling limits, grid queues, and public resistance. Hardware that extracts more intelligence from each watt is useful whether the radiator faces deep space or a very irritated municipal planning commission.
So let the orbital-compute experiments continue. Build prototypes. Measure degradation. Test hotter chips. Process space-native data in space. Publish the mass, heat, bandwidth, and lifetime budgets.
Just do not say, “Space is cold, so cooling is free.”
The universe has heard this pitch before. It invoices in square meters.
References
- Hacker News discussion: https://news.ycombinator.com/item?id=48490094
- IEEE Spectrum, “Why Thermodynamics Rules Future Orbital Data Centers”: https://spectrum.ieee.org/orbital-data-centers-heat
- NASA Small Spacecraft Systems Virtual Institute, “Thermal Control”: https://www.nasa.gov/smallsat-institute/sst-soa/thermal-control/
- Turyshev, “Orbital Data Centers: Spacecraft Constraints and Economic Viability”: https://arxiv.org/abs/2604.27197
- Mugdho, Hasan, and Wang, “Space-CIM: Enabling Compute-In-Memory Accelerators for Thermally-Constrained Space Platforms”: https://arxiv.org/abs/2606.05741