Brief

Orbital Data Centers: Beyond the Grid

Space-based computing could emerge as a strategic supplement to terrestrial infrastructure.

著者：Blaine Pellicore, Padraic Brick, and Kristie Tagawa

First published on 7月 17, 2026
min read


概要 Terrestrial power and permitting constraints on data centers are increasing interest in space-based computing. Orbital data centers could represent a modest but strategically important share of global compute capacity by 2040. Launch cost and cadence will determine whether orbital data centers become economically viable. Space-based data centers would bypass the power and permitting constraints slowing terrestrial expansion, but they introduce a different set of cost and engineering trade-offs. The key question is whether those trade-offs become economically favorable as launch costs decline. While the technology is still nascent, Bain & Company analysis suggests that orbital data centers could begin scaling in the early 2030s (see Figure 1). Start-ups such as Starcloud demonstrated in-orbit compute capabilities in 2025, and SpaceX filed for approval of an “orbital data center system” with the US Federal Communications Commission in January 2026. If terrestrial infrastructure constraints continue tightening, even a small share of global compute capacity could become strategically important. Figure 1 Orbital data centers could begin scaling in the early 2030s Source: Bain Orbital Data Center Model Orbital vs. terrestrial data centers The economics of orbital computing hinge on three variables: launch, satellite infrastructure, and operating costs. For this analysis, we assume that compute hardware costs are comparable on a per-kilowatt basis, whether deployed on Earth or in orbit. Today, a constellation of orbital data centers amounting to 100 megawatts would cost roughly 50% more per kilowatt than a comparable terrestrial facility because of launch requirements and the additional infrastructure needed in orbit, including solar arrays, radiators, propulsion, and avionics. But, once deployed, orbital systems could operate with materially lower ongoing energy and cooling costs. If launch and satellite costs decline as expected, the economics could begin to favor orbital deployments by the mid-2030s. Launch. Transporting a data center into space is a significant cost for orbital infrastructure and remains the primary reason space-based computing is not yet economically competitive. Today, placing a 100-megawatt system into orbit would cost more than building and powering a comparable terrestrial facility. That equation changes materially if reusable launch systems reduce cost per kilogram while compute, solar, and cooling systems become lighter and more efficient. Together, those trends could reduce launch costs per unit of compute by as much as 80% by 2035, according to our analysis. Satellite infrastructure (non-compute components). The infrastructure required in orbit functions as the equivalent of a terrestrial facility shell, but at far higher cost today. Solar arrays and radiator systems remain at an early stage in the manufacturing curve, though costs should decline as satellite power density improves and production scales. As next-generation computing systems consume more power per rack, fixed infrastructure can also be spread across more usable compute capacity. Under those conditions, orbital economics could approach parity with terrestrial systems. Operating costs. Operating economics could become one of the largest advantages of orbital computing. Most of the annual operating budget for terrestrial data centers is spent on electricity and cooling. In orbit, solar energy is continuously available, and heat can be dissipated radiatively, reducing ongoing operating costs substantially after deployment. That gap could widen further if terrestrial operators increasingly rely on behind-the-meter generation and grid upgrades to secure power capacity (see Figure 2). Figure 2 Orbital and terrestrial computing economics could converge in the 2030s Note: Assumes five-year compute lifetime Source: Bain Orbital Data Center Model Engineering challenges Space-based computing avoids many terrestrial bottlenecks, including grid interconnection delays, substation construction, and permitting timelines. But those constraints are replaced by a different challenge: reducing the amount of satellite mass required for each kilowatt of compute. Lighter server racks, more efficient solar power systems, and lower-mass cooling infrastructure all improve the economics of each launch. A next-generation GPU rack, for example, needs to weigh about half as much as current designs while delivering greater compute output. Over time, higher-performance chips could allow satellite infrastructure weight to remain relatively stable even as compute capacity per rack increases. Solar power is abundant in orbit, where satellites can access sunlight for much of the day. But the power system itself adds significant launch weight through panels, structural supports, and batteries. The economics improve if these systems become both lighter and more efficient, allowing satellites to generate more usable power per kilogram launched. Cooling is more challenging than intuition suggests because the vacuum of space provides no air to carry heat away. Instead, heat must be dissipated through radiator systems, which add both mass and cost. Advances in materials, thermal efficiency, and chip operating tolerance could reduce the size and weight of those systems over time, strengthening launch economics. Connectivity remains a limiting factor. In the near term, orbital systems will be better suited for inference—running trained AI models to generate outputs—than for large-scale model training. Training requires tightly coordinated GPU communication with extremely low latency, which is difficult to achieve across distributed satellites using current intersatellite links. Inference workloads are typically more tolerant of latency and naturally more distributed, making them better aligned with early orbital deployments. Launch economics The biggest challenge in scaling orbital data centers is launch economics. Three factors matter most: payload capacity, cost per kilogram, and total annual launch cadence. Heavy-lift launch vehicles will be a key element in whether orbital compute can scale. At current satellite mass estimates, deploying 1 gigawatt of compute capacity would require roughly 1,000 Falcon 9 launches per year—far beyond today’s launch cadence. A higher-capacity system such as Starship could reduce that requirement dramatically. SpaceX is targeting payloads of 150–200 metric tons over time for Starship, vs. about 23 metric tons for Falcon 9 today. At current Falcon 9 launch costs of approximately $600 per kilogram, placing a 100-megawatt system into orbit would cost more than the compute hardware itself. For orbital economics to become competitive, launch costs would likely need to fall closer to $50–$100 per kilogram. Achieving that reduction depends on highly reusable launch systems, larger payloads, and continued advances in satellite efficiency. Starship would need to move well beyond today’s demonstrated reuse rates while SpaceX simultaneously expands launch infrastructure, manufacturing capacity, and propellant logistics. Regulatory approvals and site development will also shape how quickly launch capacity can scale. Signposts to watch Orbital compute will become more viable if three conditions converge: Terrestrial infrastructure constraints worsen, satellite systems become lighter and more capable, and launch economics improve. The strongest demand signal would be a sustained slowdown in terrestrial data center expansion. Longer grid interconnection queues, rising electricity costs, and permitting delays would make alternative sources of compute capacity more attractive. A second indicator is the evolving cost for launch. Getting costs down will require reducing the amount of infrastructure needed for each kilowatt of compute, particularly across solar arrays, batteries, radiators, and server systems. Key factors include higher reuse rates, faster turnaround times, greater annual launch cadence, and declining launch costs per kilogram. Together, these metrics determine whether orbital deployments can scale economically. Orbital data centers are unlikely to scale because of any single breakthrough. Their trajectory depends on whether terrestrial constraints tighten at the same time launch economics and satellite efficiency improve. Companies that track those shifts early will be better positioned if orbital infrastructure moves from experimental to commercially relevant.

概要

Terrestrial power and permitting constraints on data centers are increasing interest in space-based computing.
Orbital data centers could represent a modest but strategically important share of global compute capacity by 2040.
Launch cost and cadence will determine whether orbital data centers become economically viable.

Space-based data centers would bypass the power and permitting constraints slowing terrestrial expansion, but they introduce a different set of cost and engineering trade-offs. The key question is whether those trade-offs become economically favorable as launch costs decline.

While the technology is still nascent, Bain & Company analysis suggests that orbital data centers could begin scaling in the early 2030s (see Figure 1). Start-ups such as Starcloud demonstrated in-orbit compute capabilities in 2025, and SpaceX filed for approval of an “orbital data center system” with the US Federal Communications Commission in January 2026. If terrestrial infrastructure constraints continue tightening, even a small share of global compute capacity could become strategically important.

visualization — Orbital data centers could begin scaling in the early 2030s

Orbital vs. terrestrial data centers

The economics of orbital computing hinge on three variables: launch, satellite infrastructure, and operating costs. For this analysis, we assume that compute hardware costs are comparable on a per-kilowatt basis, whether deployed on Earth or in orbit.

Today, a constellation of orbital data centers amounting to 100 megawatts would cost roughly 50% more per kilowatt than a comparable terrestrial facility because of launch requirements and the additional infrastructure needed in orbit, including solar arrays, radiators, propulsion, and avionics. But, once deployed, orbital systems could operate with materially lower ongoing energy and cooling costs. If launch and satellite costs decline as expected, the economics could begin to favor orbital deployments by the mid-2030s.

Launch. Transporting a data center into space is a significant cost for orbital infrastructure and remains the primary reason space-based computing is not yet economically competitive. Today, placing a 100-megawatt system into orbit would cost more than building and powering a comparable terrestrial facility. That equation changes materially if reusable launch systems reduce cost per kilogram while compute, solar, and cooling systems become lighter and more efficient. Together, those trends could reduce launch costs per unit of compute by as much as 80% by 2035, according to our analysis.

Satellite infrastructure (non-compute components). The infrastructure required in orbit functions as the equivalent of a terrestrial facility shell, but at far higher cost today. Solar arrays and radiator systems remain at an early stage in the manufacturing curve, though costs should decline as satellite power density improves and production scales. As next-generation computing systems consume more power per rack, fixed infrastructure can also be spread across more usable compute capacity. Under those conditions, orbital economics could approach parity with terrestrial systems.

Operating costs. Operating economics could become one of the largest advantages of orbital computing. Most of the annual operating budget for terrestrial data centers is spent on electricity and cooling. In orbit, solar energy is continuously available, and heat can be dissipated radiatively, reducing ongoing operating costs substantially after deployment. That gap could widen further if terrestrial operators increasingly rely on behind-the-meter generation and grid upgrades to secure power capacity (see Figure 2).

Engineering challenges

Space-based computing avoids many terrestrial bottlenecks, including grid interconnection delays, substation construction, and permitting timelines. But those constraints are replaced by a different challenge: reducing the amount of satellite mass required for each kilowatt of compute. Lighter server racks, more efficient solar power systems, and lower-mass cooling infrastructure all improve the economics of each launch.

A next-generation GPU rack, for example, needs to weigh about half as much as current designs while delivering greater compute output. Over time, higher-performance chips could allow satellite infrastructure weight to remain relatively stable even as compute capacity per rack increases.

Solar power is abundant in orbit, where satellites can access sunlight for much of the day. But the power system itself adds significant launch weight through panels, structural supports, and batteries. The economics improve if these systems become both lighter and more efficient, allowing satellites to generate more usable power per kilogram launched.

Cooling is more challenging than intuition suggests because the vacuum of space provides no air to carry heat away. Instead, heat must be dissipated through radiator systems, which add both mass and cost. Advances in materials, thermal efficiency, and chip operating tolerance could reduce the size and weight of those systems over time, strengthening launch economics.

Connectivity remains a limiting factor. In the near term, orbital systems will be better suited for inference—running trained AI models to generate outputs—than for large-scale model training. Training requires tightly coordinated GPU communication with extremely low latency, which is difficult to achieve across distributed satellites using current intersatellite links. Inference workloads are typically more tolerant of latency and naturally more distributed, making them better aligned with early orbital deployments.

Launch economics

The biggest challenge in scaling orbital data centers is launch economics. Three factors matter most: payload capacity, cost per kilogram, and total annual launch cadence.

Heavy-lift launch vehicles will be a key element in whether orbital compute can scale. At current satellite mass estimates, deploying 1 gigawatt of compute capacity would require roughly 1,000 Falcon 9 launches per year—far beyond today’s launch cadence. A higher-capacity system such as Starship could reduce that requirement dramatically. SpaceX is targeting payloads of 150–200 metric tons over time for Starship, vs. about 23 metric tons for Falcon 9 today.

At current Falcon 9 launch costs of approximately $600 per kilogram, placing a 100-megawatt system into orbit would cost more than the compute hardware itself. For orbital economics to become competitive, launch costs would likely need to fall closer to $50–$100 per kilogram.

Achieving that reduction depends on highly reusable launch systems, larger payloads, and continued advances in satellite efficiency. Starship would need to move well beyond today’s demonstrated reuse rates while SpaceX simultaneously expands launch infrastructure, manufacturing capacity, and propellant logistics. Regulatory approvals and site development will also shape how quickly launch capacity can scale.

Signposts to watch

Orbital compute will become more viable if three conditions converge: Terrestrial infrastructure constraints worsen, satellite systems become lighter and more capable, and launch economics improve.

The strongest demand signal would be a sustained slowdown in terrestrial data center expansion. Longer grid interconnection queues, rising electricity costs, and permitting delays would make alternative sources of compute capacity more attractive.

A second indicator is the evolving cost for launch. Getting costs down will require reducing the amount of infrastructure needed for each kilowatt of compute, particularly across solar arrays, batteries, radiators, and server systems. Key factors include higher reuse rates, faster turnaround times, greater annual launch cadence, and declining launch costs per kilogram. Together, these metrics determine whether orbital deployments can scale economically.

Orbital data centers are unlikely to scale because of any single breakthrough. Their trajectory depends on whether terrestrial constraints tighten at the same time launch economics and satellite efficiency improve. Companies that track those shifts early will be better positioned if orbital infrastructure moves from experimental to commercially relevant.