Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarHabr AI→ original

Orbital Data Center at 13 kW: Radiator Up to 36 m², Launch Costs 3.4–13.5× Over Budget

An orbital data center faces more than launch costs. In a vacuum, heat leaves only by radiation, so the system needs large radiators. A 13 kW compute module with support systems (15.6 kW total), with a radiator at 40–60 °C, requires 16–36 m² of double-sided panel, depending on the orbit. Meanwhile, launching a 1 MW satellite on Falcon 9 costs 3.4–13.5× more than the permissible budget. The project therefore has to be evaluated as a whole system, not by the price per kilogram.

AI-processed from Habr AI; edited by Hamidun News
Orbital Data Center at 13 kW: Radiator Up to 36 m², Launch Costs 3.4–13.5× Over Budget
Source: Habr AI. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceHabr AI ↗
Editor-in-Chief
Key metric #1An ideal radiator at 80 °C emits 880 W/m², while a real one emits less
Budget for 1 MW$250–1000 per kg for a 40-tonne spacecraft
Key metric #3In geostationary orbit, ping is about 600 ms, and launch costs are 3–5× higher

Debates about data centers in space usually come down to one question: can Starship make launches cheap enough? But delivering kilograms to orbit is only part of the bill. Alexander Sapronov, head of Edge AI at MWS's R&D center, worked through what else would have to go to space along with the servers, using a 13 kW compute module as an example.

The core problem is simple: a vacuum has no air or water, so heat from chips can only be shed by radiation, which requires radiators of considerable area. For a 13 kW module with a radiator at 40–60 °C, that means 16 to 36 m² of double-sided panel. For a 1 MW data center, the Falcon 9 launch alone costs 3.

4–13.5× more than the permissible budget.

Three different economics of orbital data centers

A space data center does not have to be a copy of AWS. Vyacheslav Turyshev, a physicist at NASA's Jet Propulsion Laboratory (JPL), proposed dividing such projects into three classes. C1 means processing data directly on the satellite: the spacecraft itself recognizes objects in images and downlinks only classes, coordinates, and metadata, not raw images.

In a study using Sentinel-2 imagery, this processing cut the volume of transmitted data by 99.7–99.99% in some tests, and by 99.

49% for 3D reconstruction, which is nearly 200-fold. However, in that experiment the computation ran on Earth. C2 means computing within an already operating communications constellation, on its platforms and control systems.

C3 is a general-purpose data center for terrestrial users that must compete with ground-based facilities on compute price. The closest example is SpaceX's Starmind constellation, but its specifications are hard to rely on: Elon Musk's companies' plans change often.

For class C3, everything comes down to money. Building and equipping a terrestrial data center costs $10–40 thousand per kilowatt, excluding servers from the comparison. If each kilowatt carries 40 kg of equipment, a 1 MW spacecraft weighs 40 tonnes. To fit within the budget of a terrestrial data center, manufacturing and launch must cost $10–40 million, or $250–1000 per kilogram. According to Turyshev's estimate, a Falcon 9 launch alone is 3.4–13.5× more than this figure, and that is without the satellite, communications, or operations.

How much heat, and where the power comes from

To the 13 kW compute load, the author added another 20% for communications, attitude control, and coolant pumps, reaching 15.6 kW. Solar arrays were chosen as the power source: an RTG of that output would be too heavy and expensive, and space reactors are still in development. Gallium arsenide and silicon cells are under consideration, while perovskite is excluded because its long-term durability is unclear. In Earth's shadow, the module runs on lithium-ion batteries.

For radiators, estimates often use an ideal blackbody at 80 °C, which emits about 880 W per square metre. Sapronov calls this an unattainable maximum: it ignores real materials and heating from the Sun and Earth. The calculation assumes an emissivity of 0.

85, and in low orbit it adds a thermal flux from Earth of 150–240 W/m². The radiator emits from both sides, so the physical area is half the emitting area. For example, at 60 °C the result is 18 m² in a terminator orbit, 24 m² in a typical low orbit, and 16–18 m² in geostationary orbit.

At 40 °C the area grows to 25, 36, and 21–23 m² respectively. The hotter the radiator, the more compact it is, but the hotter the chip gets, which is harmful to silicon.

What the orbit changes

In low Earth orbit up to 800 km, delays run to tens of milliseconds, but the satellite regularly passes into Earth's shadow: about 5,500 times a year, and at 500 km for up to 35 minutes per eclipse. In a terminator sun-synchronous orbit (about 700 km), the spacecraft is sunlit 97% of the time, so the battery can be smaller, but thermal management becomes harder. In geostationary orbit (35,800 km), there are about 90 eclipses a year lasting 70 minutes each. The battery endures fewer cycles, but with deeper discharges. Radiation is higher there, launches cost 3–5× more (about $3,250 per kg to low orbit versus $8,900 to geostationary transfer orbit), and real ping is about 600 ms.

The practical takeaway: for projects like C3, looking at launch price alone is not enough. Mass and area of panels, batteries, and radiators must be counted. Scenarios C1 and C2, where computation saves communication bandwidth, look more grounded. A caveat: the calculations are simplified and describe a hypothetical module, not a specific project.

Story

Key Questions Answered

What are the key details on Three different economics of orbital data centers?

A space data center does not have to be a copy of AWS. Vyacheslav Turyshev, a physicist at NASA's Jet Propulsion Laboratory (JPL), proposed dividing such projects into three classes. C1 means processing data directly on the satellite: the spacecraft itself recognizes objects in images and downlinks only classes, coordinates, and metadata, not raw images.

How much heat, and where the power comes from?

To the 13 kW compute load, the author added another 20% for communications, attitude control, and coolant pumps, reaching 15.6 kW. Solar arrays were chosen as the power source: an RTG of that output would be too heavy and expensive, and space reactors are still in development. Gallium arsenide and silicon cells are under consideration, while perovskite is excluded because its long-term durability is unclear. In Earth's shadow, the module runs on lithium-ion batteries.

What the orbit changes?

In low Earth orbit up to 800 km, delays run to tens of milliseconds, but the satellite regularly passes into Earth's shadow: about 5,500 times a year, and at 500 km for up to 35 minutes per eclipse. In a terminator sun-synchronous orbit (about 700 km), the spacecraft is sunlit 97% of the time, so the battery can be smaller, but thermal management becomes harder. In geostationary orbit (35,800 km), there are about 90 eclipses a year lasting 70 minutes each.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…