In-Orbit Servicing Feasibility for Orbital Compute

Could robotic in-orbit servicing -- modular GPU replacement, component swaps -- extend the effective lifetime of orbital compute satellites and materially reduce amortized costs?

Answer

In-orbit servicing is speculative and unproven for LEO component-level repair. The current state of the art is limited to GEO satellite life extension through fuel transfer or station-keeping assistance. No demonstrated capability exists for LEO component-level repair or module replacement airandspaceforces-oos-2026.4. Current servicing vehicles cost $37-55M each and serve one client at a time [satnews-starfish-52m.1, satnews-starfish-52m.2] — obviously uneconomical for $3-8M compute satellites.

This page explores a speculative mature constellation servicing model that differs fundamentally from today's GEO paradigm: a fleet of reusable robotic servicing satellites carrying magazines of replacement GPU modules, with capex amortized over hundreds of visits and GPU module cargo delivered in bulk at Starship-era launch costs ($75-500/kg). Under this model, the marginal per-visit cost could reach $50-200K — two orders of magnitude below current servicing economics. However, this model depends on a chain of collectively undemonstrated assumptions (see "Critical assumptions" below): reusable servicers performing 300-500 visits, standardized robotically-accessible GPU modules, reliable robotic thermal/electrical/data connector mating in vacuum, and Starship-class cargo costs. None of these capabilities exist today, and the earliest plausible timeline for the full system is 2035-2040.

In the restructured cost model, failure costs are captured in effective lifetime (central: 4.1 years) rather than as a separate opex line. Servicing's primary impact would be extending effective lifetime by restoring capacity to degraded satellites. Under the mature servicing model's assumptions, extending central effective lifetime from 4.1 to 4.5-5.0 years would reduce amortized capex by ~18-26% — shifting the central 2040 TCO ratio from ~1.5x toward roughly 1.1-1.25x terrestrial. These projections are illustrative only — they show what mature servicing could achieve if all assumptions are met, not what is likely or expected. The numerical precision should not obscure the speculative nature of the underlying technology chain.

The most plausible near-term path remains software-managed graceful degradation (already assumed in the optimistic scenario), with mature constellation servicing as a potential second-generation improvement in the 2035+ timeframe. The two approaches are complementary: software degradation management extends effective lifetime without physical intervention, while servicing could further extend it by restoring degraded satellites to full capacity.

Analysis

Why Current GEO Servicing Economics Don't Transfer to LEO Compute

The existing in-orbit servicing industry is built around GEO satellite economics: individual assets costing $300-500M with 15-20 year design lives, where fuel depletion (not hardware failure) is the primary end-of-life driver [breakingdefense-spacelogistics-2022.1, airandspaceforces-oos-2026.3]. Paying $13M/year to extend a $400M satellite's life by 5 years is obviously economical breakingdefense-spacelogistics-2022.1.

Orbital compute satellites operate under fundamentally different economics:

The $242M minimum client satellite cost for viable servicing sciencedirect-oos-economic-value.1 exceeds LEO compute satellite costs by 30-80x. Applying current GEO servicing economics -- expendable vehicles at $37-55M each, one client per vehicle [satnews-starfish-52m.1, satnews-starfish-52m.2] -- to LEO compute satellites is the wrong reference class. It yields per-visit costs of $7-15M mccalip-space-dc.1, which is 1.4-3x the cost of full satellite replacement.

The Module-Swap Cost Problem (Current Technology)

For in-orbit servicing to work with today's technology, a servicer vehicle must:

1. Launch to the correct orbital plane (dedicated launch or rideshare)
2. Perform rendezvous and proximity operations with the target satellite
3. Dock with the target (requiring compatible interfaces)
4. Robotically remove the failed GPU module
5. Install a replacement module (with mechanical, electrical, thermal, and data connections)
6. Verify the repair and undock
7. Transit to the next target or deorbit

Steps 1-3 alone represent the bulk of current servicing vehicle capability. Steps 4-6 require robotic manipulation precision that has been demonstrated in laboratory settings sciencedirect-modular-reconfigurable-spacecraft.1 but not in orbit. The MRV/RSGS system airandspaceforces-oos-2026.1 will demonstrate robotic manipulation in GEO in 2026, but for a pre-designed MEP installation, not arbitrary component swaps.

Derived cost estimate for a near-term LEO GPU-swap servicing mission, based on current servicer pricing satnews-starfish-52m.2 extrapolated for additional robotic capability, model satellite capex, and Starship-era launch costs. Per-satellite replacement cost is $2.9M-$8.4M (manufacturing + GPU hardware + launch, from the model). McCalip's independent estimate mccalip-space-dc.1 yields a similar range:

• Servicer vehicle (Otter-class with robotic arm): $20-40M amortized over 5-10 visits = $4-8M per visit
• Replacement GPU module (100 kW_IT equivalent): $2.5-5M
• Launch cost (servicer + module to correct orbit): $0.5-2M
• Total per servicing visit: $7-15M

This is potentially 2-3x the cost of simply replacing the entire satellite ($2.9-8.4M). With current technology and economics, servicing is more expensive than replacement. This estimate applies to the current GEO-derived servicing paradigm; the mature constellation model below projects dramatically lower per-visit costs.

Mature Constellation Servicing Model

The analysis above applies current-generation servicing economics to a problem that would only arise at constellation scale in the 2035+ timeframe. The correct reference class for a mature orbital compute constellation (10,000+ satellites in similar sun-synchronous orbits) is fundamentally different from today's one-off GEO servicing missions.

The model: reusable servicing satellite fleet. Rather than expendable servicing vehicles visiting one client each, a scaled constellation would deploy a fleet of purpose-built robotic servicing satellites. Each carries a magazine of replacement GPU modules, visits dozens or hundreds of client satellites over its own multi-year operational life, and is periodically resupplied with fresh modules via bulk cargo shipments.

Key structural differences from current GEO servicing:

Per-visit cost breakdown for the mature model:

1. Servicing satellite capex amortization. A purpose-built servicing satellite (bus + robotic arm + docking mechanism + electric propulsion) costs an estimated $50-100M to build and launch. If it performs 300-500 service visits over a 5-7 year operational life (roughly one visit per week within a single orbital plane), the amortized capex per visit is $100-330K.

2. GPU module cargo cost. The replacement module itself -- GPU hardware only, no bus, no solar arrays, no radiators -- might mass 5-10 kg/kW_IT (compute hardware plus packaging for robotic handling). At Starship-era launch costs of $75-500/kg and delivered in bulk cargo shipments, the launch cost per module is $375-5,000/kW_IT. For a module serving ~10-50 kW_IT of compute, this is $4K-250K per module. The hardware cost of the module itself (GPU + rad-hardened packaging) adds $2,500-5,000/kW_IT, or $25K-250K per module.

3. Propellant per visit. Within a single orbital plane, phasing maneuvers between nearby satellites require minimal delta-v (tens of m/s). With efficient electric propulsion, propellant cost per visit is negligible relative to other costs.

4. Resupply operations. Servicing satellites are periodically resupplied with fresh GPU module magazines via cargo carriers launched as Starship rideshare payloads. At $75-500/kg, delivering a batch of 10-20 modules costs $50K-500K, amortized across the batch.

Estimated marginal per-visit cost: $50-200K in the mature model. This is two orders of magnitude below the $7-15M estimated for current-technology servicing, and well below the $2.9-8.4M cost of full satellite replacement. The economics work because the servicing satellite's fixed cost is spread over hundreds of visits, and the variable cost per visit is dominated by the GPU module hardware (not the servicing operation itself).

Critical assumptions this model requires:

• Satellites designed from inception with standardized, robotically-accessible GPU module bays
• Module interfaces that support reliable robotic connection/disconnection of electrical, data, and thermal paths in vacuum
• Servicing satellites capable of hundreds of autonomous docking cycles without failure
• A constellation large enough (thousands of satellites per orbital plane) to keep servicing satellites busy
• Starship-class launch costs ($75-500/kg) for bulk cargo delivery

These assumptions are individually plausible for a 2035-2040 timeframe but collectively undemonstrated. The model is speculative engineering extrapolation, not proven capability.

Impact on the Orbital Compute Business Case

Important caveats for all scenarios below:

1. The scenarios use simple division (capex / effective_lifetime) for illustrative purposes. The main TCO model uses CRF-based amortization at 15% WACC, which yields higher annual costs — the servicing-adjusted ratios below are approximate and would be somewhat less favorable with full CRF treatment.
2. Every scenario below depends on the mature constellation servicing model's full assumption chain being met (reusable servicers with 300-500 visit lifetimes, standardized robotic GPU module interfaces, Starship-era cargo costs). These are individually plausible for 2035-2040 but collectively undemonstrated. No robotic component swap has been performed in orbit airandspaceforces-oos-2026.4, and autonomous module exchange has TRL 3-4 sciencedirect-modular-reconfigurable-spacecraft.1.
3. The numerical estimates should be read as "if all assumptions hold, then..." — not as projections of likely outcomes.

In the restructured cost model, satellite failure costs are captured in effective lifetime rather than as a separate opex line item. Central effective lifetime is 4.1 years, reflecting the capacity-weighted impact of bus failures, GPU attrition, and SDC overhead over a nominal 5-year physical lifetime (derived from a structured reliability model with separate terms for each mechanism). Servicing's primary impact would be extending this effective lifetime by restoring capacity to degraded satellites rather than requiring full replacement.

How servicing extends effective lifetime. A satellite whose GPUs degrade from radiation damage loses compute capacity over time. Without servicing, this degradation reduces the capacity-weighted effective lifetime below the physical lifetime. With servicing, failed or degraded GPU modules can be replaced in-situ, restoring the satellite to near-full capacity and potentially extending useful operation beyond the original physical lifetime (bus and power systems permitting).

Scenario A: No servicing (current model). Effective lifetime remains at central 4.1 years. Central 2040 amortized capex: 49,122 / 4.1. Central 2040 TCO ratio: ~1.5x.

Scenario B: Near-term servicing at current-technology costs ($3-5M/visit). This scenario uses purpose-built but first-generation servicing infrastructure (reusable vehicles with 20-50 visits each, co-orbital module depots). Per-visit cost of $3-5M compares to full satellite replacement at ~$5M, so the savings per incident are modest. The main benefit is extending the effective lifetime of satellites whose bus, solar arrays, and thermal systems are still functional but whose GPUs have degraded. If this extends effective lifetime from 4.1 to ~4.2-4.5 years, amortized capex falls by 8-15%. Central 2040 TCO ratio: ~1.7-1.8x.

Scenario C: Mature constellation servicing ($50-200K/visit). With marginal servicing costs far below replacement cost, the operator services every degraded satellite as soon as GPU replacement is worthwhile. GPU modules are replaced perhaps 2-3 times over a satellite's physical lifetime, keeping compute capacity near 100% until the bus, solar arrays, or thermal system reaches end of life. If this extends effective lifetime from 4.1 to 4.5-5.0 years:

• Amortized capex at 4.5 years: 49,122 / 4.5 = ~$12,891/kW_IT/year (vs ~$15,260 at 4.1 years)
• Amortized capex at 5.0 years: 49,122 / 5.0 = ~$11,602/kW_IT/year
• Reduction in amortized capex: ~18% (at 4.5 years) to ~26% (at 5.0 years)
• Adding fixed opex of $200/kW_IT/year plus modest servicing opex (~$100-300/kW_IT/year for module costs amortized across the fleet):
  • TCO at 4.5 years: ~$13,191-$13,391/kW_IT/year
  • TCO at 5.0 years: ~$11,902-$12,102/kW_IT/year
  • Central 2040 terrestrial TCO: ~10,698/kW_IT/year
  • TCO ratio: ~1.23-1.25x (at 4.5 years) to ~1.11-1.13x (at 5.0 years)

Scenario D: Theoretical maximum (servicing enables effective lifetime equal to terrestrial GPU depreciation period of 5 years, at negligible marginal servicing cost). Amortized capex: 49,122 / 5.0 = ~$11,602/kW_IT/year. TCO: ~$11,802/kW_IT/year. TCO ratio: ~1.10x. This scenario is maximally speculative — it requires all mature servicing assumptions plus negligible per-visit costs plus no design-for-servicing mass penalty.

Assessment of the ~1.1-1.25x range: The improvement from ~1.5x to ~1.1-1.25x would be substantial if achieved, but it depends on a speculative technology chain that has no demonstrated precedent. The entire chain — from expendable GEO station-keeping (today's state of the art) to reusable LEO robotic module swap at scale — represents multiple TRL jumps across different subsystems. Combined with the simplification of using capex/lifetime division rather than full CRF math, the ~1.1-1.25x figures should be understood as an optimistic boundary on what mature servicing could theoretically deliver, not as an expected outcome. A more conservative reading is that servicing could narrow the central TCO ratio by 15-30%, significant but not transformative, and only in the 2035+ timeframe after substantial capital investment in servicing infrastructure.

Mass and Complexity Penalties of Serviceable Design

Designing satellites for in-orbit servicing imposes penalties that partially offset servicing savings:

• Standardized interfaces add mass (connectors, grapple fixtures, alignment mechanisms): estimated 5-10% increase in satellite dry mass
• Module isolation requires redundant power/data buses and fault-tolerant boundaries between swappable and non-swappable components
• Thermal interface management is particularly challenging: GPU modules must have separable thermal connections to the radiator system while maintaining adequate conductivity during operation
• Structural rigidity requirements increase when modules must survive robotic removal forces

These penalties increase satellite capex by an estimated 5-15%, partially offsetting the lifetime extension benefit. However, in the mature constellation model, where the per-visit servicing cost is very low ($50-200K), the capex penalty is more than compensated by the lifetime extension: a 10% capex increase with an 18-32% increase in effective lifetime (4.1 to 4.5-5.0 years) yields a net reduction in amortized capex of 7-19%.

The Software-Managed Degradation Alternative

Sophia Space's TILE architecture [sophia-space-seed.1, aiaa-sophia-space.1] points toward an alternative approach that avoids the servicing cost problem entirely: treat individual compute units as expendable and manage degradation through software.

This approach:

• Routes around failed GPUs without physical intervention (SOOS redistributes workload)
• Accepts graceful capacity decline over the satellite lifetime rather than maintaining peak capacity
• Replaces entire satellite batches on a scheduled cycle (every 5-6 years) rather than repairing individual failures
• Avoids serviceable-design mass penalties (no robotic interfaces, no module bays)

The optimistic scenario in the current cost model already assumes effective software degradation management (5.0-year effective lifetime). Software degradation management and physical servicing are complementary, not competing: software management is the near-term approach (available with first-generation satellites), while physical servicing is a potential second-generation enhancement that could further extend effective lifetime by restoring degraded satellites to full capacity rather than merely routing around failures.

Timeline for LEO Compute Servicing Capability

The 2035+ timeline for commercial LEO compute servicing means it cannot affect the initial orbital compute deployment window. However, a constellation designed for servicing from inception could begin benefiting from servicing infrastructure as it matures, potentially improving the economics of second- and third-generation satellite replacements.

Key Uncertainties and Blockers

1. No demonstrated autonomous module swap in orbit. The most advanced robotic servicing capability (MRV/RSGS) targets pre-designed MEP installation on cooperative GEO satellites, not arbitrary component replacement on LEO compute platforms [airandspaceforces-oos-2026.1, airandspaceforces-oos-2026.4].

2. Orbital plane matching is less costly for concentrated constellations. Unlike current GEO servicing where targets are scattered, a compute constellation with thousands of satellites per orbital plane reduces inter-visit delta-v to small phasing maneuvers. This fundamentally changes the visit-rate economics: a servicing satellite could realistically visit one satellite per week within a single plane, enabling the high visit counts (300-500 per lifetime) that the mature model requires.

3. Thermal interface separation and reconnection in vacuum is unsolved. GPU modules require high-conductivity thermal paths to radiators. Creating separable thermal interfaces that maintain adequate conductivity after multiple connect/disconnect cycles in vacuum, subject to thermal cycling from -150C to +80C, is an open engineering challenge militaryembedded-modular-gpu.1.

4. Servicing satellite reliability at scale is unproven. The mature model assumes a servicing satellite can perform hundreds of autonomous docking-undocking-module-swap cycles over a multi-year lifetime without failure. Current rendezvous and docking technology has been demonstrated for single missions, not sustained high-tempo operations.

5. The business case competes with improving alternatives. Launch costs are declining rapidly. By the time servicing infrastructure matures, full-satellite replacement may cost $1-2M (at $100/kg launch), narrowing the cost advantage of servicing. However, servicing's value is not just cost per replacement -- it also avoids the downtime and capacity loss of deploying a replacement satellite, and it extends the useful life of the non-GPU components (bus, solar arrays, thermal system) which may still be functional.

6. Serviceable design penalties reduce the net benefit. The 5-15% capex increase from serviceable design partially offsets the lifetime extension, though the net effect is still positive in the mature model.

7. Regulatory framework for LEO servicing is undefined. No FCC or international framework exists for autonomous robotic operations on third-party assets in LEO. However, a constellation operator servicing its own satellites faces fewer regulatory barriers than third-party servicing spacenews-oos-road-to-market.2.

Evidence

Current State of In-Orbit Servicing Technology

E1. Northrop Grumman's MEV-1 docked with Intelsat 901 in February 2020 -- the first commercial satellite life extension mission. MEV-2 docked with Intelsat 10-02 in April 2021. Both operate in GEO, providing station-keeping by physically attaching to the client satellite's engine nozzle. Neither performs component repair or module replacement. Intelsat pays approximately $13M/year for MEV service, or roughly $65M for a 5-year life extension -- compared to $300-500M to build and launch a replacement GEO satellite. -- breakingdefense-spacelogistics-2022, bcsatellite-mev

E2. Northrop Grumman is transitioning from MEV to two new platforms: the Mission Robotic Vehicle (MRV), equipped with a DARPA-funded robotic arm (the RSGS Integrated Robotic Payload developed by the Naval Research Laboratory), and the smaller Mission Extension Pod (MEP), approximately the size of a dishwasher, which provides 6 years of electric propulsion for station-keeping. MRV can carry and install multiple MEPs. The MRV/RSGS system launches in 2026. -- airandspaceforces-oos-2026, breakingdefense-spacelogistics-2022

E3. Four U.S. government-funded on-orbit servicing missions are planned for 2026, all targeting GEO: (1) SpaceLogistics MRV with RSGS robotic arm (DARPA), (2) Astroscale U.S. Refueler for first U.S. hydrazine refueling in GEO (Space Force), (3) Tetra-5 for autonomous rendezvous/docking/refueling (Air Force Research Lab), (4) Kamino hydrazine transfer vehicle (Defense Innovation Unit). -- airandspaceforces-oos-2026

E4. Starfish Space's Otter satellite-servicing vehicle launched to LEO on SpaceX Transporter 14 in June 2025 (Otter Pup 2 mission) to attempt the first commercial docking in LEO. The Otter is an ESPA-class vehicle (~200 kg), approximately 10x smaller than traditional servicers, using electric propulsion with autonomous rendezvous (CETACEAN computer vision, CEPHALOPOD guidance, Nautilus universal capture mechanism for non-equipped satellites). Three operational Otters scheduled for 2026 launch. -- space-com-starfish-otter, satnews-starfish-52m

E5. The U.S. Space Force awarded Starfish Space $52.5M for "Deorbit-as-a-Service" for the Proliferated Warfighter Space Architecture (PWSA) LEO constellation -- the first operational LEO deorbit services contract. A separate $54.5M contract funds a second Otter for GEO operations, with 2028 delivery. An earlier $37M contract funded the first GEO Otter. Total contract value across three Otter vehicles: ~$144M. -- satnews-starfish-52m, breakingdefense-starfish-otter-2

E6. NASA's OSAM-1 (On-orbit Servicing, Assembly, and Manufacturing 1) was cancelled in March 2024 due to "continued technical, cost, and schedule challenges." OSAM-2 (Archinaut, for in-orbit 3D printing and assembly) was concluded in 2023 without a flight demonstration. These cancellations reflect the difficulty of demonstrating autonomous robotic manipulation in space even with government funding. -- nasa-osam-1, wikipedia-osam-1

E7. ESA's EROSS IOD (European Robotic Orbital Support Services In-Orbit Demonstrator), led by Thales Alenia Space, targets a 2026 demonstration mission. It will validate rendezvous, capture, docking, refuelling, and payload exchange between two cooperative spacecraft. The ASCEND study (space data centers) explicitly plans to use EROSS robotic technology for modular assembly of orbital data center infrastructure, with a proof-of-concept by 2031 and initial deployment by 2036. -- thales-ascend, cordis-eross-iod

Economics of LEO Servicing vs. Replacement

E8. SpaceLogistics' Rob Hauge stated that the GEO servicing market exists because "every year about 10 to 20 [GEO satellites] reach their end of life because they run out of fuel." GEO satellites cost $300-500M to build and launch, making $13M/year servicing fees economically rational. The LEO market lacks equivalent economics: individual LEO satellites cost $250K-$1.2M to build, with launch costs of $100K-$3M each. -- airandspaceforces-oos-2026, motley-fool-starlink-replacement

E9. SpaceNews analysis (2024) concluded that for LEO constellations, "the low cost of launching satellites has reached a point where the risk and cost of OOS operations do not justify the benefits." Operators find it "more practical and economical to launch new, more advanced satellites" rather than service existing ones. LEO satellites have 3-5 year lifespans and ~$500K manufacturing costs, making fleet renewal the dominant strategy. -- spacenews-oos-road-to-market

E10. Acta Astronautica analysis (2020) found that on-orbit servicing is commercially viable when the client satellite costs more than $242M and the servicing architecture costs less than $140M. Below these thresholds, replacement is more economical. LEO compute satellites at $3-8M each fall well below the $242M viability threshold. -- sciencedirect-oos-economic-value

E11. Starfish Space Otter vehicles cost $37-54.5M per spacecraft (based on Space Force contract values), excluding launch costs. Each Otter performs limited missions (deorbit, relocation, station-keeping) before requiring replacement itself. For component-level servicing (module removal and installation), additional robotic manipulation hardware would increase vehicle mass, complexity, and cost. -- satnews-starfish-52m, breakingdefense-starfish-otter-2

E12. See "The Module-Swap Cost Problem" in the Analysis section below for a derived cost estimate of near-term LEO servicing visits, based on current servicer pricing [satnews-starfish-52m.2] and Starship-era launch costs.

Modular Satellite Architecture for Compute

E13. Sophia Space's TILE platform uses 1m x 1m x 1cm modular compute slabs, each with integrated solar panels and passive radiative cooling. Sophia envisions three tiers: single tiles on host spacecraft, clusters of ~40 tiles, and full-scale orbital data centers of ~2,500 tiles. The company plans to "launch, maintain, and eventually replace" arrays, with hardware refreshed approximately every 6 years. The Sophia Orbital Operating System (SOOS) routes around failed tiles without human intervention. -- sophia-space-seed, aiaa-sophia-space

E14. Sophia Space's architecture treats individual tiles as expendable. When a tile fails, SOOS redistributes workload across remaining tiles rather than attempting physical repair. The 30-year orbital data center lifecycle is achieved through periodic launch of replacement tile batches, not in-orbit repair of individual tiles. This is architecturally closer to terrestrial hyperscaler practice (replace failed servers) than to traditional satellite servicing. -- aiaa-sophia-space

E15. The ASCEND feasibility study envisions "modular space infrastructures assembled in orbit using robotic technologies" from EROSS IOD, with a timeline of: robotic demo 2026, proof-of-concept 2031, initial deployment 2036, and 1 GW scale before 2050. This is the most concrete roadmap for robotically-assembled orbital compute infrastructure, but it targets initial assembly, not ongoing component-level maintenance. -- thales-ascend

E16. Current TRL for autonomous module swap in orbit is 3-4 (laboratory validated only). No flight demonstration of autonomous compute module replacement has been conducted. — sciencedirect-modular-reconfigurable-spacecraft

Servicing Market and Timeline

E18. The on-orbit satellite servicing market was valued at $2.4B in 2023 and is projected to reach $5.1B by 2030 (11.5% CAGR). GEO dominates the addressable market. Active debris removal and orbit adjustment are the fastest-growing segments. Key companies: Maxar Technologies, Astroscale, SpaceLogistics (Northrop Grumman), Airbus, Thales Alenia Space. -- marketsandmarkets-oos

E19. SpaceNews analysis projects that commercial in-orbit servicing will emerge "primarily supported by governmental institutions" rather than through pure market demand. In-orbit assembly and maintenance services are characterized as "emerging only in the long term, driven by large-scale, expensive infrastructure projects." -- spacenews-oos-road-to-market

E20. Current in-orbit servicing capabilities are limited to: life extension via station-keeping (MEV-1, MEV-2), planned fuel transfer (Astroscale, Tetra-5, Kamino), planned robotic inspection and MEP installation (MRV/RSGS), and planned deorbit services (Starfish Space Otter). No demonstrated capability exists for: component-level repair, module removal and replacement, or electrical/data connector mating in orbit. -- airandspaceforces-oos-2026, spacenews-oos-road-to-market