| Processor | NVIDIA Jetson Orin NX or Nano | NVIDIA Jetson AGX/IGX Thor or Orin, dual cards | Clustered NxA + NxN modules | 224 NxA / NxN cards · 14 nodes |
| AI compute | Up to 157 TOPS | 4,000+ TFLOPS per module | Up to 16,000 TFLOPS distributed | Up to 896,000 TFLOPS per rack |
| Envelope | 0.5U · 50 × 100 × 100 mm | 2U · 200 × 100 × 100 mm | ESPA-class · EnduroSat FRAME-15 | Full-height rack · 16-card nodes |
| Storage | Resilient filesystems, custom EDAC | Dual NVMe per card · RAID · BTRFS/ZFS · checkpointing | Terabyte-scale, pooled across modules | Disaggregated, pooled by node BMC |
| Radiation | 50+ kRad TID, hardened by system design | 50+ kRad TID, hardened by system design | Module-level hardening + bus shielding | Card-level hardening · redundant torus paths |
| Vibration | 25 GRMS | 25 GRMS | Launch-qualified with bus | Qualified per node |
| Power | Smart arbitration, rad-hard supervisor | Smart arbitration, rad-hard supervisor | 3.4 kW peak platform power | 16 kW+ peak per rack · liquid & conduction cooled |
| Configurations | Modular · redundant · distributed | Single or dual card · hybrid Thor + Orin · redundant · distributed | Kubernetes node; constellation-ready | Local + global 2D torus · Kubernetes cluster |
| Expansion | Native NxN expansion ecosystem | NxN ecosystem via interposer | Hosted payload slots; Scanway imager | Mezzanine node-to-node interconnect |
| Readiness | TRL-9 | TRL-8 | TRL-8 | Architecture · 2026 |
| Heritage | Deimos 2024 · Phobos 2026 | Titan, Transporter-19 | Launch Q1 2027 | Same card as Titan, Phobos, Deimos |
| Best for | CubeSats, EO payloads, OBC + AI in one box | High-end ISR, SAR/HSI, multi-tenant compute | Orbital data center, telecom, CDN, SDA | Multi-rack orbital, lunar and Mars infrastructure |