Orbital Data Center Economics

Launch Cost Trajectory

Historical and projected $/kg to LEO: from Space Shuttle ($54,500/kg) through Falcon 9 ($2,940/kg) to Google's Starship estimates ($60/kg at 10x reuse, $15/kg at 100x). Google's $200/kg threshold marked.

Active/OperationalProjected/TestingIn DevelopmentRetired

Space vs Terrestrial Comparison

Orbital and ground data centers compared metric by metric, with the advantage flagged for each.

Orbital (Space)Terrestrial (Ground)

PUE (Power Usage Effectiveness)

Comparable to state-of-the-art hyperscale (Starcloud estimate)
1.10 (Google fleet average, 2023)

Starcloud white paper, Google 2024 Environmental Report

Cooling Operating Cost

None expected (radiative cooling to space)Advantage
Liquid or air cooling

SpaceX prospectus (Jun 12, 2026)

Water Consumption

ZeroAdvantage
17B gallons/year (US DCs, 2023)

Lawrence Berkeley Lab

Solar Irradiance

1,361 W/m²Advantage
~200-300 W/m² effective

NASA GSFC, Wikipedia

Solar Capacity Factor

Nearly continuous sunshine (dawn-dusk sun-synchronous LEO)Advantage
24.0% (US utility-scale PV, 2025)

Google preprint (arXiv 2511.19468v2), EIA Electric Power Monthly Table 6.07.B

Google Solar Advantage

8x annual energy productionAdvantage
Baseline

Google Research Blog

Land Use

Zero land footprintAdvantage
Hundreds of acres per GW-scale DC

Industry consensus

Grid Connection

Not needed (self-powered)Advantage
Major bottleneck (years to connect)

Industry analysis

Maintenance Access

Not easily repaired or upgraded once deployed
Full 24/7 accessAdvantage

SpaceX prospectus (Jun 12, 2026)

Hardware Replacement

Failures can mean permanent capacity loss or replacing the infrastructure
Standard IT refresh cyclesAdvantage

SpaceX prospectus (Jun 12, 2026)

Latency to End Users

20-40 ms (LEO round-trip)
<1 ms (on-premises fiber)Advantage

Industry standard

Scalability

Limited by launch cadence: about 180 Starship launches a year to sustain a ~20% learning rate
Limited by power/land/permits

Google preprint (arXiv 2511.19468v2, Jun 17, 2026)

Environmental Impact

Zero emissions (solar), debris riskAdvantage
CO2 emissions, water use, land use

ESA, IEA

Terrestrial Data Center Market Context

$61B invested in DC construction (2025), 1,189 hyperscale DCs globally, 415 TWh energy consumption (2024), $424.3B hyperscaler cash capex (2025, five companies).

DC Construction Market Size
$261.3B
Year:
2025
Trend:
Projected to reach $662.7B by 2033 at a 12.7% CAGR (2026 to 2033)
Global DC CapEx Forecast
$1.7 trillion
Year:
By 2030
Trend:
Global data center capex expected to approach $1 trillion in 2026
Hyperscale DCs Worldwide
1,189
Year:
Q1 2025
Trend:
Hyperscale operators hold 44% of worldwide data center capacity
Global DC Energy Consumption
415 TWh
Year:
2024
Trend:
1.5% of global electricity; grew 12% per year over the last five years
Source:
IEA
Projected DC Energy Consumption
945 TWh
Year:
2030
Trend:
Base Case; about 15% growth per year from 2024 to 2030
Source:
IEA
Delayed DC Projects

75 projects / ~$130B blocked or delayed

Year:
Q1 2026
Trend:
Largest single quarter on record, roughly matching all of 2025
Top 5 Hyperscaler CapEx

$424.3B cash capex (Amazon $131.8B, Alphabet $91.4B, Microsoft $83.1B, Meta $69.7B, Oracle $48.3B)

Year:
2025
Trend:
2026 guidance is $730B for Microsoft, Alphabet, Meta and Amazon only (midpoints; Microsoft and Meta include finance leases, Amazon's is cash capital expenditures; Oracle gives no calendar-year guidance)
Source:
Company 10-K and 10-Q cash-flow capex (Oracle on the four fiscal quarters to February 2026); Microsoft FY2027-Q1 call, Alphabet Q2 2026 call, Meta Q3 2026 release, Amazon Q3 2026 release
Average PUE (industry)
1.52
Year:
2026
Trend:
Down from 2.5 in 2007; gains remain gradual

Power Advantage: Space Solar

Space solar: 1,361 W/m², nearly continuous sunshine, 8x more productive than Earth. Solar array specific power evolution from ISS (27 W/kg) to advanced thin-film (>200 W/kg).

Space (Orbit)Earth (Surface)

Solar Irradiance (constant)

1,361 W/m² (solar constant)
~1,000 W/m² peak (sea level, clear day)

Wikipedia, PVEducation

Atmospheric Absorption Loss

0% (no atmosphere)
~25% absorbed/scattered

Wikipedia Solar Irradiance

Effective Average Irradiance

1,293-1,361 W/m² (LEO sun-sync)
~200-300 W/m² (location-dependent average)

NASA GSFC, S&P Global

Capacity Factor

Nearly continuous sunshine (dawn-dusk sun-synchronous LEO)
24.0% (US utility-scale PV, 2025)

Google preprint (arXiv 2511.19468v2), EIA Electric Power Monthly Table 6.07.B

Night/Weather Downtime

0-5% eclipse (orbit-dependent)
50-75% (night + weather + seasons)

Orbital mechanics

Annual Energy Output (per m²)

~11,200 kWh/m²/year
~1,400 kWh/m²/year (good locations)

Calculated from irradiance × capacity factor

Google's 8x Productivity Claim

8x more power per panel per year vs Earth
Baseline 1x

Google Research Blog

ISS Original Solar Arrays

27 W/kg specific power
-

NASA Solar Power Technologies

ISS iROSA Arrays (Redwire)

75.3 W/kg specific power
-

Wikipedia ROSA, Redwire

Advanced Thin-Film (target)

150-250 W/kg specific power
-

ScienceDirect

ISS Total Solar Power

~120 kW (end-of-life with iROSA)
-

NASA ISS Facts

Starcloud Solar Power

Not publicly disclosed
-

Starcloud

Degradation Rate

~1-2% per year (radiation)
~0.5% per year (weather/UV)

Industry standard

The Cooling Challenge

The ISS rejects up to 70 kW with six radiator units of 23.3 m by 3.4 m. A 20°C black plate radiates about 838 W per m². A NASA study put a 200 kW droplet radiator at about 500 kg against 3,500 kg for heat pipes.

ISS Active Thermal Control (EATCS)

70 kW heat rejection

External Active Thermal Control System: ammonia loops and radiators, designed for 35 kW per loop

NASA ATCS Overview

ISS Radiator Units

6 units, 23.3 m by 3.4 m each

Two radiator wings of three units each; each unit has eight panels and weighs about 1,123 kg

NASA ATCS Overview

Space Ambient Temperature

2.725 K (-270.4°C)

Cosmic microwave background, measured by COBE as a near-perfect blackbody

NASA COBE

Heat Rejection Method in Space

Radiation only (Stefan-Boltzmann law)

Conduction and convection to the environment are not available in space (Starcloud); the ISS releases heat by radiation to space (NASA)

Starcloud white paper, NASA ATCS Overview

Share of Power Dissipated as Waste Heat

Essentially all

Nearly all the power the solar arrays generate must leave as heat, so radiators are sized for gigawatts at data center scale

Starcloud white paper

Heat Radiated by a Plate at 20°C

~838 W per m² (both sides)

A 1 m by 1 m black plate radiating to deep space; roughly three times the electricity solar panels generate per square meter

Starcloud white paper

Heat-Pipe Radiator Mass

13 to 18 kg per kW

At about 300 K, excluding the interface mechanism; a 200 kW heat load would need about 3,500 kg of radiators

NASA CR 185164 (1989)

Advanced Liquid Droplet Radiators

~500 kg for 200 kW

Conceptual design, including generators, collectors, piping and structure, against about 3,500 kg of heat-pipe radiators; sprays droplets to radiate heat

NASA CR 185164 (1989)

Starcloud Approach

Radiators less than half the size of the solar arrays

White paper design: a 4 km by 4 km deployed solar array with a very large lightweight deployable radiator

Starcloud white paper

Google Suncatcher Approach

81-satellite cluster of 1 km radius

Heat pipes and radiators on each satellite; Google lists thermal management among the challenges still to be worked out

Google preprint (arXiv 2511.19468)

Launch Provider Comparison

14 vehicles from the Space Shuttle to Starship V3. Cost per kg only where NASA or Google prints it.

VehicleOperatorPayload to LEO (kg)Total CostCost per kgReusabilityStatus
Starship (100x reuse, Google analysis)SpaceXn/dn/d$15 or less100x component reuse (SpaceX target)Projected
Starship (10x reuse, Google analysis)SpaceXn/dn/d$60 or less10x component reuseProjected
Falcon HeavySpaceX63,800~$97M$1,520Partially reusableActive
Falcon 9SpaceX22,800~$67M$2,940Booster reuse (34 times as of Mar 31, 2026)Active (about 620 orbital launches by Mar 31, 2026)
Space ShuttleNASA27,500~$1.5B per mission$54,500Partial (orbiter + SRBs)Retired 2011
Atlas V (401)ULA9,800Not disclosedNot disclosedExpendableActive
Delta IV HeavyULAn/dNot disclosedNot disclosedExpendableRetired 2024 (final launch Apr 9, 2024)
Ariane 5ArianeGroupn/dNot disclosedNot disclosedExpendableRetired 2023 (final flight Jul 6, 2023)
Ariane 6 (A62)ArianeGroup10,300Not disclosedNot disclosedExpendableActive (first launch Jul 9, 2024)
Ariane 6 (A64)ArianeGroup21,600Not disclosedNot disclosedExpendableActive
ElectronRocket Labup to 300~$9.2M (average revenue per launch, H1 2026)Not disclosedExpendable (first stage recovery in development)Active (75 launches by Dec 31, 2025)
NeutronRocket Lab13,000Not disclosedNot disclosedReusable-ready (reusable configuration)In development (delivery to the launch pad targeted for Q4 2026)
Vulcan CentaurULA10,800-27,200Not disclosedNot disclosedExpendableActive (last flight Feb 12, 2026 after a solid rocket motor anomaly; next launch set for Nov 13, 2026)
New GlennBlue Origin45,000Not disclosedNot disclosedReusable first stageNG-3 mishap Apr 19, 2026 (FAA closed the investigation May 22, 2026); a May 2026 static-fire anomaly destroyed a vehicle and damaged the SLC-36 pad
Long March 5CASC (China)25,000Not disclosedNot disclosedExpendableFlown (maiden flight completed)
Starship V3SpaceX100,000Not disclosedNot disclosedFully reusable (design)Testing; 12th flight test in May 2026 debuted V3; payload delivery to orbit expected in 2H 2026

Google launch cost threshold: $200/kg

Launch price below which launch stops being the limiting cost, per Google's preprint.

US Data Center Electricity Use (Terrestrial Analog)

LBNL historical estimates: 58 TWh (2014) to 176 TWh (2023). 2028 dashed lines are LBNL scenarios (325 to 580 TWh), not a single-point forecast. Measured analog for the grid constraint orbital proposals cite. Not an orbital TAM.

Historical estimate2028 LBNL scenario

Terawatt-hours per year.

Big-Four Hyperscaler Capex (Measured)

Cash capex from SEC filings, 2019 to 2025. Combined 2025 cash capex exceeded $370B. Finance leases are excluded, so totals run below company guidance. Ground-side spending analog, not orbital revenue.

PUE Analog: Survey vs Hyperscale vs Theoretical Orbit

Uptime Institute 2024 survey average PUE 1.56 and Google 2023 fleet PUE 1.10 are measured. The 1.0 orbital figure is a theoretical floor (no convective HVAC). No commercial orbital DC has published a PUE.

1.56
Uptime survey 2024
1.10
Google fleet 2023
1.00
Orbital theoretical floor

Datacenter GPU Thermal Design Power

NVIDIA datasheet TDP: A100 SXM 400 W, H100/H200 SXM 700 W, B200 SXM 1,000 W. Specified heat, not in-orbit draw. Every watt in vacuum must be radiated. Starcloud-1 flew an H100-class GPU.

US Industrial Electricity Price (EIA, Measured)

Average industrial retail tariff, 2014 to 2025. Rose from 6.67 cents/kWh in 2020 to 8.62 cents in 2025. Ground-side energy-cost analog. Orbital sites would not pay this tariff.

PJM Capacity Prices (Power Scarcity Analog)

PJM base residual auction RTO clearing prices. 2025/26 cleared at $269.92 and the 2026/27, 2027/28 and 2028/29 auctions each cleared at the price cap ($329.17, $333.44 and $325.00) after a decade of low prices. Measured signal of data center load outrunning Northern Virginia supply. Not an orbital price.

US Interconnection Queues (Siting Delay Analog)

Berkeley Lab Queued Up: about 2,600 GW waiting at end-2023, more than double the installed US fleet. Typical waits are years. Most queued megawatts never get built. Terrestrial siting analog.

Terrestrial DC Equipment Lead Times

DOE and industry quoted ranges as of 2025: large transformers 24 to 48 months, heavy-duty turbines 36 to 60 months. Shape of the ground-side delay, not a forecast of orbital hardware lead times.

ISS Solar Power (Only MW-Class Orbital Analog)

NASA STMD: six iROSA wings raised combined ISS generation to more than 250 kW. Each wing is more than 28 kW at beginning of life (Redwire/NASA) or more than 20 kW in NASA's 2023 blog. Measured flight analog, not an orbital DC nameplate.

ItemValueKind
Unshaded remainder of original ISS arraysAbout 95 kW combinedmeasured
Each iROSA wing (NASA 2023 blog)>20 kWmeasured
Each iROSA wing (BOL)>28 kW BOLmeasured
Six iROSA wings (Redwire)>120 kW combined BOLmeasured
ISS combined generation with six iROSA>250 kWmeasured
Two additional iROSA (requested)About 15 kW more once installedplanned

Ask Sterling

Register for a premium account to gain access to Sterling AI.

Get Started

Things you can ask Sterling:

Summarize Tesla's latest earnings reportWhy did NVIDIA's margins expand?Compare Apple vs Microsoft's cash flowWhat's driving EV industry growth?
Menu
Favorites
Orbital Data Center Economics: Market Data | Sterling