Space DC Platform Comparison
Technical specifications for all announced and operational orbital data center platforms — Starcloud, Axiom, K2 Space, Google Suncatcher, OrbitsEdge, Lonestar, HPE SBC, Aethero, and ASCEND.
| Company | Platform | Mass | Orbit | Solar Power | Compute | Storage | Comms | Launch Date | Status | Confidence |
|---|---|---|---|---|---|---|---|---|---|---|
| Starcloud | Starcloud-1 (Lumen 1) DCD, CNBC, SkyRocket DB, Starcloud.com | 60 kg | 325 km LEO | 1 kW (Astro Digital Corvus-Micro bus) | NVIDIA H100 (80 GB HBM3, ~4 PFLOPS FP8) | H100 HBM3 only (no additional SSD disclosed) | Ka-band TT&C; Starlink relay for data backhaul | Nov 2, 2025 | Operational (11-month lifetime) | High |
| Starcloud | Starcloud-2 SpaceNews, DCD, Crusoe Press, NVIDIA | Not disclosed (est. larger than SC-1) | Sun-synchronous (altitude TBD) | ~100 kW (100x SC-1, per company) | Multiple H100s + NVIDIA Blackwell (specific chip TBD); Crusoe Cloud layer | Not disclosed | Optical ISL (laser sat-to-sat) + Ka-band TT&C | Late 2026 / 2027 | Planned; Crusoe partnership confirmed | Medium |
| Axiom Space | ODC Node (AxODC) Axiom Space, Phison Blog, Kepler, Microchip | ~260 kg (unconfirmed — no official spec found) | ISS orbit (408 km) | ISS power allocation (not publicly specified) | Microchip PIC64-HPSC (8x RISC-V @ 1 GHz, 2 TOPS int8, 1 TFLOPS bf16) | >1 PB Phison Pascari enterprise SSDs (confirmed) | Kepler 10 Gbps optical (initial; 100 Gbps planned) + Skyloom 2.5 Gbps OCT | 2027 | In development; ISS integration | Medium-High |
| Axiom Space | DCU-1 (AxDCU-1) Red Hat Press, Axiom Space | Not disclosed | ISS orbit (408 km) | ISS power allocation | Red Hat Device Edge (MicroShift/RHEL/Ansible) — underlying hardware not specified | Not disclosed | ISS communications | Spring 2025 (target); no launch confirmation found | Integration / delayed | Low-Medium |
| K2 Space | GRAVITAS (Mega Class) K2 Space, PRNewswire, SES Press | Not disclosed (payload capacity ~1,000 kg) | MEO (orbit-raise from LEO via electric propulsion; specific altitude TBD) | 20 kW (twin 10 kW solar arrays — confirmed) | Custom national security + commercial payloads | Not disclosed | SES meoSphere relay (modular software-defined MEO network) | Q1 2026 | Integration; $60M STRATFI contract | Medium-High |
| OrbitsEdge | SatFrame / Edge1 OrbitsEdge, HPE, FactoriesInSpace | ~2 kg | LEO (altitude TBD) | ~10 W | HPE Edgeline EL8000 (15 TOPS) | Not confirmed (256 GB claim unverified in sources) | Ground + relay (standard sat comms) | 2026 | Orbital demo planned with HPE | Medium |
Project Suncatcher Google Research Blog, DCD | Not disclosed (81 satellites total) | ~650 km sun-synchronous orbit (SSO) | Solar (8x productivity vs Earth — Google claim); per-sat power TBD | Google Trillium TPU v6e (count per satellite not disclosed; ≥15 krad TID) | Not disclosed | 1.6 Tbps bidirectional optical (bench test); 800 Gbps unidirectional per link | Early 2027 (2 prototypes with Planet Labs) | Prototyping / lab testing | Medium-High | |
| ASCEND (EU) | Orbital DC Demonstrator Thales Alenia Space, CNBC, ASCEND Project | Not disclosed | 1,400 km (confirmed from feasibility study) | Solar (23 GW ultimate target); requires new 'eco-launcher' 10x less emissive | TBD (feasibility stage) | TBD | TBD | 2026 demo (EROSS IOD robotic assembly); full: 13 modules by 2036, 1,300 by 2050 | Feasibility complete; demonstrator planned | Medium |
| Lonestar Data | Freedom Lonestar, Phison Blog, Tom's Hardware | Not confirmed (described as 'shoebox-sized') | Lunar surface (via IM-2 Athena lander, not IM-1) | Solar (lander power system) | Microchip PolarFire FPGA + RISC-V processor (not GPU-class) | 8 TB Phison Pascari enterprise SSD (confirmed) | Ground relay via lander | IM-2 lander (launched Feb 2025; landing early Mar 2025) | Launched (corrected from IM-1 to IM-2) | High |
| HPE | Spaceborne Computer-2 HPE, ISS National Lab, TweakTown | Not disclosed | ISS orbit (408 km) | ISS power | HPE Edgeline / ProLiant edge server (software-hardened COTS) | 130 TB confirmed (KIOXIA 960GB + 1.024TB NVMe + 30.72TB SAS SSDs) | ISS communications | v1: Aug 2017-Jun 2019 (615 days); v2: Feb 2021; 130TB refresh: Jan 30, 2024 | Operational on ISS (ongoing) | High |
| Aethero | Deimos (1.5U CubeSat) Aethero, NanoSats.eu, SpaceNews | ~2 kg (1.5U standard) | LEO | Solar (not specified) | NVIDIA Jetson Orin NX (100 TOPS — confirmed; first space-rated 100 TOPS) | Not disclosed | Weak transmitter; limited bandwidth; low-capacity battery | Aug 2024 | Operational in orbit | High |
| Sophia Space | TILE Module Sophia Space, TechCrunch, PerAspera, FoundersToday | Not disclosed (form factor: 1 m² × 1 cm thick) | 600-1,000 km sun-synchronous orbit | 92% solar-to-compute efficiency (patented passive cooling + solar design) | 4× NVIDIA Jetson Orins per TILE (confirmed) | Not disclosed | Not disclosed | 2027-28 orbital demo | Ground testing; $13.5M raised | Medium |
| Aetherflux | Galactic Brain TechCrunch, The Register, DCD, PRNewswire | Not disclosed | LEO (altitude TBD) | Space solar + infrared power beaming to ground receivers | Not disclosed (GPU-class inference focus) | Not disclosed | Not disclosed | Q1 2027 operational target; ~30 sats per Falcon 9 launch | Development; $60M raised | Low-Medium |
| NTT + SKY Perfect JSAT | Space DC Satellite NTT R&D, DCD | Not disclosed | Multi-orbit: HAPS + LEO + GEO | Solar (not specified) | Silicon photonics + fiber + analog ICs; IOWN optical network tech | Not disclosed | Optical (IOWN technology) | 2025 target (no launch confirmation found); ops start 2026 | Development; launch unconfirmed | Low-Medium |
Specifications from official disclosures, press releases, and analyst reports. Some values are targets or estimates. Confidence reflects data verifiability.
Space Compute Hardware
All known compute hardware deployed or planned for orbital data centers — GPUs, TPUs, FPGAs, edge processors, and radiation shielding approaches.
| Chip / Module | Vendor | Type | Performance | Rad Tolerance | Power (W) | Deployed By | Status |
|---|---|---|---|---|---|---|---|
| H100 SXM NVIDIA Datasheet | NVIDIA | GPU | 80 GB HBM3, 3.35 TB/s BW, ~4 PFLOPS FP8 | Not rad-hard (COTS) | 700 | Starcloud-1 | In orbit (Nov 2025) |
| Blackwell B200 NVIDIA | NVIDIA | GPU | Next-gen (est. 2-3x H100) | COTS (shielded) | 1,000 | Starcloud-2 (planned) | Planned 2026-27 |
| Vera Rubin Space Module NVIDIA Newsroom, Tom's Hardware | NVIDIA | GPU Module | 25x H100 compute (inference) | Custom space design | TBD | Orbital DC partners | Announced Mar 2026 |
| Trillium TPU v6e Google Research | TPU | Custom AI inference/training | Survives ≥15 krad(Si); HBM fails ~2 krad | TBD | Project Suncatcher | Testing / 2027 launch | |
| Jetson Orin NX NVIDIA Developer, Aethero | NVIDIA | Edge GPU | 20-157 TOPS (configurable) | COTS + Plasteel shielding | 10-40 | Aethero Deimos, NxN-ECM | In orbit (Aug 2024) |
| NxA-ECM Aethero | Aethero | Edge Module | 275-550 TOPS | Plasteel shielded COTS | TBD | Aethero next-gen | Q4 2025 |
| PIC64-HPSC Microchip, RISC-V Intl | Microchip | Processor | 8x RISC-V X280 @ 1 GHz, 2 TOPS int8, 1 TFLOPS bf16 | Space-qualified (TBD krad) | TBD | Axiom ODC Node, NASA | In production |
| Versal XQR ACAP Spirit Electronics, ESA | AMD/Xilinx | FPGA+AI | 400 AI engines, DSP-focused | ≥30 krad TID, MIL-STD-883 Class B | TBD | Military/aerospace | Flight-qualified |
| Kintex UltraScale XQR AMD | AMD/Xilinx | FPGA | High-performance reconfigurable | Rad-tolerant (20nm) | TBD | Various defense | Flight-qualified |
| RH12 (12nm FinFET) GlobalFoundries, BAE | BAE / GlobalFoundries | Custom ASIC | Custom rad-hard IC platform | Full rad-hard by design | TBD | Military/intelligence | Production 2025+ |
| GR765 (NOEL-V) Frontgrade | Frontgrade | Processor | RISC-V space processor | Space-qualified | TBD | ESA missions | Shipping 2025-26 |
| VA4 MCU VORAGO | VORAGO | MCU | ARM Cortex-M based | 300 krad(Si) TID, SEL >110 MeV | Low | LEO missions | Q1 2026 shipments |
| SAMD21RT Microchip | Microchip | MCU | ARM Cortex-M0+ (rad-tolerant) | Radiation-tolerant | Low | CubeSats, small sats | In production |
| ZSOM-F01 ZES | Zero Error Systems | SoM | Radiation-tolerant COTS FPGA SoM | COTS with error correction | Low | Various | Q1 2026 shipments |
| Edgeline EL4000 HPE, ISS National Lab | HPE | Edge Server | Commercial edge server (software-hardened) | Software-hardened COTS | Varies | HPE SBC-1 & SBC-2 on ISS | Flight heritage (615 days) |
Space-qualified and COTS compute hardware for orbital data centers. Sources: manufacturer datasheets, press releases, NASA/ESA documentation.
Orbit Type Comparison
LEO vs MEO vs GEO vs Lagrange points — latency, solar illumination, radiation, debris risk, launch cost, and best use cases for each orbit type.
| Parameter | LEO200 - 2,000 km | MEO2,000 - 35,786 km | GEO35,786 km | LagrangeL1 - L5 points |
|---|---|---|---|---|
| One-way Latency | 1-10 ms | 30-60 ms | 240-300 ms | >300 ms (1.3s for L2) |
| Round-trip Latency | 2-20 ms | 60-120 ms | 480-600 ms | >600 ms |
| Solar Illumination | 60-75% (eclipsed 25-40%) | 80-95% | ~99% (short eclipse seasons) | ~100% (near-constant) |
| Radiation Environment | Lower (magnetosphere protection) | Van Allen belt: ~100x LEO dose | High: 100-1,000x LEO dose | Minimal (solar wind only) |
| Debris Risk | CRITICAL (500-800 km peak) | Low-moderate | Very low | Negligible |
| Tracked Debris Objects | ~40,230 total (Apr 2025) | ~Few hundred | <100 significant | ~0 |
| Launch Cost (relative) | Baseline ($2,600/kg Falcon 9) | 2-3x LEO cost | 3-4x LEO cost | 4-5x LEO cost |
| Orbital Period | 90-120 minutes | 2-12 hours | 24 hours (stationary) | Varies (halo orbits) |
| Ground Contact per Pass | 5-15 minutes | Hours | Continuous (fixed position) | Near-continuous |
| Atmospheric Drag | Significant below 500 km | Negligible | — | — |
| Best Use Case | Low-latency compute, EO processing, broadband | Navigation, regional comms, defense | Broadcast, fixed comms, weather | Deep space relay, astronomy |
| Key DC Players | Starcloud, Axiom, OrbitsEdge, Aethero, Google Suncatcher | K2 Space (GRAVITAS), SES | Legacy telecom (Intelsat, Viasat) | NASA JWST, ESA missions |
| Deorbit Requirement | 5 years (FCC rule, effective Sep 2024) | 25+ years (varies) | Graveyard orbit (+300 km) | — |
LEO highlighted as the primary orbit for near-term space data center deployments. Sources: standard orbital mechanics references, FCC filings, industry estimates.
Optical Inter-Satellite Links
Bandwidth specifications for satellite laser communication systems — Starlink ISL, Amazon Kuiper, China records, Skyloom, Kepler, Google Suncatcher, and NASA TBIRD.
| System | Operator | Bandwidth | Range | Type | Status |
|---|---|---|---|---|---|
| Starlink Network Total Industry estimates | SpaceX | ~42 PB/day aggregate (est.) | Global mesh | Full constellation | Operational |
| TeraWave ISL Blue Origin | Blue Origin | Part of 6 Tbps network | LEO + MEO mesh | Sat-to-Sat + Sat-to-Ground | Announced Jan 2026; deploy Q4 2027 |
| Google Suncatcher Google Research Blog | Google Research | 1.6 Tbps bidirectional / 800 Gbps unidirectional | Short range (bench) | Bench-scale demonstrator | Lab testing; flight 2027 |
| Starlink V3 ISL NextBigFuture, ISPreview | SpaceX | 1 Tbps downlink / 160 Gbps uplink per sat | TBD | Sat-to-Sat + Sat-to-Ground | V3 sats in deployment |
| China Laser Starcom IEEE Spectrum, SCMP | China (various institutes) | 400 Gbps (record) | 640 km | Sat-to-Sat (record demo) | Demonstrated March 2025 |
| NASA TBIRD NASA Goddard | NASA / MIT Lincoln Lab | 200 Gbps | LEO to ground | Space-to-Ground (optical) | Demonstrated Jun 2023; deorbited Sep 2024 |
| Starlink ISL (V2 Mini) Hackaday, ArXiv | SpaceX | 100 Gbps per link | Thousands of km | Sat-to-Sat (optical) | Operational (10,000+ sats) |
| Project Kuiper OISL AboutAmazon, IEEE Spectrum | Amazon | 100 Gbps | Up to 2,600 km demonstrated | Sat-to-Sat (optical) | Demonstrated Dec 2023; production sats 2025 |
| Skyloom/NEC WARP OCT NEC Press, SatNews | Skyloom + NEC | 100 Gbps | Multi-orbit | Sat-to-Sat (advanced) | In development |
| Kepler WARP Terminal Kepler Space | Kepler Communications | 100 Gbps+ | LEO-to-LEO relay | Optical data relay | 10 sats launching Jan 2026 |
| Skyloom V'ger Skyloom, SatNews | Skyloom (now IonQ) | 10 Gbps base | Multi-orbit | Sat-to-Sat (SDA compliant) | Production 2025-26 |
Optical inter-satellite links and space-to-ground laser communication systems, sorted by bandwidth. Sources: operator press releases, NASA, IEEE Spectrum.
Radiation Hardening Approaches
Three paradigms for protecting computing hardware in orbit — traditional rad-hard, software-hardened COTS, and shielded COTS (Cosmic Shielding Plasteel).
Traditional Rad-Hard
Approach
Radiation-hard-by-design ICs using specialized fabrication
Cost Premium vs COTS
5-10x higher (industry estimate)
Performance vs COTS
1-3 generations behind (significant penalty)
Radiation Tolerance (TID)
100-300+ krad(Si) by design
SEE (Single Event Effect) Protection
Built into silicon design
Mission Lifetime Extension
Designed for 5-15 year missions
Mass Impact
Standard IC weight
Lead Time
12-18 months typical
Flight Heritage
Decades (Apollo era onward)
Key Suppliers
BAE Systems (RH12), Microchip (SAMD21RT), Frontgrade (GR765), VORAGO (VA4)
Government Support
Long DOD/NASA heritage programs
Best For
Critical military/deep space missions requiring absolute reliability
Software-Hardened COTS
Approach
COTS hardware + adaptive software that manages radiation events
Cost Premium vs COTS
~1x (standard COTS hardware cost)
Performance vs COTS
Full COTS performance (throttled during radiation events only)
Radiation Tolerance (TID)
Varies by component; software compensates
SEE (Single Event Effect) Protection
Software detection and recovery
Mission Lifetime Extension
615 days demonstrated on ISS (HPE SBC-1)
Mass Impact
Standard + some redundancy
Lead Time
Standard procurement
Flight Heritage
615 days ISS (9/20 SSDs failed, zero unrecoverable compute errors)
Key Suppliers
HPE (Edgeline EL4000), general Linux-based approaches
Government Support
NASA ISS programs
Best For
Cost-sensitive LEO missions with ISS-like radiation environment
Shielded COTS (Plasteel)
Approach
COTS hardware + nanocomposite material shielding (Plasteel)
Cost Premium vs COTS
Near cost parity with aluminum; 15x cost savings vs traditional rad-hard
Performance vs COTS
Full COTS performance (constant)
Radiation Tolerance (TID)
60% reduction in radiation exposure vs aluminum
SEE (Single Event Effect) Protection
10x reduction in SEE rates
Mission Lifetime Extension
8x lifespan vs conventional materials
Mass Impact
40x mass savings vs equivalent metal shielding
Lead Time
~2 weeks (Cosmic Shielding claim)
Flight Heritage
Tested on Space Forge satellite; Aethero Deimos in orbit
Key Suppliers
Cosmic Shielding Corp (exclusive Plasteel manufacturer)
Government Support
$4M Pentagon TACFI contract (Oct 2025)
Best For
Commercial LEO constellations needing COTS performance + protection
Approach
TraditionalRadiation-hard-by-design ICs using specialized fabrication
Software-HardenedCOTS hardware + adaptive software that manages radiation events
ShieldedCOTS hardware + nanocomposite material shielding (Plasteel)
Cost Premium vs COTS
Traditional5-10x higher (industry estimate)
Software-Hardened~1x (standard COTS hardware cost)
ShieldedNear cost parity with aluminum; 15x cost savings vs traditional rad-hard
Performance vs COTS
Traditional1-3 generations behind (significant penalty)
Software-HardenedFull COTS performance (throttled during radiation events only)
ShieldedFull COTS performance (constant)
Radiation Tolerance (TID)
Traditional100-300+ krad(Si) by design
Software-HardenedVaries by component; software compensates
Shielded60% reduction in radiation exposure vs aluminum
SEE (Single Event Effect) Protection
TraditionalBuilt into silicon design
Software-HardenedSoftware detection and recovery
Shielded10x reduction in SEE rates
Mission Lifetime Extension
TraditionalDesigned for 5-15 year missions
Software-Hardened615 days demonstrated on ISS (HPE SBC-1)
Shielded8x lifespan vs conventional materials
Mass Impact
TraditionalStandard IC weight
Software-HardenedStandard + some redundancy
Shielded40x mass savings vs equivalent metal shielding
Lead Time
Traditional12-18 months typical
Software-HardenedStandard procurement
Shielded~2 weeks (Cosmic Shielding claim)
Flight Heritage
TraditionalDecades (Apollo era onward)
Software-Hardened615 days ISS (9/20 SSDs failed, zero unrecoverable compute errors)
ShieldedTested on Space Forge satellite; Aethero Deimos in orbit
Key Suppliers
TraditionalBAE Systems (RH12), Microchip (SAMD21RT), Frontgrade (GR765), VORAGO (VA4)
Software-HardenedHPE (Edgeline EL4000), general Linux-based approaches
ShieldedCosmic Shielding Corp (exclusive Plasteel manufacturer)
Government Support
TraditionalLong DOD/NASA heritage programs
Software-HardenedNASA ISS programs
Shielded$4M Pentagon TACFI contract (Oct 2025)
Best For
TraditionalCritical military/deep space missions requiring absolute reliability
Software-HardenedCost-sensitive LEO missions with ISS-like radiation environment
ShieldedCommercial LEO constellations needing COTS performance + protection
Three approaches to protecting electronics in the space radiation environment. Sources: BAE Systems, HPE ISS experiments, Cosmic Shielding Corp, industry estimates.
Data Throughput: In-Orbit Processing vs Downlink
HPE demonstrated a 30,000x data reduction by processing in orbit. EO satellites generate 40-80 TB/day but ground contact windows are only 5-15 minutes per orbit.
Planet Labs Daily EO Data
30 TB/day
200+ Dove and SkySat satellites imaging Earth daily
Planet Labs
NISAR Daily Data Volume
80 TB/day
NASA-ISRO SAR satellite; launched Jul 30, 2025; 100 PB over 3-year mission
NASA/ISRO
Global EO Data Archive (total)
807 PB
Cumulative Earth observation data stock; growing ~100 PB/year
ScienceDirect (2023)
EO Archive Storage CO2
4,101 tonnes CO2/year
Environmental cost of storing 807 PB on ground
ScienceDirect (2023)
Ground Contact Window (LEO)
5-15 minutes per pass
Typical LEO orbit passes over ground station 6-8 times/day
Standard orbital mechanics
HPE Data Reduction Demo
30,000x reduction (2.8 GB → 92 KB)
ISS edge computing demonstration; process data in orbit, downlink results only
HPE (internal benchmark)
NASA TBIRD Optical Downlink
200 Gbps
Fastest space-to-ground laser comms; 4.8 TB in 5 min error-free (Jun 2023)
NASA Goddard, MIT LL
Starlink V3 Per-Satellite Downlink
1 Tbps
Next-gen Starlink satellites with enhanced throughput
NextBigFuture, ISPreview
Starlink Network Aggregate
~42 PB/day (estimated)
Full Starlink constellation aggregate data throughput
Industry estimate
Typical LEO Sat RF Downlink
1-10 Gbps
Standard Ka/Ku-band satellite downlink capacity
Industry standard
Data Processed On-Orbit (Ubotica)
11 missions flown
CogniSAT AI compression for ESA/NASA; reduces downlink by orders of magnitude
Ubotica
In-Orbit Processing Advantage
Eliminates 90-99% of raw data downlink
Only transmit insights/results rather than raw imagery
HPE, Ubotica, industry consensus
Key data throughput and processing metrics driving the case for in-orbit computing. Sources: NASA, Planet Labs, HPE, industry estimates.