Data Centre In Space (Part 2): The Engineering Bottlenecks & Economics Behind Orbital Computing
Space-based AI data centres offer sustainable power and scale in theory, but these structures face complex engineering challenges like thermal management, maintenance, and communication.
Published : July 31, 2026 at 12:02 PM IST
As the stress caused by AI data centres becomes too big to ignore, the industry has finally started to look at possible ways to mitigate the risks. One such remedy is akin to a sci-fi movie plot where we take the compute to outer space and harness the energy of the sun to power the workload. From American companies like Google, Axiom Space, Starcloud and xAI to Indian startups such as Agnikul and TakeMe2Space, a growing number of organisations believe the concept deserves serious consideration.
According to Starcloud's white paper, gigawatt-scale orbital data centres represent one of the most ambitious space projects ever proposed. It attributes their feasibility to four major trends: declining launch costs, rising electricity demand, the rapid growth of AI compute requirements, and increasingly affordable satellite connectivity. The company believes orbital data centres are both economically viable and necessary to support AI's long-term growth sustainably.
However, in reality, the concept is not without its challenges. While most of the requirements for a working orbital data centre are within reach, several engineering challenges like thermal management, maintenance work, and high-bandwidth space-to-ground communication still await solutions. Additionally, finances may take a little longer to become realities, and Google acknowledges them in its own paper.
ORBITAL DATA CENTRES INVOLVE COMPLEX ENGINEERING
Talking to ETV Bharat, Ronak Kumar Samantray, founder and CEO of Hyderabad-based Indian space tech company TakeMe2Space, explains that the biggest challenge of an orbital data centre is that all the moving parts become interdependent in space.
He says that while data centres on Earth can rely on the grid, large-scale cooling systems, physical maintenance teams, and frequent hardware replacement, in orbit, power, thermal control, compute, storage, communications, launch mass, radiation tolerance, and orbital operations all have to work within the limits of a spacecraft.
Apart from challenges around launch schedules, mission qualification, system reliability, regulatory approvals, orbital debris mitigation, and customer trust, Samantray specifically highlights sustained energy for high-performance computing, thermal management, and communications as major obstacles.
“The question is not only whether compute can work in space. The real question is whether compute, power, cooling, networking, data security, and commercial demand can come together as one viable system,” he says. “That is why we are building this in phases, starting with in-orbit AI processing and then scaling towards larger orbital data centre capability over time.”
Google, in its paper, discusses the major ingredients required for a working orbital data centre and explores the feasibility by addressing fundamental challenges in inter-satellite communication, orbital formation control, radiation hardness, and other aspects. It proposes a system design that utilises fleets of satellites equipped with solar arrays and advanced communication technologies to create a space-based machine learning environment that can achieve performance roughly comparable to a terrestrial data centre.
While Google believes to have practical solutions for most of the obstacles, the paper highlights three bottlenecks that require testing as well as solutions, namely thermal management, on-orbit reliability and repair strategies, and high-bandwidth ground communications.
BOTTLENECKS AWAITING SOLUTIONS
Thermal Management: In space, heat cannot be removed by air like it is on Earth. Talking about the misconception associated with the “free cooling” part in the vacuum of space, Ronak Kumar Samantray of TM2Space called thermal management one of the biggest challenges for an orbital data centre.
“Space does not offer cooling in the way people understand it on Earth. In vacuum, there is no air and therefore no convection. Heat has to be moved through the spacecraft and rejected through radiation. That means thermal design, radiator surfaces, coatings, power budgets, compute loads, and orbital conditions all have to be engineered together. For us, thermal management is not a side problem; it is one of the core design constraints,” he says.
In simpler words, heat normally escapes when hot air rises or when cooler air particles bump into the warm object, carrying the heat away. However, space is a vacuum with no air particles to carry the heat. The only way for heat to escape is by converting itself into light waves (thermal radiation). Since light waves do not need molecules to travel, the heat can escape into space as infrared light—similar to how sunlight travels from the Sun to Earth.
Based on this principle, NASA uses an Active Thermal Control System (ATCS) to get rid of waste heat on the International Space Station. It circulates water through an internal loop to absorb heat from the cabin, experiments, and equipment. The heat is then transferred via exchangers to an external ammonia-based loop, which dissipates it into space, helping maintain stable operating temperatures.
However, to account for the immense heat generated by data centres, the thermal management system needs an upgrade. Google, in its paper, called thermal management one of the biggest challenges for deploying power-dense Tensor Processing Units (TPUs) in space. It says that since vacuum conditions make heat dissipation difficult, advanced thermal materials, passive cooling systems, and efficient heat transport mechanisms would be essential.
On-Orbit Reliability and Repair: In a terrestrial data centre on Earth, a technician can physically access a rack, replace a server, fix a cable, or check a cooling system, but the same liberty is not available in space.
According to Google, the "simplest solution is redundant provisioning", meaning adding extra backup parts and extra capacity right from the start so that the system keeps working even when some parts break. Samantray agrees with it and adds some other measures to the mix.
Highlighting the need for safeguards at multiple levels for machines in orbit, he says, “The system has to be designed to protect itself, recover from faults, and continue operating with minimal human intervention. He also highlighted a three-point approach for the same, which is detailed in the table below:
“For an orbital data centre, the design philosophy is simple: assume that faults can happen, detect them early, contain them quickly, and recover without depending on physical intervention,” Samantray says.
As far as maintenance work is concerned, Samantray emphasises that it should be remote and software-led, autonomous where possible, redundancy-driven (just like what Google suggested) so that critical systems cannot depend on a single point of failure, and modular by design to support easy replacement or fresh upgrades.
High Bandwidth Ground Communication: As for the final bottleneck, Google highlighted the need for reliable high-speed optical communications between satellites and ground stations for efficient working of large-scale space computing.
“Developing robust optical satellite-ground communications will also be critical for scaled operation, but will necessitate overcoming challenges including atmospheric turbulence, high-speed relative motion errors, and precision beam tracking,” it says, adding that it is an active field of development, with the current leader being NASA’s TeraByte Infrared Delivery (TBIRD) mission, which demonstrated 200 Gbps ground-LEO communications in 2023.
For the pilot project, Google's paper suggests that radio communication would be sufficient, as it is less affected by atmospheric interference. However, optical links will eventually be required to enable high-bandwidth communication with ground stations.
LAUNCH ECONOMICS AND FINANCES
Beyond the technicalities and science behind the working of a data centre in space, launch economics and financials are obviously the single major aspect that may make or break the orbital computing dream.
Google's observations consider the project to be financially sound and expect it to become even more feasible as launch costs continue to drop.
“Launch costs are a critical part of overall system cost,” Google says, highlighting that a learning curve analysis suggests launch to low-Earth orbit (LEO) may reach around $200/kg by the mid-2030s, by which time the launch economics would start to make sense.
Google says its analysis compared the mass-to-power ratios of satellites such as Starlink V2, Starlink V1, OneWeb, and Iridium, finding that if launch costs to Low Earth Orbit (LEO) fall to $200 per kg, the cost of launching power infrastructure could be roughly comparable to the electricity costs of terrestrial data centres on a per-kilowatt basis.
The paper estimated this could range from around $810 to $7,500 per kW per year, depending on satellite design. It also noted that historical SpaceX launch data showed a roughly 20 per cent learning rate, suggesting launch costs could drop below $200 per kg by around 2035 if launch frequency increases significantly, with even $300 per kg making large-scale orbital data centres more viable.
Google further argued that based on Starship's public specifications, launch costs could eventually fall to below $60 per kg, or even below $15 per kg with extensive hardware reuse, although achieving such costs would depend on SpaceX and other launch providers successfully deploying highly reusable, cost-effective rockets.
Additionally, Google also suggests the need for new design approaches for individual satellites which would minimise the mass and improve the cost further. It envisions highly integrated satellite designs that combine computing, power, and cooling systems into a single architecture, just like how system-on-chips (SoC) in smartphones combine CPU, GPU, and a cellular modem integrated onto a microchip.
However, achieving the same is easier said than done and would require significant research and technological advances, which would only arrive once the world adopts the idea of an orbital data centre after solving the remaining challenges while staying financially sound.
While tech behemoths are confident about orbital data centres, they exist in the future and hence cannot solve the inherent problems right now. Several industry experts argue that the immediate solutions to today's data centre crisis already exist on Earth. The third part of the series will look at these solutions as well as discuss whether the industry should focus on fixing terrestrial infrastructure before reaching for the stars.
| READ ABOUT OTHER ASPECTS OF ORBITAL DATA CENTRES |
| Data Centre In Space (Part 1): Why The AI Industry Wants To Take The Cloud Into Orbit |
| [You are here] 👉 Data Centre In Space (Part 2): The Engineering Bottlenecks & Economics Behind Orbital Computing |
| Data Centre In Space (Part 3): Why The Real Solution To AI Infra Load May Still Be On Earth |
| Data Centre In Space (Part 4): Why India's TakeMe2Space Is Betting On Orbital Computing |
