
~5×
Energy per Installed kW
Near-continuous sun in a dawn-dusk orbit (95–100% capacity factor against about 20% on the ground), but space arrays cost €200–350/W against €1–2/W
Setcoin Intelligence · Space Infrastructure · Feasibility Study · 7 Interactive Models
Can AI compute move to orbit, and what would a kilowatt-hour there really cost? Twenty-two chapters on launch economics, orbital power, cooling, radiation, carbon, regulation and ten companies, with seven interactive models and Setcoin's recalculation of the figures that do not add up.
Open to everyone: the first 3 chapters in full, with exhibits and models; a free account adds their Excel models. Space subscription: all 22 chapters, every model and Excel file, 7-day free trial, then €199 a month. Already a reader? Sign in.
The study in 60 seconds
Radiators, launch mass, radiation and latency, and what a pilot and a constellation cost, in one minute.
01 · The answer
The in-orbit data centers market is projected to reach $1.77 billion by 2029 and explode to $39.09 billion by 2035—a staggering 67.4% CAGR—driven by AI's insatiable energy demands outpacing terrestrial infrastructure capacity. Data centers already consume 17 GW in the US, projected to reach 35 GW by 2030, with AI workloads potentially demanding 60-100 GW.
This feasibility study provides institutional investors with the analytical depth required for early-stage positioning: technical architecture assessments for each major player, comparative cost modeling between orbital and terrestrial deployments, launch economics sensitivity analysis, and thermal management engineering constraints that determine commercial viability.
Key insight: Starcloud's November 2025 launch of an H100 GPU satellite validated in-orbit computing. Google's Project Suncatcher aims for 80+ satellite constellation by early 2030s. The race is no longer theoretical—first-mover advantages are being established now.
Setcoin's reading: technical feasibility is high-confidence (80–90%); the commercial case is the conditional one (65–75%). At €250/W the study's own output figures give about €2.2/kWh of generation capital, not the €0.108/kWh it states, and an orbital node only emits less over its life than a 300 g/kWh grid above about 16 kW. The interactive models let you test both.
19 chapters + 3 appendices · 316 min · updated 2026-02-24
Try the model
Change the array cost, launch price, degradation and life, and compare an orbital solar kilowatt-hour with a datacentre power purchase agreement on Earth. The defaults are the study's figures; your changes stay in your browser.
Interactive model
What does a kilowatt-hour from an orbital array really cost, and what price per watt would match a datacentre PPA?
| Item | Value |
|---|---|
| Lifetime output per kW, space | 110,719 kWh |
| Lifetime output per kW, ground | 26,280 kWh |
| Capital per kW, space | €250,000 |
| Space cost per kWh | €2.26 |
| Ground cost per kWh | €0.076 |
| Space €/W for PPA parity | €8.86 |
What moves “Space solar, cost per kWh” most: each input 20% lower (left of a pair) and 20% higher
This is one of 7 interactive models in the report. The first 3 chapters of the report are open to everyone; a Space subscription, with a 7-day free trial, opens every model and its Excel file.
02 · Orbital Advantages
Each orbital advantage is real, and each comes with a cost the study quantifies

~5×
Near-continuous sun in a dawn-dusk orbit (95–100% capacity factor against about 20% on the ground), but space arrays cost €200–350/W against €1–2/W

No water
No water or chillers, but a 2 MW facility needs about 4,000 m² of radiators, adding 20–40 tons of launch mass

Isolated
Beyond the reach of floods, fire and physical intrusion; exposed instead to radiation, debris and jamming

>16 kW
Launch is two-thirds of lifecycle emissions; below about 16 kW per node, orbit emits more than a 300 g/kWh grid (Setcoin recalculation)
03 · Coverage Areas
Deep-dive technical feasibility assessments, competitive intelligence, and financial projections across the orbital computing ecosystem

High-compute, low-I/O batch workloads ideal for orbital deployment. Molecular folding simulations, Monte Carlo financial modeling, and AI weight training where 500ms latency penalty is irrelevant for 48-hour jobs.

Earth observation satellites generate 10 GB/second using synthetic-aperture radar. In-orbit processing eliminates downlink bottlenecks, enabling real-time wildfire detection and distress-signal response.

Data sovereignty requirements for jurisdictionally complex scenarios. Secure global data storage independent of terrestrial infrastructure, immune to geopolitical access risks.

Redundant infrastructure immune to terrestrial risks—earthquakes, floods, grid failures, and regional conflicts. High-availability computing with guaranteed uptime independent of Earth-based events.

AWS Ground Station, Azure Orbital, and Google Cloud partnerships create seamless terrestrial-orbital hybrid architectures. Global coverage without ground station latency.

Lonestar's $120M deal for lunar-orbiting data storage. Earth-Moon L1 positioning for deep space mission support and ultra-secure archival storage 60,000 km from the Moon.
04 · Hidden Value
Insights that challenge consensus narratives and reveal asymmetric risk-reward profiles
Local permitting opposition blocks new terrestrial facilities in US and Europe. Orbital deployment sidesteps land use conflicts, NIMBY resistance, and grid connection bottlenecks that delay ground-based expansion 3-5 years.
No Grid Queue
GPU obsolescence cycles (5-6 years) align with satellite operational lifespans. Unlike terrestrial facilities requiring continuous upgrades, orbital deployments can refresh entire constellations with next-gen hardware on natural replacement schedules.
Synchronized Upgrade Cycles
05 · Risk Analysis
Key constraints that could delay timelines, increase costs, or invalidate business models
LEO radiation degrades electronics. Radiation-hardened hardware costs 10-100x consumer equivalents. HPE's Spaceborne Computer demonstrated challenges—error-correcting software essential but adds latency.
LEO orbital latency: 60-190ms round-trip vs. 10-50ms terrestrial CDN. Eliminates latency-sensitive applications. Only batch processing, AI training, and satellite-to-satellite workloads viable.
Saarland University research: orbital data centers may create 10x greater emissions than terrestrial when including rocket launches and reentry pollution. Ozone layer impact from burning spacecraft components.
No on-site technicians. Hardware failures require robotic servicing or complete replacement. Space debris collision risk. Counterspace technologies (jamming systems) pose national security concerns.
06 · Technology Insights
First NVIDIA H100 deployed in space. 100x more powerful than any previous orbital GPU. Running Google Gemma LLM in orbit proves large language models viable for space-based inference. Blackwell integration planned for future launches.
80+ satellite constellation with TPU accelerators and free-space optical inter-satellite links. Two prototype satellites launching early 2027. Testing distributed ML tasks in orbit—if successful, operational nodes by early 2030s.
12 AI satellites launched May 2025 as first phase. 744 TOPS per satellite, 8B-parameter AI models, 100 Gbps laser links. China racing to establish orbital computing dominance with centralized gigawatt-scale megacenter by 2035.
Thales Alenia Space leading European consortium. First in-orbit demonstration 2028. Robotic assembly using EROSS IOD technology. Target: 1 GW computing power by 2050. €82M private investment since 2020.
Starcloud's stated plan: a 5-gigawatt orbital data center with 4km x 4km solar and cooling arrays in sun-synchronous orbit, scaling to megascale by 2035. Its claimed 10x lifecycle carbon saving depends on power per node; the study's carbon model lets you test it.
$120M Sidus contract for 6 data-storage satellites. 15-petabyte L1 positioning launching 2027. Each subsequent satellite doubles capacity. Delay-Tolerant Network (DTN) tested successfully for Solar System Internet architecture.
07 · Launch cost
Launch cost per kilogram to low Earth orbit by vehicle, as used in the study's financial model.
US$ per kg
Study chapter 7 (technology providers); Starship figure is a company target.
Try the model
From a single pilot to a forty-satellite constellation: development, satellites, launch and ground, with the study's own figures as defaults.
Interactive model
What does a pilot, a small constellation or a large one cost to build and launch, and where does the money go?
| Line | Your case | Share | Report reference |
|---|---|---|---|
| Development | €75M | 49.3% | €60–90M pilot · €80–120M constellation |
| Manufacturing | €40M | 26.3% | €35–45M pilot · €350–650M for 10–20 |
| Launch | €12M | 7.9% | €7–12M pilot · €150–350M for 10–20 |
| Ground segment | €20M | 13.1% | €15–25M pilot · €120–180M for 10–20 |
| Insurance | €5.2M | 3.4% | 8–15% of insured value |
| Last spacecraft, manufacturing | €40M | Learning curve 85–90% | |
| Total | €152M | 100% | €50–150M pilot · €1.0–2.0B for 10–20 |
What moves “Total capex” most: each input 20% lower (left of a pair) and 20% higher
Try the model
Size the solar array for a given amount of computing, with degradation over the satellite's life. Try a 1 PFLOPS node or the study's 1 kW reference case.
Interactive model
How much solar array, battery and radiator does a given amount of orbital compute need?
| Item | Value |
|---|---|
| Compute power | 1.3 kW |
| Total load | 2.0 kW |
| Array power at launch | 2.7 kW |
| Array area | 6.6 m² |
| Battery capacity | 2.43 kWh |
| Battery mass (175 Wh/kg) | 14 kg |
| Radiator area (3 m²/kW) | 6.0 m² |
| Reference satellite (Appendix A) | 10–15 kW array, 25–35 m², 8–12 kWh battery |
What moves “Power at end of life” most: each input 20% lower (left of a pair) and 20% higher
08 · Coverage Universe
Deep-dive analysis of the leading orbital computing developers with comparative technology and business model assessments
NVIDIA-Backed LEO DC
H100 in Orbit Nov 2025
Project Suncatcher
Demo 2027
Orbital Data Center Nodes
Launch 2025
Cislunar Storage
$120M Sidus Deal
EU ASCEND Program
Demo 2028
2,800-Satellite Fleet
12 Launched 2025
Azure Space / Orbital
Partner Strategy
Galactic Brain
Q1 2027 Target
Leo Satellite Initiative
Development
On-Orbit Compute Lease
April 2025
Check it yourself
The study's findings move with launches, contracts and filings. These Setcoin tools show the current public record behind them.
8 large constellations against what they are licensed or filed for: satellites launched, re-entered and net pace over twelve months, and the date each milestone is reached at that pace.
See how fast the large constellations are filling low orbit, the same shells an orbital datacenter would use.
Open the monitor →Type an operator’s name and see what the public satellite catalogues record: satellites launched, in orbit, operating and re-entered. The full check adds US contracts, SBIR awards, UK and NOAA licences and list screening.
Check a developer’s claimed demonstration satellites against the catalogue record.
Check an operator →607 new US federal space awards and European tenders from the last twelve months, by part or service bought, buyer, country and stage, with the tenders still open.
See which agencies are buying satellites, platforms and space R&D.
Open the radar →More than a thousand cited R&D awards and prime contracts in space and aerospace, by organisation and readiness gate, from public procurement and grant data.
Open the space record →244 tracked capital profiles on one ladder, from seed to construction loan, by stage, ticket size and sector, including space.
Open the explorer →Places where emerging-sector sites sit inside active power pipelines, joint tender pools and shared counterparties across sectors.
Open Crosslink →Setcoin Crosslink
The same technologies, supply chains and capital show up in other sectors on the Atlas. Leads to explore, not findings.
Research partners, licensors or licensees for the same technology outside this sector
Research partners, licensors or licensees for the same technology outside this sector
Research partners, licensors or licensees for the same technology outside this sector
Research partners, licensors or licensees for the same technology outside this sector
185 grants, trials or patents in robotics, biotech work on it
61 grants, trials or patents in fusion work on it
Questions
The in-orbit data centre market is projected at $1.77B by 2029 and $39.09B by 2035 — a 67.4% CAGR — as AI energy demand outpaces terrestrial capacity (US data centres at 17 GW today, 35 GW by 2030, with AI workloads potentially demanding 60-100 GW).
Near-continuous solar power (each installed kilowatt produces about five times the energy of a ground panel), radiative cooling with no water (terrestrial data centres consume 1.2 trillion litres a year) and physical isolation. Each has a cost: space arrays cost €200–350/W against €1–2/W, a 2 MW facility needs about 4,000 m² of radiators, and launch is two-thirds of lifecycle emissions.
Not today. At €250/W, a 1 kW space array produces about 115,000 kWh over 15 years, about €2.2/kWh of generation capital against €0.06–0.10/kWh for large terrestrial renewable PPAs. The study states €0.108/kWh, which its own figures do not support; that price would need an array cost near €12/W. Setcoin recalculation.
Only above a certain size. Launch accounts for about 65% of lifecycle emissions (400–800 t CO₂ per satellite). At equal load, an orbital node emits less over 15 years than a 300 g/kWh grid only above about 16 kW per node; at the study's 10 kW example it emits more. The study's 90% reduction compares a 10 kW node with a ~150 kW terrestrial load.
Technically yes, with high confidence (80–90%): an H100 GPU has operated in low Earth orbit since November 2025. Regulatory feasibility is also high. Commercial feasibility is the weakest (65–75%) and depends on launch costs near $200/kg, cheaper space-grade solar arrays and workloads that tolerate 60–190 ms latency.
The study costs a single-satellite pilot at €50–150M and a 10–20 satellite constellation at €1.0–2.0B, with launch and ground segments taking up to 45% of capex. A 100–500 kW prototype phase (2025–2027) is estimated at $150–250M.
High-compute, low-I/O batch workloads: AI model training, satellite data processing (10 GB/s SAR downlink avoidance), sovereign cloud, disaster recovery, cloud edge extension with AWS/Azure/Google, and cislunar archival storage. LEO latency of 60-190 ms rules out latency-sensitive applications.
Launch cost. Costs fell 95%+ from $54,500/kg (Shuttle) to $2,720/kg (Falcon 9), but Google's Suncatcher team estimates orbital datacenters need under $200/kg by 2035 — a further 7x drop that depends on Starship ($93-250/kg projected). Thermal management adds 20-40 tons of radiator mass per 2 MW.
Starcloud (H100 in orbit November 2025), Google Project Suncatcher (2027 demo), Axiom Space, Lonestar Data, Thales Alenia (EU ASCEND, 2028 demo), ADA Space (2,800-satellite fleet), Microsoft Azure Space, Aetherflux, Amazon/AWS and Kepler Communications.
The first three chapters are open to everyone in full, with no account; the rest can be previewed. A space sector subscription, with a free trial, opens all 22 chapters, the seven interactive models and their Excel files.
This request form is for informational purposes and does not constitute an offer to sell or solicitation of an offer to buy any securities. The research materials contain forward-looking statements and projections based on current market conditions subject to change. Access to institutional research is granted at the sole discretion of Setcoin Group. All information provided is kept strictly confidential in accordance with our privacy policy.
1. Read the first 3 chapters in full, no account needed. 2. Create a free account (two minutes, no card) to download their Excel models. 3. Subscribe to Space for every chapter, model and Excel download: 7-day free trial, then €199 a month or €1,490 a year.