Orbital Data Centers: Why Starcloud’s Project Is Impractical

The surging demand for AI has led to a global race for technological dominance, impacting energy consumption, trade agreements, and even inspiring ambitious, albeit questionable, projects like orbital data centers. AI currently accounts for nearly 2% of total energy generation, a figure projected to double by 2030. This demand has driven up energy costs and spurred growth in nuclear energy stocks. The most extreme proposals, however, come from billionaires who envision solar-powered data centers in Earth orbit, leveraging reusable rockets to reduce launch costs. Starcloud, for instance, secured $170 million in funding in March 2026 for such an idea.

The Flawed Vision of Orbital Data Centers

The concept of placing data centers in orbit, while seemingly innovative, faces significant practical and engineering challenges.

Power Requirements and Scale

A modern data center relies on powerful server racks, such as the Nvidia NVL-72, which houses 72 GB200 GPUs, 36 CPU cores, and 17 terabytes of RAM. This single rack provides 720,000 teraflops (FP8) of compute at a cost of approximately 120 kilowatts of electricity, equivalent to the power consumption of 60 suburban homes. These racks are essential for handling the massive data processing required by large language models.

In contrast, existing satellite hardware, like the AMD Versal System-on-Chip in Starlink satellites, is significantly less powerful (a thousand times less) and consumes far less power (a hundred times less), with a peak power generation of only 5 kilowatts. This is nowhere near the demands of a single Nvidia NVL-72 server.

Starcloud's white paper proposes putting 5 gigawatts' worth of these servers in space, with Nvidia server racks arriving in 40-megawatt containers. This would be eight times more powerful than the largest operational data center on Earth today. The primary goal is likely on-orbit processing of Earth observation satellite data, as beaming down terabytes of high-resolution imagery and synthetic aperture radar data is becoming a bottleneck. This would allow for processing images in space and only sending down relevant data.

Engineering Challenges

Building a satellite capable of housing such a data center presents numerous engineering hurdles:

Power Generation and Size

A 5-gigawatt solar panel system, using high-efficiency panels generating 400 watts per square meter, would require an area of 12.5 million square meters. Starcloud's rendered concept uses 16 million square meters, which is nearly 5,000 times the surface area of the International Space Station (ISS) solar panels. The ISS took decades to assemble, highlighting the immense scale and complexity of such a project.

Heat Dissipation

The 5 gigawatts of power generated would also produce 5 gigawatts of heat. In the vacuum of space, heat cannot be dissipated through convection. The only way to remove it is through radiation using radiator panels. The Stefan-Boltzmann equation dictates the required radiator area. While a perfect blackbody has an emissivity of 1, radiators in space are typically coated with white materials like AZ93 (emissivity of 0.92) to minimize solar absorption while maximizing heat emission.

Starcloud aims to maintain server temperatures at 20°C, similar to terrestrial data centers. At this temperature, radiators emit 380 watts per square meter. To dissipate 5 gigawatts of heat, a radiator would need to be four kilometers tall and nearly a kilometer wide, occupying the central part of the satellite.

Fluid Circulation

Such a massive radiator would require tiny coolant channels circulating fluid between the servers and the radiators. Starcloud's paper vaguely states that "a workable design is possible without heat pumps," ignoring the immense fluid flows required. If a coolant like glycol were used, entering the radiator at 35°C and returning at 5°C, it would necessitate circulating 68,870 kilograms of fluid per second. This is equivalent to emptying an Olympic swimming pool in 40 seconds or the pumping power of 134 Space Shuttle RS-25 rocket engine turbopumps. The paper fails to address the constant maintenance required in orbit, or the risk of micrometeorites piercing these high-mass flow rate lines, as seen with the ISS's ROSA solar panels.

Orbital Stability

Large, flat panels extending from a satellite pose significant challenges for maintaining a stable orbit. The ISS regularly performs boost burns to counteract aerodynamic drag. For a satellite of this proposed mass, with hundreds of thousands of tonnes of fluid flowing to its extremities, Earth's uneven gravity would constantly pull its far-flung edges, adding to off-course forces. Satellites typically use inertia wheels for orientation control, but the moment of inertia for such a massive and spread-out satellite would be enormous, a detail also overlooked by Starcloud's white paper. Furthermore, their renders show all surfaces facing the sun, which would expose radiators to constant solar absorption, decreasing efficiency. A more sensible design would use an X-shape, with solar panels facing the sun and radiators edge-on.

Material Degradation in Orbit

Space environments rapidly degrade materials. Atomic oxygen in the ionosphere chemically attacks surfaces, while the Van Allen belts trap high-energy particles that damage radiators and solar panels. Ultraviolet radiation, solar flares, cosmic rays, and orbital debris also contribute to degradation. The ISS has tested over 1,500 material samples since 2001 to identify resilient materials. While AZ93, a common radiator coating, is resistant, its emissivity can still drop from 0.92 to 0.90 over time. Orbital debris strikes are also a constant threat, leading to leaks and coolant loop drainage on the ISS.

Radiation Effects on Electronics

Ionizing radiation can burn out transistors or flip bits of information in satellite electronics, leading to data corruption. Hewlett Packard Edge Computer Servers on the ISS mitigate this by running three instances of the same calculation on different nodes and comparing results, tripling power draw and mass. Space-rated chips are significantly slower than commercial hardware due to the hardening required to withstand cosmic rays. Starcloud's white paper dismisses this issue, claiming "logic devices are resistant to radiation" and that their larger server containers would require comparatively less shielding due to a higher volume-to-surface ratio. This implies that smaller data center satellites would struggle even more.

Cost and Feasibility

Starcloud's cost estimates are highly optimistic, particularly regarding launch weights and costs.

Mass Discrepancies

Starcloud claims a single 100-tonne launch can deliver a 40-megawatt server container, equating to 400 watts of compute per kilogram launched. However, the Nvidia servers they reference weigh 1,400 kilograms and draw 120 kilowatts, achieving only 88 watts per kilogram. Starcloud's figures are four times higher than even SpaceX's TERAFAB project, which aims for 100 watts per kilogram for space data center hardware.

For solar panels, Starcloud quotes 1,000 watts per kilogram. Realistic advanced lightweight ROSA arrays achieve 100 watts per kilogram, while lab-stage micrometers-thick silicon panels might reach 300 watts per kilogram. A more realistic weight for 5 gigawatts of solar panels would be 50,000 tons, not the 5,000 tons quoted by Starcloud.

Starcloud's white paper provides no mass figures for their radiators. Based on calculations, a radiator 4 kilometers tall and 840 meters wide, using current-generation panels (10 kg/sqm), would weigh 33,600 tonnes. Even with advanced carbon-carbon radiators (4 kg/sqm), it would still be 13,440 tons.

Considering all components (pumps, coolant, shielding, fuel, inertia wheels, structures), Starcloud's station would exceed 113 million kilograms, more than an aircraft carrier in orbit and over six times the total mass ever launched into space.

Launch Costs

While SpaceX's Starship lacks a public price menu, a known customer, Voyager Technologies, signed a deal for a $90 million delivery into orbit, translating to $900 per kilogram. For Starcloud's estimated mass, this would result in $102 billion in launch costs alone. Starcloud, however, quotes $30 per kilogram, a figure that would barely cover fuel and launch operations.

Economic Viability and Risks

AI data centers operate under intense competitive pressure, with chips typically having a 2-4 year lifespan before needing upgrades. An orbital data center, even with protective measures, would likely experience more failures and a shorter useful lifespan due to the harsh space environment.

Many AI companies are expected to fail to return a profit, with some analysts predicting OpenAI could run out of funds by 2028. If terrestrial AI companies fail, their computational power and energy generation infrastructure remain on Earth for other uses. In contrast, Starcloud envisions recovering server containers at the end of their life, but the cost of sending an empty rocket to orbit offers little discount. A serious fault in an orbital data center could render the entire satellite useless, leaving nothing behind.

Alternative Concepts and Future Applications

While Starcloud's proposal appears flawed, other concepts for space-based computing are being explored. Google's "Suncatcher" project, for example, addresses some of Starcloud's design mistakes by proposing a constellation of smaller satellites.

Google's Suncatcher

Google's approach involves a network of smaller data center satellites that combine their computational power. This necessitates efficient communication, with laser networking being a key technology. However, the inverse square law makes laser communication challenging without fiber optic cables, as signal power drops drastically with distance. Therefore, keeping satellites close together and maximizing inter-satellite connections is crucial.

Google has calculated a unique bounded orbit for pods of 81 satellites to orbit together, navigating Earth's complex gravity without heavy support structures. These satellites would orbit in a Sun-Synchronous orbit, continuously harvesting solar energy. However, this is an increasingly busy orbit, and data center constellations would exacerbate the problem of space debris. SpaceX's Starlink constellation performed 300,000 collision avoidance maneuvers in 2025 alone. A bounded orbit of 81 satellites means a single collision avoidance maneuver could trigger adjustments for all 81 satellites, multiplying the burden.

Liquid-Droplet Radiators

Another promising technology is liquid-droplet radiators, which eliminate heavy pipes and radiator structures by spraying a stream of coolant oil directly into the vacuum of space. Each droplet's surface acts as a radiative surface, drastically changing the physics of heat dissipation.

The Real Motivation: Military Applications

Despite the technical and economic challenges, the question remains: why put computers in space? Space is becoming a new layer of the internet, with terabytes of sensitive military data being transmitted. Events like the conflict in Ukraine, where autonomous drones are used with satellite intelligence, highlight the future of warfare. Processing this information quickly is a critical military asset.

Space-based data centers, with their uninterruptible power, could become near-impossible targets for most adversaries, making them highly valuable strategic assets. The military, with its deep pockets, is likely the primary driver for such ambitious projects.

The Value of Technical Journalism

In a world filled with hype and speculative ventures, reliable technical science journalism, such as that provided by IEEE Spectrum, is crucial. It helps in understanding emerging technologies and identifying legitimate breakthroughs. For instance, IEEE Spectrum has covered how AI can provide real value, such as in particle physics research, where machine eyes might detect patterns missed by humans, leading to new discoveries. They also reported on a new cancer treatment developed with CERN, using particle acceleration technology for precise, high-power radiation delivery with less damage to healthy tissue. Such publications offer insights into the work of accomplished engineers and provide a valuable resource for staying informed about technological advancements.