Data Centers in Space
A Comprehensive Research Report

Table of Contents

• Executive Summary
• Key Questions Answered
  • Why would anyone put data centers in space?
  • Can commercially available hardware survive in orbit?
  • Is it economically viable today?
  • What is the technology readiness level?
• Core Findings
  Why Put Data Centers in Space?

  Technical Architecture & Thermal Management

  Power Systems in Orbit

  Cooling Mechanisms & Physics

  Radiation Tolerance & Reliability

  Connectivity & Networking

  Economics & Cost Structure

  Market Use Cases & Strategic Positioning

  Key Players & Projects
• Contradictions & Debates
  • 1. Economic Viability Is Deeply Disputed
  • 2. PUE Claims Are Contested in Definition
  • 3. Radiation Tolerance: COTS vs. Rad-Hard
  • 4. Passive vs. Active Cooling Preference
  • 5. Thermal Scaling: Favorable vs. Unfavorable
• Deep Analysis
  • The Scaling Gap: From Watts to Megawatts
  • Thermal Architecture for Orbital Compute
  • The Convection Penalty
  • Radiative View Factor Degradation in Clusters
  • Environmental & Debris Considerations
  • Legal, Regulatory, & Geopolitical Dimensions
  • Alternatives & Competitors
• Implications
  • For Investors & Strategists
  • For Engineers
  • For Policymakers
• Future Outlook
  • Optimistic Scenario
  • Base Case
  • Pessimistic Scenario
• Unknowns & Open Questions
• Evidence Map

Executive Summary

The concept of deploying data centers in orbit has transitioned from speculative fiction to early-stage engineering analysis, driven by the convergence of explosive terrestrial data center power demand (projected to rise 165% by 2030 versus 2023 levels [13]), declining launch costs (with projections of $200/kg by the mid-2030s under SpaceX Starship's reusability trajectory [12]), and advances in radiation-tolerant computing hardware (Google TPU TID tolerance exceeding 2 krad(Si) [12]). However, the concept remains deeply immature, best characterized at Technology Readiness Level 3–4, with first orbital compute demonstrations (Starcloud-1 carrying an NVIDIA H100 GPU) only launching in late 2025 [13].

The central engineering constraint is thermal management. In vacuum, convective cooling vanishes entirely, leaving radiation as the sole passive heat dissipation mechanism [5], [6], [7], [9], [11]. Experimental evidence shows electronics temperatures up to 66% higher in vacuum than at atmospheric pressure for passively cooled heat sinks [11], and silicon chip dependability drops approximately 10% for every 2 °C increase beyond junction temperature thresholds of 80–90 °C [9]. Over 55% of electronics failures are temperature-related [7]. Rejecting megawatt-class heat loads via radiation alone requires radiator areas measured in thousands of square meters, massing tens of thousands of kilograms — a scaling challenge that no existing space system has addressed.

The economics are starkly unfavorable today: a modeled 81-satellite constellation carries an estimated 10-year total cost of ownership (TCO) of $380 million versus $81.2 million for an equivalent terrestrial facility — a 4.7× premium [12]. Starcloud's own economic framing, which compares $175 million in terrestrial electricity costs against "tens of millions" in launch and solar infrastructure [13], appears to omit significant cost categories including radiator mass, structural mass, insurance, ground stations, and replacement cycles.

The field is best understood not as a near-term commercial opportunity but as a long-horizon research and development effort with genuine strategic rationale — sovereignty, grid independence, and potential for power abundance in orbit — that may yield specialized, niche applications before (if ever) achieving general-purpose viability. The terrestrial competition is not standing still: $6.7 trillion in projected global data center investment by 2030 [13], nuclear-powered campuses (960 MW for $650 million [13]), and 20-year power purchase agreements are locking in massive capacity that orbital systems must demonstrably outperform.

Key Questions Answered

Why would anyone put data centers in space?

The motivation rests on several structural pressures affecting terrestrial data centers:

• Power abundance: The Sun delivers approximately 1,360 W/m² at 1 AU [4], [13]. A dawn-dusk sun-synchronous orbit at 650 km provides >95% illumination duty cycle with eclipses under 5 minutes [12], effectively eliminating the intermittency that plagues terrestrial solar. Terrestrial grids face multi-year interconnection delays, and nuclear-powered data center campuses command $650 million acquisitions for 960 MW [13].
• Cooling via radiation: The vacuum of space eliminates convective cooling entirely [2], [4], [6], [7], [9], but the cosmic background at ~2.7 K [8] provides an effectively infinite cold sink for radiative heat rejection — a fundamentally different cooling physics than terrestrial facilities.
• Land and permitting scarcity: Terrestrial data center permitting can take years; the report references $6.7 trillion in projected global investment by 2030 [13], reflecting enormous demand that is constrained by physical and regulatory capacity on Earth.
• Sovereign compute: Orbital platforms could enable data jurisdiction avoidance and neutral-zone compute, enabling nations without domestic hyperscale capacity to access strategic compute autonomy [13].

Can commercially available hardware survive in orbit?

Partially. Google's internal testing indicates TPU HBM subsystems tolerate >2 krad(Si) total ionizing dose, with expected LEO exposure of ~150 rad(Si)/year, yielding a 2.67× safety margin over a 5-year mission [12]. This suggests that at least for Google's Trillium v6e architecture, standard aluminum spacecraft structure provides sufficient shielding without additional radiation hardening. However, this result is architecture-specific: different compute chips, memory technologies, and advanced process nodes (3nm, 2nm) may have substantially different radiation responses. Single-event upsets (SEUs), distinct from total ionizing dose degradation, are not addressed [12].

Is it economically viable today?

No. Even under optimistic assumptions (launch at $200/kg, COTS hardware, 5-year operations), the space-based TCO exceeds terrestrial equivalents by approximately 4.7× [12]. The viability threshold requires launch costs at or below $200/kg [12], a significant reduction from current SpaceX Falcon 9 costs of approximately $2,700/kg [12].

What is the technology readiness level?

TRL 3–4 overall [13]. Individual components (solar arrays, heat pipes, deployable radiators, radiation-tolerant processors) are at higher TRLs (5–9), but no integrated orbital data center system has been demonstrated. Starcloud-1, launched November 2025, carried an NVIDIA H100 GPU described as ~100× more compute than previously operated in orbit [13].

Core Findings

Why Put Data Centers in Space?

Power Demand Crisis on Earth

Terrestrial data center power demand is projected to rise 165% by 2030 versus 2023 levels [13]. The United States alone has 150 GW of data center capacity in the pipeline [13]. Grid interconnection delays stretch multiple years. Nuclear-powered data center campuses command premium valuations: Talen Energy sold a 960 MW nuclear campus to AWS for $650 million [13]; Microsoft signed a 20-year PPA for Three Mile Island restart [13]; Aalo Atomics broke ground on a 10 MW small modular reactor (SMR) at Idaho National Laboratory in August 2025 [13]. Global data center investment is projected at $6.7 trillion by 2030 [13].

In orbit, the Sun delivers approximately 1,360 W/m² at 1 AU [4], [13], with on-orbit solar array efficiencies of 20–25% [13] (triple-junction cells achieving 32% beginning-of-life efficiency [12]). A dawn-dusk sun-synchronous orbit at 650 km provides >95% illumination duty cycle with eclipses under 5 minutes [12], effectively eliminating intermittency. There is no grid to connect, no permitting delay, and no land to acquire.

Cooling Physics in Vacuum

The vacuum of space eliminates convective cooling entirely [2], [4], [6], [7], [9]. The cosmic microwave background at approximately 2.7 K [8] serves as an effectively infinite cold sink, making radiative heat rejection fundamentally available without power input [8]. However, the Stefan-Boltzmann T⁴ relationship [1], [4] means that radiator area requirements grow rapidly with power levels, and space environmental stressors — extreme temperature fluctuations, space dust deposition, vacuum ultraviolet radiation, and atomic oxygen erosion — degrade radiative surface performance over time [8].

Source 8 explicitly notes "fundamental and material-requirement differences between terrestrial and space-based radiative cooling" [8], meaning that terrestrial radiative cooling material research cannot be directly transferred to space applications.

Carbon Accounting Arbitrage (Speculation)

No source in the evidence base directly quantifies the carbon footprint of launching and operating an orbital data center versus a terrestrial one. Starcloud frames orbital solar as eliminating ongoing fuel and grid costs [13], but does not account for launch emissions, manufacturing footprint, or end-of-life disposal. This remains an open question.

Technical Architecture & Thermal Management

Orbital Location Tradeoffs

The sources implicitly address orbital selection through the lens of thermal environment, radiation exposure, and connectivity:

• LEO (Low Earth Orbit, ~400–2,000 km): Periodic eclipses impose thermal cycling stress [7]. Atomic oxygen erodes materials [8]. Micrometeorite/orbital debris risk is highest. Latency to ground is lowest. The modeled satellite operates at 650 km altitude [12].
• GEO (Geostationary Orbit, ~35,786 km): Near-continuous sunlight, coldest effective heat sink view, but higher latency. Not directly analyzed for data center applications in these sources.
• Dawn-dusk sun-synchronous orbit at 650 km: Provides >95% illumination duty cycle with eclipses under 5 minutes [12], optimizing both power availability and thermal stability.

Radiation exposure differs by orbit: ~150 rad(Si)/year in LEO [12], with higher doses in the Van Allen belts (MEO) and lower doses above them (GEO/HEO). Debris density is highest in LEO [12].

Physical Design: Modular Satellite Racks

The Biswas analysis [12] models a specific architecture for Google's Project Suncatcher: four Trillium TPU v6e chips producing 1,200 W of compute heat plus 150 W avionics and ~100 W external heating, for 1,450 W total, managed by:

• Two 2.0 m² deployable aluminum 6061-T6 honeycomb radiator panels (4.0 m² total)
• Eight copper-water vapor chamber heat pipes (effective thermal conductivity: 50,000 W/(m·K))
• Gallium-Indium-Tin (Ga-In-Sn) liquid metal thermal interface material
• Z-93 white thermal control coating (ε=0.91 beginning-of-life, degrading to ε=0.85 end-of-life after 5 years; solar absorptance from 0.15 to 0.25)
• System thermal resistance of 0.300 K/W [12]

The design tolerates single heat pipe failure, with junction temperature rising only to 114.7°C, still within the 125°C limit [12]. Eclipse transients cause <1°C variation due to the radiator's 54 kg thermal mass and 44-minute time constant [12].

Starcloud targets 40 MW modules scaling to gigawatt orbital clusters within Starship payload volumes [13]. The proposed architecture exploits favorable radiation shielding scaling laws: shielding mass scales with surface area (∝ r²) while compute scales with volume (∝ r³), making higher-power platforms proportionally easier to protect [13].

Microgravity Effects on Hardware

Microgravity introduces specific challenges:

• Heat pipe performance degradation: Microgravity reduces the temperature-controlling precision of heat pipes due to vapor bubble migration and stagnation [9]. Two-phase thermal components can experience startup problems in microgravity [9].
• PCM behavior: At least for wax-based phase-change materials, initial experimental evidence from a 2024 CubeSat mission indicates that microgravity does not impact the orientation of the wax on heat sinks [3]. However, this is a limited data point; broader effects on two-phase heat transport remain unaddressed.
• Bubble dynamics: In microgravity, bubbles do not buoyantly rise, potentially causing vapor lock in heat pipes and fluid loops [7]. This is an underexplored risk for space-based data center cooling.

Launch Mass Constraints

A modeled 81-satellite constellation would have satellites at 415 kg each [12], totaling approximately 33,600 kg to LEO. Larger proposed constellations (500+ satellites [12]) would multiply this figure. Thermal control subsystem mass typically ranges from 2% (passive designs) to 10% (active designs) of spacecraft dry mass [4]. For a large orbital data center with massive radiator arrays, thermal subsystem mass could be tens of thousands of kilograms.

Power Systems in Orbit

Solar power is the only viable energy source discussed for orbital data centers. Key parameters:

Starcloud frames orbital power economics as follows: for a 40 MW data center over a five-year GPU lifecycle, terrestrial electricity costs would total approximately $175 million, whereas launch and solar infrastructure for an equivalent orbital system would be "on the order of tens of millions of dollars" with no ongoing fuel or grid costs [13]. However, this comparison appears to significantly underestimate total system cost by omitting radiator mass, structural mass, shielding, ground stations, insurance, replacement hardware, and operational overhead [12].

Energy storage is not addressed in detail by any source. Batteries would be needed for the <5-minute eclipses in the modeled orbit [12], but storage for longer eclipse periods (as in other orbits) is not discussed.

Cooling Mechanisms & Physics

Fundamental Physics

Radiative heat transfer governed by the Stefan-Boltzmann law (Q = σ ε A [T⁴_rad − T⁴_space], where σ = 5.67×10⁻⁸ W/(m²·K⁴)) is the sole method for rejecting waste heat from a satellite in vacuum [1], [4], [6], [7], [9]. Convection is entirely absent [2], [4], [10], [11]. The cold sink of deep space (~2.7 K background [8]) provides a significant thermal gradient, but the T⁴ relationship means radiator area requirements grow rapidly with power levels [1], [4].

The spacecraft heat balance equation is: q_solar + q_albedo + q_planetshine + Q_gen = Q_stored + Q_out,rad [6]. Environmental heat inputs include solar irradiance (1,360 W/m² [4]), Earth's infrared emission (239 ± 28 W/m² [4]), and albedo (0.25–0.35 fraction of solar irradiance [4]).

Available Cooling Technologies

Radiator Area Calculations at Scale

Using the Stefan-Boltzmann law and source parameters, the mass implications of data-center-scale thermal rejection can be estimated:

• Per-satellite: The modeled architecture uses 4.0 m² of aluminum honeycomb radiator at ~3.3 kg/m² [4], rejecting 1,200 W of TPU heat at a radiator heat flux of 362.5 W/m² [12].
• At 1 MW: ~2,759 m² of radiator area at 362.5 W/m², massing approximately 9,100 kg of radiator panel alone (at 3.3 kg/m²) — a rough lower bound excluding structure, heat pipes, and deployment mechanisms.
• At 40 MW (Starcloud target): ~110,000 m² of radiator area [13], massing approximately 363,000 kg of radiator panels — comparable to the mass of the entire ISS (~420,000 kg total). This represents a fundamental scaling challenge that no source addresses with validated engineering analysis.

Note: These calculations are the author's inference from source parameters; no source directly performs them. The estimates assume ideal emissivity (ε ≈ 1), zero environmental heat absorption, and no view-factor losses — all of which would increase the required area in practice [4].

Experimental PCM Evidence

The most rigorous experimental evidence comes from a 2025 peer-reviewed study:

• Electronics in vacuum displayed temperatures 32.8% higher than those in atmosphere due to absent convective heat transfer [10].
• At 0.00025 Pa (high-grade vacuum), temperatures exceeded atmospheric results by up to 66% for a 3D-printed SS316L plate-fin heat sink at 3.5–7 W power levels [11].
• Adding just 6 grams of paraffin wax PCM reduced electronics temperatures by up to 18.0 °C in vacuum, compared to only 12.3 °C in atmospheric pressure [10]. The PCM was more effective in vacuum than in atmosphere.
• PCM doubled electronics operating time under both pressure conditions at high power [10].

A separate 2024 CubeSat mission confirmed that microgravity does not impact wax PCM orientation on heat sinks [3], though quantitative performance data was not published.

Scalability caveat: The PCM experiments tested only 6 grams at power levels relevant to satellite electronics (watts, not kilowatts) [10]. The gap between this experimental scale and data-center-scale cooling requirements is enormous and unaddressed.

Thermal Control Operating Requirements

Standard satellite electronics operating temperature ranges [4]:

• General electronics: −10°C to +50°C
• Command & Data Handling (C&DH) boxes: −20°C to +60°C
• Batteries: 0°C to 15°C (narrowest range, often the thermal design driver)

NASA GSFC GOLD Rules mandate a 5 K margin from predicted temperatures to component limits [4]. A thermal design that fails for even 0.1% of the mission lifetime is considered a failed design [4].

Advanced Thermal Materials

High-thermal-conductance materials offer pathways for improved heat transport [9]:

Radiation Tolerance & Reliability

Radiation Effects on Electronics

The total ionizing dose (TID) environment in LEO is approximately 150 rad(Si)/year [12]. Google's TPU HBM subsystems tolerate >2 krad(Si), yielding a 2.67× safety margin over a 5-year mission [12]. This suggests that at least some COTS hardware can survive LEO with standard aluminum spacecraft shielding.

However, this finding is architecture-specific and may not generalize to the most advanced process nodes (3nm, 2nm) where transistor geometries are more susceptible to single-event effects. The ±20% uncertainty in TPU TDP assumptions [12] and the omission of single-event upset (SEU) analysis reduce confidence in the radiation tolerance claims.

Reliability: The Temperature-Failure Nexus

The reliability data is sobering:

• >55% of electronics failures are temperature-related, per US Air Force data [7].
• Failure rate increases exponentially with operating temperature, per MIL-HNBK2178 [7].
• Silicon chip dependability drops ~10% for every 2 °C increase beyond a junction temperature threshold of 80–90 °C [9].
• A 10°C increase in temperature can lead to a twofold rise in electronic component failure rates [11].
• Each 1°C reduction improves reliability by decreasing failure rates by 4% [11].

This exponential degradation curve means that thermal management is not merely an engineering optimization but an existential reliability requirement. In an orbital context where hardware replacement costs are orders of magnitude higher than on Earth, maintaining junction temperatures well below threshold becomes an economic imperative.

The modeled satellite maintains junction temperature at 111.4°C — within 13.6°C of the 125°C limit [12]. This slim margin, combined with ±20% TDP uncertainty, indicates that the design operates near the edge of the thermal envelope.

Thermal Interface Material Degradation

Thermal interface materials (TIMs) are exposed to high doses of gamma radiation over typical spacecraft design lifetimes of 5–10 years, causing increased thermal contact resistance, cracking, or delamination [9]. This degradation mechanism is particularly concerning for long-duration orbital data center missions where TIM replacement would be costly or impossible. Multiple degradation pathways (VUV, atomic oxygen, gamma radiation) operate on similar timescales [8], [9], potentially causing correlated degradation of thermal management components.

Connectivity & Networking

Connectivity is a significant gap in the evidence base. No source provides detailed analysis of downlink bandwidth constraints, which is critical for the viability of orbital data centers:

• The Biswas analysis [12] assumes optical ground links (citing NASA TBIRD heritage) to minimize thermal impact from RF transmitters, but does not quantify achievable bandwidth or latency for AI inference workloads.
• The BSV report [13] describes SpaceX Starlink v3 as adding AI processing capability to create a "global edge-compute layer" with inter-satellite laser links, but does not analyze round-trip latency or bandwidth of routing data through orbital compute nodes versus terrestrial fiber.
• Latency from LEO (~650 km) to ground would be approximately 2–4 ms one-way, versus ~1–3 ms for major terrestrial data center interconnects, but the cumulative latency through multiple hops is not analyzed.

Critical gap: The viability of orbital inference workloads depends entirely on the data transfer economics between space and Earth, which is unquantified in the available evidence.

Economics & Cost Structure

Launch Cost Trajectory

Total Cost of Ownership

The Biswas analysis [12] provides the most detailed TCO model:

• 81-satellite constellation, 10-year TCO: $380 million [12]
• Equivalent terrestrial facility, 10-year TCO: $81.2 million [12]
• Premium: approximately 4.7× [12]
• 5-year operational costs: $0.3M per satellite [12], implying highly autonomous operations

Starcloud's economic framing compares $175 million in terrestrial electricity costs over a 5-year GPU lifecycle against "tens of millions" in launch and solar infrastructure [13]. However, this comparison appears to omit major cost categories: radiator mass, structural mass, insurance, ground stations, replacement hardware, and operational overhead.

Comparison with Terrestrial Hyperscale

The terrestrial PUE comparison warrants scrutiny: the Biswas analysis uses a terrestrial average of 1.58 [12], but modern hyperscale facilities routinely achieve PUEs of 1.10–1.20, which would eliminate the claimed space PUE advantage.

Market Use Cases & Strategic Positioning

Orbital Strategy Is Bifurcating Into Two Models [13]:

1. Large centralized orbital platforms (exemplified by Starcloud): Concentrated compute in a small number of large satellites or modular clusters, targeting 40 MW modules scaling to gigawatt orbital clusters within Starship payload volumes [13]. This model favors latency-tolerant training workloads and benefits from favorable radiation shielding scaling laws [13].
2. Distributed edge compute via existing constellations (exemplified by SpaceX Starlink v3): Spreading kW-class compute across thousands of mass-produced LEO satellites with inter-satellite laser links [13]. This model favors latency-sensitive inference workloads and proximity to users.

The BSV report argues these are complementary rather than competitive — terrestrial hyperscale for training and bulk inference, orbital platforms for specialized workloads (sovereign compute, space-based sensing), and distributed constellations for low-latency edge inference [13].

High-Compute Tasks

• AI model training: Latency-tolerant, sustained high-power workloads. Starcloud frames this as a primary target [13]. However, the thermal challenges of sustaining high utilization are extreme.
• Earth observation preprocessing: Processing data where it is generated (in orbit) could reduce downlink bandwidth requirements [13].
• Military/defense workloads: Strategic autonomy and resilience against ground-based threats.

Sovereign & Regulatory Use

• Data jurisdiction avoidance: Data processed in orbit may not fall under any single nation's legal framework — a potential advantage or regulatory risk.
• Neutral-zone compute: Nations without domestic hyperscale capacity could access orbital compute [13].
• Strategic autonomy: Independence from terrestrial grid and permitting constraints.

Key Players & Projects

Contradictions & Debates

1. Economic Viability Is Deeply Disputed

This is the most significant contradiction across sources:

• Biswas models a 10-year TCO of $380M for an 81-satellite constellation versus $81.2M for terrestrial equivalents (4.7× premium), concluding the standalone business case is "not viable" [12].
• Starcloud compares $175M in terrestrial electricity costs alone against "tens of millions" in launch/solar costs [13], suggesting orbital is cheaper — but this comparison omits most non-power cost categories.
• The BSV report acknowledges that terrestrial constraints (grid interconnection delays, $650M nuclear campus acquisitions, $6.7 trillion in projected investment) create structural demand for alternatives [13], but does not demonstrate that orbital is the most cost-effective alternative versus undersea or Arctic facilities.

Assessment: The Biswas analysis is more rigorous but published on LinkedIn without peer review. The Starcloud/BSV framing is investor-oriented with promotional bias risk. The truth likely lies between these positions but closer to Biswas's skepticism.

2. PUE Claims Are Contested in Definition

The Biswas analysis claims a space-based PUE of 1.42 versus a terrestrial average of 1.58 [12], and separately references a "space PUE of 1.17 from optimized accounting." Modern hyperscale facilities achieve PUEs of 1.10–1.20, which would negate the claimed space advantage. The terrestrial baseline of 1.58 may not reflect current best practices.

3. Radiation Tolerance: COTS vs. Rad-Hard

Google's TPU radiation test results (>2 krad(Si) tolerance, ~150 rad(Si)/year LEO dose, 2.67× safety margin) are presented as evidence that COTS hardware can survive LEO [12]. However, this confidence is uncertain: the ±20% uncertainty in TPU TDP assumptions [12], the omission of single-event effects, and the architecture-specific nature of the results limit generalizability.

4. Passive vs. Active Cooling Preference

Sources for small spacecraft emphasize passive methods as preferred due to power, mass, and budget constraints [5], [7]. Sources detailing advanced systems note that active thermal control (pumped loops, variable-emissivity radiators) offers >200% thermal performance improvement [6], implying passive methods alone are insufficient. Source 9 notes that thermal control is already "one of the key bottlenecks that restrict the level of spacecraft design" [9]. For a data center, active cooling would almost certainly be required, making the power budget for thermal management a critical design driver.

5. Thermal Scaling: Favorable vs. Unfavorable

Starcloud argues that larger satellites benefit from favorable shielding scaling laws: shielding mass scales with surface area (∝ r²) while compute scales with volume (∝ r³) [13]. However, the radiative cooling relationship works in the opposite direction: radiator area scales with heat rejection (approximately proportional to compute power), while the structural mass of those radiators accumulates linearly. Whether the favorable shielding argument or the unfavorable radiator area argument dominates at scale is unresolved.

Deep Analysis

The Scaling Gap: From Watts to Megawatts

The most critical gap illuminated by the entire evidence base is the scaling problem. Current thermal control technologies are designed for spacecraft consuming watts to perhaps hundreds of watts:

The jump from the modeled satellite [12] to Starcloud's target [13] is approximately 33,000×. The jump from the experimental heat sink study [11] to Starcloud is roughly 6 million-fold. No validated engineering analysis bridges these scales.

The terrestrial data center industry provides a partial analog: rack power densities have risen from 5–10 kW to 30–50 kW for AI training, with frontier systems exceeding 100 kW per rack [13]. Air cooling becomes insufficient above 10–15 kW per rack [13], driving adoption of immersion cooling, direct-to-chip liquid loops, and rear-door heat exchangers. This body of engineering knowledge exists for terrestrial applications; no equivalent knowledge base exists for orbital compute.

Thermal Architecture for Orbital Compute

Combining evidence across all sources, the thermal path from chip junction to radiative surface would require [9], [12]:

1. Low-resistance thermal interface materials — degraded by gamma radiation over 5–10 years [9]; modeled Ga-In-Sn liquid metal TIM achieves R_th,system = 0.300 K/W [12]
2. High-conductance heat spreaders — vapor chambers (50,000 W/(m·K) effective [12]) or APG (1,700 W·m⁻¹·K⁻¹ in-plane [9])
3. Heat transport — heat pipes (effective thermal conductivity several thousand times higher than copper [7]) or pumped fluid loops (subject to microgravity degradation [9])
4. Radiative rejection — deployable radiator panels with Z-93 or equivalent coatings (ε=0.91 BOL, degrading to 0.85 EOL [12]), subject to space environmental degradation [8]

Each stage introduces failure modes, mass penalties, and performance uncertainties. The reliability chain is only as strong as its weakest link.

The Convection Penalty

The 32.8–66% temperature increase in vacuum versus atmosphere [10], [11] is a fundamental physical fact that quantifies the cost of operating in space. On Earth, even passively cooled electronics benefit from natural convection. In orbit, this free cooling pathway is entirely absent. This means:

• Higher baseline operating temperatures for the same power density
• Earlier onset of the 10%-per-2°C reliability degradation curve [9]
• Greater dependence on active thermal management
• Larger radiator areas than an equivalent-air-cooled terrestrial facility
• An inherent power density disadvantage that undermines one of the key arguments for orbital compute

Radiative View Factor Degradation in Clusters

When multiple compute satellites operate in close proximity — as would be required for Starcloud's gigawatt-scale orbital clusters [13] — adjacent radiators would partially view each other rather than cold space, degrading radiative efficiency. This "view factor degradation" is a known problem in spacecraft thermal engineering that is not addressed in any source [12], [13]. At the extreme, tightly clustered orbital data center modules could partially self-heat, creating a thermal management death spiral that limits the achievable compute density per unit volume of orbit.

Environmental & Debris Considerations

No source comprehensively addresses the environmental impact of orbital data centers:

• Launch emissions: Not quantified by any source. An 81-satellite constellation at 415 kg each adds approximately 33,600 kg to LEO [12]. Larger constellations multiply this figure.
• Space debris: The assumption of <5% radiator area loss from micrometeorite/orbital debris impacts [12] and 25-year deorbit compliance [12] does not address the aggregate impact of large constellations on the orbital environment.
• End-of-life disposal: The 25-year deorbit plan mentioned [12] does not include disposal costs in the TCO analysis.
• Lifecycle carbon footprint: Not compared between orbital and terrestrial options by any source.

Legal, Regulatory, & Geopolitical Dimensions

The evidence base provides minimal direct coverage of legal and regulatory issues, but the following can be inferred:

• Outer Space Treaty: Data processed in orbit would likely fall under the jurisdiction of the launching state [12], but this is not directly analyzed.
• Data jurisdiction: Orbital compute platforms could enable data jurisdiction arbitrage — processing data beyond any single nation's legal reach. This is a potential advantage for some users and a regulatory challenge for others.
• Spectrum allocation: Not addressed. Orbital data centers would require significant communication bandwidth for data transfer.
• Liability for debris: Under the Outer Space Treaty and Liability Convention, the launching state is liable for damage caused by its space objects. Large orbital data center constellations would increase debris risk.
• Militarization: Not directly addressed, but the BSV report [13] notes defense applications and the potential for orbital compute to enable strategic autonomy.

Alternatives & Competitors

The BSV report [13] frames the competitive landscape:

• Terrestrial nuclear-powered data centers: $650M for 960 MW nuclear campus [13]; 20-year PPAs for nuclear restart [13]; SMRs at Idaho National Laboratory [13]. These represent the most direct terrestrial competitor.
• Distributed edge compute: The Starlink v3 model embeds inference capability in existing communication satellites [13], competing with dedicated orbital data centers.
• Undersea data centers: Not discussed in the sources but represent a terrestrial alternative for cooling and land constraints.
• Arctic/Antarctic facilities: Not discussed.
• Fusion-powered terrestrial centers: Not discussed.

The BSV report argues that terrestrial alternatives are complementary to rather than competitive with orbital compute, but this framing has not been validated by comparative cost analysis.

Implications

For Investors & Strategists

• The 4.7× TCO premium [12] means orbital data centers are not viable as drop-in replacements for terrestrial facilities under current economics. Investment theses must be predicated on (a) continued launch cost declines toward $200/kg [12], (b) premium pricing for specialized use cases (sovereign compute, defense), or (c) structural advantages not captured in simple TCO comparisons (permitting speed, grid independence).
• Starcloud's November 2025 H100 launch [13] is a meaningful demonstration milestone but should not be conflated with commercial viability. Starcloud-2 (October 2026) introducing AWS Outposts hardware suggests a partnership strategy that could de-risk the business model through ecosystem integration [13].
• The terrestrial competition is massive and accelerating: $6.7 trillion in projected investment by 2030 [13], nuclear-powered campuses, and 20-year PPAs.

For Engineers

• Thermal management is the primary engineering constraint. The 66% temperature increase in vacuum [11] and the 13.6°C margin to TPU junction temperature limits [12] indicate that thermal design operates near the edge of feasibility. Radiator area, mass, and reliability dominate the design space.
• Radiation tolerance findings for Google TPUs [12] are encouraging but architecture-specific. The 2.67× safety margin is modest for a 5-year unattended mission.
• The 55% failure attribution to temperature [7] and the exponential temperature-failure relationship [7], [9], [11] make thermal management reliability the single most important determinant of orbital data center uptime.
• Thermal interface material degradation over 5–10 years [9] creates a maintenance cycle that is impractical without robotic servicing capability.

For Policymakers

• Orbital compute platforms could enable data jurisdiction arbitrage and neutral-zone processing, raising novel legal questions under the Outer Space Treaty.
• Space debris policy will become increasingly relevant as orbital compute scales.
• The geopolitical implications of orbital compute sovereignty — which nations can deploy and control orbital data processing — are significant but entirely unaddressed in the evidence base.

Future Outlook

Optimistic Scenario

Launch costs reach $200/kg by 2030–2035 under aggressive Starship deployment [12]. Advanced thermal management solutions (carbon nanotube conductors at 6,000 W·m⁻¹·K⁻¹ [9], variable-emissivity radiators [9], next-generation PCM systems) scale from demonstrated kW levels to multi-MW modules. Starcloud demonstrates 40 MW modules within Starship payload volumes [13]. Specialized use cases — sovereign compute for nations without domestic hyperscale capacity, space-domain awareness processing, latency-tolerant AI training — generate premium revenue streams. Orbital data centers capture 1–3% of global compute capacity by 2040. SpaceX Starlink v3 creates a complementary distributed edge compute layer for low-latency inference [13].

Key assumption: Launch cost trajectory holds; thermal scaling breakthroughs occur; regulatory frameworks develop.

Base Case

Launch costs decline to $500–1,000/kg by 2035, short of the viability threshold [12]. Starcloud and similar ventures demonstrate technology at small scale (single-digit MW) but struggle to close the 4.7× TCO gap [12]. Terrestrial alternatives (nuclear SMRs, grid-scale batteries, improved PUE) absorb most demand growth. Orbital compute remains niche — limited to demonstration missions, defense applications, and specialized sovereign compute where cost is secondary to strategic value. The Starlink edge compute model succeeds modestly by embedding inference capability in existing communication satellites [13] rather than building dedicated orbital data centers.

Key assumption: Launch costs decline but not fast enough; terrestrial alternatives continue improving.

Pessimistic Scenario

Launch costs plateau above $1,500/kg due to Starship development delays or market failures. Radiation-induced hardware degradation proves more severe than Google's TPU tests suggest, particularly for newer process nodes. Thermal scaling proves intractable at MW+ levels — radiator mass and area dominate satellite design, squeezing out compute payload. A high-profile orbital compute satellite failure (thermal runaway, debris impact, or radiation event) erodes investor confidence. Regulatory frameworks remain undefined, creating legal uncertainty. Terrestrial alternatives (fusion power, advanced cooling, underwater facilities) capture all incremental demand. The concept remains perpetually "5–10 years away."

Key assumption: Multiple technical and economic barriers compound; no breakthrough occurs.

Unknowns & Open Questions

1. Downlink bandwidth and latency: No source quantifies achievable data transfer rates from an orbital data center to terrestrial users, or round-trip latency versus fiber for inference workloads. This is arguably the most critical unanswered question.
2. Thermal scaling to MW/GW: Experimental data covers watts [10,11], modeling covers kilowatts [12], and business plans target megawatts to gigawatts [13]. No validated engineering analysis bridges these scales.
3. Microgravity effects on two-phase cooling at scale: Only a single CubeSat PCM experiment [3] and one reference to a numerical study [7] exist. Large-scale two-phase heat transport in microgravity is uncharacterized.
4. Radiation effects on advanced process nodes: Google's TPU results [12] are encouraging but may not generalize to 3nm, 2nm, or beyond.
5. Cost comparison of space-based vs. terrestrial cooling per kW of compute: No source provides this critical metric.
6. View factor degradation in clustered configurations: Multiple satellites operating in close proximity would degrade each other's radiative efficiency. Not addressed.
7. Long-duration (10+ year) degradation of radiator surfaces: VUV, atomic oxygen, and micrometeorite damage compound over time [8]. No integrated degradation model is provided.
8. Robotic servicing feasibility: Can orbital compute platforms be serviced, upgraded, or repaired robotically? The $0.3M per-satellite 5-year operations budget [12] implies fully autonomous operations, but technical feasibility is not demonstrated.
9. Regulatory and legal framework: Who has jurisdiction over data processed in orbit? How is spectrum allocated? How does the Outer Space Treaty apply?
10. Lifecycle carbon footprint comparison: Is orbital compute actually lower-carbon when launch emissions, manufacturing, and disposal are included?
11. Terrestrial competition trajectory: Nuclear SMRs, fusion, and advanced cooling may dramatically alter the terrestrial cost baseline by 2035.
12. Variable-emissivity radiator maturity: Identified as a key research direction [9], but no data on current performance or TRL is provided.
13. Carbon nanotube manufacturability: The theoretical 6,000 W·m⁻¹·K⁻¹ conductivity [9] is meaningless without scalable manufacturing.

Evidence Map

References

1. ↩ Advanced Cooling Systems for Space - https://numberanalytics.com/blog/advanced-cooling-systems-for-space
2. ↩ Cooling techniques for satellite systems - https://thermal-engineering.org/cooling-techniques-for-satellite-systems
3. ↩ Space electronics cooling experiment tests wax-based heat sinks on orbiting satellite - https://mechse.illinois.edu/news/76332
4. ↩ ENAE 691 Satellite Design: Thermal Control - https://ntrs.nasa.gov/api/citations/20230001953/downloads/ENAE 691 Spring23 Thermal Cottingham.pdf
5. ↩ Satellite Thermal Control System Design Tutorial - https://east-space.com/satellite-thermal-control-system-design-tutorial
6. ↩ State-of-the-Art of Small Spacecraft Technology — Chapter 7: Thermal Control - https://nasa.gov/smallsat-institute/sst-soa/thermal-control
7. ↩ Review of Electronic Cooling and Thermal Management in Space and Aerospace Applications - https://mdpi.com/2673-4591/89/1/42
8. ↩ Radiative Cooling Materials for Spacecraft Thermal Control: A Review (Abstract and Metadata) - https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202506795
9. ↩ Review on Thermal Management Technologies for Electronics in Spacecraft Environment - https://sciencedirect.com/science/article/pii/S277268352400013X
10. ↩ Experimentally investigating phase change material behaviour in satellite electronics thermal control under vacuum and atmospheric pressure - https://sciencedirect.com/science/article/pii/S0017931024012134
11. ↩ Investigating the performance of a heat sink for satellite avionics thermal management: From ground-level testing to space-like conditions - https://sciencedirect.com/science/article/pii/S0017931025004788
12. ↩ Space-Based Data Center Infrastructure: A Multi-Physics Thermal Analysis for AI Computing in Low Earth Orbit - https://linkedin.com/pulse/space-based-data-center-infrastructure-multi-physics-biswas-phd-xfipc
13. ↩ BSV Insights 0002: Kilowatts to Compute — The Convergence of Data Centers and Power on Earth and in Orbit - https://balerionspace.substack.com/p/bsv-insights-0002-kilowatts-to-compute