Orbital Data Center Risks

Full Sector Timeline

Milestones from HPE SBC-1 (2017) through Starcloud-1 (Nov 2025) and Google's Suncatcher prototype (Oct 2026) to projected cost parity (mid-2030s).

Jul 22, 2026

FCC adopts Space Modernization Report and Order (FCC 26-47; adopted Jul 22, released Jul 23, 2026)

Adopts a new Part 100 built on a licensing assembly line and will sunset Part 25 once Part 100 takes effect. Allows conditional grants (federal coordination, commercial coordination, orbital debris deferral) and sets annual processing rounds with a Jan 1 to Oct 31 filing window

FCC.gov (FCC 26-47)

Jul 30, 2026

K2 Space raises $500M Series D at $6.8B valuation

Led by Kleiner Perkins and ICONIQ, with CapitalG, Lightspeed, Altimeter, Spark Capital, Sands Capital, ARK Invest and accounts advised by T. Rowe Price Associates participating; K2 has raised over $1 billion in total

K2 Space release (Jul 30, 2026)

Oct 1, 2026

Google's Suncatcher prototype satellite launches on Transporter-18

Built with Planet; Google confirmed contact and said the satellite is operating as expected. Its Oct 30, 2026 post reports the first response from a data center class TPU in orbit and names twin interconnected satellites in 2027 as the next milestone; Trillium TPUs were tested to 15 krad(Si) on the ground (Nov 2025)

Google blog (Oct 1 and Oct 30, 2026); Google Research blog (Nov 4, 2025)

Nov 2, 2026

ESA counts about 47,280 tracked objects and about 15,800 functioning satellites

ESA statistics (last updated Nov 2, 2026): about 47,280 objects regularly tracked by space surveillance networks, about 19,610 satellites in space and about 15,800 functioning; MASTER-8 (Feb 2026 population) estimates 68,450 objects larger than 10 cm, including about 11,300 active payloads

ESA DISCOS Space Environment Statistics (updated Nov 2, 2026)

2027 (proj)

Starcloud-2, first commercial mission, planned for sun-synchronous orbit

Starcloud: GPU cluster, persistent storage and 24/7 access, fully operational in sun-synchronous orbit by 2027. Crusoe (Oct 22, 2025) plans to run Crusoe Cloud on a Starcloud satellite scheduled to launch in late 2026 and offer limited GPU capacity from space by early 2027

Starcloud (Starcloud-2 page); Crusoe release (Oct 22, 2025)

2027 (proj)

Axiom Orbital DC Node on ISS (AxODC Node ISS) with Spacebilt, Skyloom, Phison, Microchip

Skyloom optical terminal up to 2.5 Gbps (100 Gbps a future step); Phison SSDs for over 1 petabyte; Microchip PIC64-HPSC as core controller; Axiom plans at least three interconnected ODC nodes by 2027

Axiom Space and Spacebilt release (Sep 16, 2025)

2028 (proj)

Axiom Station first module launch planned

Axiom plans to launch the first module in 2028 and the second less than a year later; together they form a four-crew, operational free-flying station

Redwire release on Axiom Space newsroom (Oct 5, 2026)

Mid-2030s (proj)

Projected cost parity: orbital vs terrestrial DCs

Google projects launch prices below $200/kg by the mid-2030s, at which point launch plus operating cost per kW-year could be roughly comparable to a terrestrial data center's reported energy cost. Its preprint says sustaining the ~20% learning rate would take about 180 Starship launches a year. Not confirmed in a primary source as of 2026-10-31: the Deutsche Bank figure.

Google Research blog (Nov 4, 2025); Google preprint arXiv 2511.19468v2 (Jun 17, 2026)

Regulatory Landscape

FCC 5-year deorbit rule, FCC Space Modernization NPRM, ITAR/EAR CSA exception, ESA Zero Debris Charter, NOAA licensing, and ITU spectrum congestion.

Regulation / PolicyAgency / BodyDateKey ProvisionsImpact on Space DCsSource
5-Year Deorbit Rule (FCC 22-74)FCC (US)Adopted Sep 29, 2022; Effective Sep 29, 2024All new satellites must deorbit within 5 years of end-of-life (replaces 25-year guideline)Limits useful satellite lifetime; increases replacement cadence and cost; favors distributed architecturesFCC.gov
Space Modernization for the 21st CenturyFCC (US)NPRM adopted 2025 (Fed. Reg. Dec 5, 2025); Report and Order and Further NPRM (FCC 26-47) adopted Jul 22, 2026, released Jul 23, 2026Overhauls Part 25 and creates a new Part 100 'licensing assembly line'; orbital debris criteria, satellite identifiability, energy venting, conditional grantsSpeeds and streamlines space DC licensing; could add design/compliance requirementsFCC.gov (FCC 26-47), Federal Register
FCC NEPA Modernization (space-based operations)FCC (US)Report and Order and Further NPRM (FCC 26-64) adopted Sep 30, 2026, released Oct 1, 2026Finds that Commission-licensed space-based operations and activities are excluded from NEPA's definition of a major federal action because their effects occur in outer space, outside US jurisdiction; spectrum authorizations for space-based operations do not create federal control over later non-federal deploymentsRemoves NEPA review from FCC licensing of orbital data center constellations, shortening one licensing stepFCC.gov (FCC 26-64, WT Docket 25-217)
ITAR/EAR License Exception CSAFCC / State Dept (US)Announced Oct 23, 2024New export control exception for commercial space activities (Lunar Gateway, Mars, CLEO)Eases international partnerships for commercial space computing; reduces export frictionFederal Register
ESA Zero Debris CharterESA (Europe)Finalized Oct 2023; Target 2030Non-binding commitment to achieve zero new debris by 2030; 12 countries signed on May 22, 2024, with 100+ organizations pledged to signMay influence European orbital DC designs; voluntary compliance standardESA (Nov 6, 2023; May 22, 2024)
NOAA Remote Sensing LicensingNOAA (US)Ongoing / EvolvingLicensing requirements for Earth observation from space; evolving to cover compute-enhanced sensingSpace-based computing with EO may trigger new licensing requirementsNOAA
ITU Spectrum CoordinationITU (International)OngoingCoordination for inter-satellite links and ground station frequencies; increasing congestionGrowing competition for Ka-band, optical, and other spectrum for space DC communicationsITU
Golden Dome AuthorizationUS Congress / DoDEnacted Jul 4, 2025 (Public Law 119-21)Missile defense appropriations in the reconciliation law (Sec. 20003): $5.6B for space-based and boost phase intercept, $7.2B for military space-based sensors, $2.55B for missile defense capabilities, available until Sep 30, 2029Creates major government demand signal for orbital computing capabilitiesPublic Law 119-21, Sec. 20003
SDA Proliferated ArchitectureSDA / Space Force2020-ongoingHundreds of small satellites in LEO mesh for missile tracking and data relayEstablishes military demand and standards for space-based computing and networkingSDA.mil

Risks & Challenges

10 risks quantified, from radiation and thermal management (the ISS cooling system is rated at 70 kW) to space debris (over 1,000 Starlink maneuvers a day).

Radiation DamageHigh

Google Trillium TPU: HBM irregularities from 2 krad(Si), no hard failures up to 15 krad(Si); COTS GPUs not rad-rated

Plasteel shielding (limits single event effects); software hardening (HPE approach); traditional rad-hard (5-10x cost)

Google preprint (arXiv 2511.19468), Cosmic Shielding, HPE, Aethero release (Jul 8, 2024)

Thermal ManagementVery High

ISS external cooling system rated at 70 kW with six radiator units of 23.3 m by 3.4 m; Starcloud's white paper sizes radiators at less than half the area of its solar arrays

Distributed small satellites (Google's 81-satellite cluster); liquid droplet radiators (NASA concept: about 500 kg against 3,500 kg of heat-pipe radiators for 200 kW); advanced heat pipe technology

NASA ATCS Overview, Starcloud white paper, NASA CR 185164

Launch Failure / CostHigh

SpaceX does not typically obtain insurance for its satellites, payloads or launch vehicles; NASA's Ames paper puts Falcon 9 at $2,940/kg

Google estimates Starship at about $60/kg with 10x reuse and $15/kg with 100x reuse; constellation redundancy; in-orbit spares

SpaceX prospectus (Jun 12, 2026), NASA Ames paper (Jones, 2023), Google preprint (arXiv 2511.19468)

Space Debris / CollisionHigh

About 47,280 tracked objects (ESA, Nov 2, 2026); Starlink satellites made over 1,000 collision avoidance maneuvers per day in 2025

5-year deorbit rule; active debris removal; lower orbit (higher drag = faster deorbit)

ESA DISCOS, SpaceX prospectus (Jun 12, 2026), FCC 22-74

LatencyMedium

LEO: 20-40 ms round-trip vs <1 ms fiber; GEO: 480-600 ms

LEO for lowest latency; mesh networking; edge caching in orbit

Physics, industry standard

ServiceabilityVery High

SpaceX's prospectus: once deployed, orbital AI compute will not be easily repaired or upgraded, so component failures could mean permanent capacity loss

Design for reliability; modular constellation (replace entire sats); autonomous fault recovery

SpaceX prospectus (Jun 12, 2026)

Regulatory UncertaintyMedium

FCC Space Modernization Report and Order adopted Jul 22, 2026 (further NPRM pending); ITAR/EAR complexity; ITU spectrum congestion growing

Engage regulators early; comply with strictest standards; diversify across jurisdictions

FCC, Federal Register

Power ScalingHigh

Starcloud's white paper sizes a 5 GW data center at a solar array of about 4 km by 4 km

Thin-film solar (150-250 W/kg target); space solar beamed power (Aetherflux approach)

Starcloud white paper, ScienceDirect

Bandwidth BottleneckMedium

Ground contact 5-15 min/pass in LEO; 200 Gbps max demonstrated (TBIRD)

Optical ISL mesh (100 Gbps-1.6 Tbps); in-orbit processing (reduce downlink 30,000x)

NASA TBIRD, HPE, Google

Technology MaturityMedium-High

First GPU in space only Nov 2025; no multi-year space DC operational data yet

Phased approach (crawl/walk/run); ISS heritage (HPE: 207 days without reset, per NASA); incremental scaling

Industry consensus, NASA (Feb 20, 2021)

Space Debris Trend

ESA counts about 47,280 tracked objects (Nov 2026) and models 68,450 objects above 10 cm, up from 54,000 in Aug 2024. Starlink satellites made over 1,000 collision avoidance maneuvers a day in 2025.

MetricValueYear / PeriodSource
Tracked Objects
Total tracked objects~47,280Nov 2, 2026ESA DISCOS Space Environment Statistics
Collision Avoidance
Starlink collision avoidance maneuversOver 1,000 per day2025SpaceX prospectus (Jun 12, 2026)
Starlink collision avoidance maneuvers65,137 (Gen 1) and 142,015 (Gen 2)Dec 2025-May 2026 (6 months)SpaceX FCC semi-annual report (Jul 1, 2026)
Starlink satellites in orbit~9,600Mar 31, 2026SpaceX prospectus (Jun 12, 2026)
Estimated Untracked
Estimated objects 1-10 cm~1,200,000Aug 2024 populationESA Space Environment Report 2025
Estimated objects 1-10 cm~1,500,000Feb 2026 populationESA Space Environment Report 2026
Estimated objects 1 mm - 1 cm~130,000,000Aug 2024 populationESA Space Environment Report 2025
Estimated objects 1 mm - 1 cm~230,000,000Feb 2026 populationESA Space Environment Report 2026
Other
Functioning satellites in orbit~15,800Nov 2, 2026ESA DISCOS Space Environment Statistics
Modelled objects >10 cm (MASTER)~54,000Aug 2024 populationESA Space Environment Report 2025
Modelled objects >10 cm (MASTER-8)~68,450Feb 2026 populationESA Space Environment Report 2026
Active payloads in modelled population~9,300Aug 2024 populationESA Space Environment Report 2025
Active payloads in modelled population~11,300Feb 2026 populationESA Space Environment Report 2026
Non-deliberate fragmentation events/year10.5 per yearAverage of the two decades to 2024ESA Space Environment Report 2025
Non-deliberate fragmentation events/year9.8 per yearAverage of the two decades to 2025ESA Space Environment Report 2026
New catalogued fragments (2024)3,000+2024ESA Space Environment Report 2026
Average LEO impact velocity~10 km/sRelative speed between objectsNASA Orbital Debris Program Office
Maximum LEO impact velocity~15 km/sRelative speed between objectsNASA Orbital Debris Program Office
FCC deorbit rule5 years post-end-of-lifeSatellites launched after Sep 29, 2024FCC 22-74
ESA Zero Debris targetStop generating space debris by 2030ESA Zero Debris CharterESA

Bull vs Bear Case

Company filings, research papers and government data for and against orbital data centers, dimension by dimension.

Market Timing
Bull

Musk, in the World Economic Forum's own recording (Jan 22, 2026), said space will be the lowest-cost place to put AI within 2 years, maybe 3 at the latest. SpaceX's prospectus says it expects to begin deploying orbital AI compute satellites as early as 2028. Google's next milestone is twin interconnected satellites in 2027 (Google, Oct 30, 2026).

Bear

SpaceX's prospectus says its orbital AI compute initiative has not yet been proven at commercial scale, or at all, and may ultimately be unsuccessful, and that the timeline and the launch cadence required may be difficult or impossible to determine.

Key Inflection Point: First commercial orbital compute workload delivering revenue (Estimate)

Energy Economics
Bull

Google: panels in a dawn-dusk sun-synchronous orbit get nearly continuous sunshine and up to 8x more solar energy per year than a mid-latitude panel on Earth (Google preprint), against a 24.0% capacity factor for US utility-scale solar PV in 2025 (EIA). Starcloud: about $0.002/kWh in orbit versus $0.045/kWh average US wholesale electricity, assuming one $5M launch per 40 MW and $0.03/W solar cells (white paper, Sep 2024).

Bear

Google: even if launch costs reach $200/kg, launched power would cost about $810/kW/yr for Starlink v2-type satellites, only roughly comparable to US data center power spend of $570 to $3,000/kW/yr (Google preprint).

Key Inflection Point: Launch prices fall to about $200/kg, which Google's learning-curve analysis reaches by the mid-2030s (Estimate)

Cooling Physics
Bull

Starcloud: deep space averages about 2.7 K, so radiators facing it can shed waste heat, and they need be less than half the size of the solar arrays (white paper, Sep 2024); no water is needed for cooling (NVIDIA blog, Oct 15, 2025). SpaceX expects radiative cooling to incur no operating costs compared with liquid or air cooling (prospectus).

Bear

NASA: the ISS external cooling system is designed to reject 70 kW in total, using six radiator units each 23.3 m by 3.4 m. Google calls thermal management a critical optimization challenge for power-dense TPUs in a vacuum (Google preprint). Starcloud itself estimates only a power usage effectiveness comparable to state-of-the-art terrestrial hyperscale data centers (white paper).

Key Inflection Point: Radiators or heat pumps proven in orbit at heat loads well above the ISS's 70 kW

Critical Unknowns

  1. 1.Starship cadence: Google's learning curve needs about 180 Starship launches a year for launch prices to fall below $200/kg by about 2035 (Google preprint)
  2. 2.Radiation and repair: Google's Trillium TPU showed memory irregularities at 2 krad(Si) and the effect of single event effects on training needs further study; SpaceX says orbital AI compute cannot be readily repaired (Google preprint; SpaceX prospectus)
  3. 3.Thermal solution maturity: the ISS external cooling system is designed to reject 70 kW, and Google calls thermal management a critical challenge for TPUs in a vacuum (NASA; Google preprint)
  4. 4.Terrestrial grid outlook: PJM's 2028/2029 auction procured 6,831 MW below its reliability requirement, while SpaceX says terrestrial energy breakthroughs could diminish orbital viability (PJM; SpaceX prospectus)
  5. 5.Defense funding: whether defense programs pay for on-orbit compute directly; P.L. 119-21 appropriations for space-based sensors and intercept run to Sep 30, 2029 (P.L. 119-21)
  6. 6.Satellite manufacturing cost: SpaceX says it believes it will be the first and only company to manufacture satellites at the scale of automotive manufacturing (SpaceX prospectus)
  7. 7.Timeline: Google's twin prototype satellites in 2027, SpaceX orbital AI compute deployment as early as 2028, and Google's launch price at or below $200/kg by about the mid-2030s (Google; SpaceX prospectus; Google preprint)

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 Risks: Market Data | Sterling