Orbital Power Usage Effectiveness

What is the orbital PUE — the ratio of total satellite power to IT load power?

What is the PUE for compute-focused satellites in LEO?

Answer

Orbital PUE ranges from 1.035 (optimistic) to 1.10 (conservative), with a central estimate of 1.05. This is dramatically lower than terrestrial PUE (1.10-1.20 for modern liquid-cooled AI facilities; legacy air-cooled fleets averaged ~1.4) because the dominant component of terrestrial overhead — cooling — is handled passively via radiative heat rejection at zero power cost.

Confidence: Medium. The DC-DC conversion efficiency range (91-97%) is sourced from three independent sources: Vicor's space-grade converter specifications (91-96%) [vicor-newspace-dc-dc.1-.4], VPT's GaN converter (95%) vpt-epc-gan-sgrb.1, and NASA's SOA survey of flight-proven PMAD systems (86-98.5%, most 93-97%) nasa-soa-power-2021.1. The housekeeping budget is cross-checked against JWST as-measured subsystem power jwst-design-2023.1 and NASA-affiliated allocation guides uah-spacecraft-design-101.1. The 1.05 central estimate is a plausible component-level estimate, still unvalidated at system scale — no orbital datacenter has been built and measured.

The overhead has two components: (1) DC-DC power conversion losses (2.5-6% of IT load), the largest contributor, where modern space-grade converters achieve 91-97% efficiency per stage; and (2) bus housekeeping subsystems (0.5-2% of IT load), covering ADCS, communications, C&DH, thermal pumps, and station-keeping (orbit-averaged). For a 100 kW_IT satellite, these collectively consume 0.4-2.1 kW.

Analysis

Power conversion chain

The largest contributor to orbital PUE is DC-DC conversion. A compute satellite must convert solar array voltage (28-100V bus) to GPU load voltage (~0.7-1.0V core). The conversion chain:

Stage 1: Solar array regulator to bus. Sequential switching shunt regulators (S3R) or direct energy transfer architectures regulate the bus. Vicor's space-grade PRM2919 buck-boost regulator achieves 96% efficiency vicor-newspace-dc-dc.3. S3R topologies achieve similar performance. A direct energy transfer architecture minimizes this loss to 2-4%.

Stage 2: Bus to load voltage. Modern space-grade DC-DC converters achieve 91-96% efficiency [vicor-newspace-dc-dc.2, vicor-newspace-dc-dc.4], with ZVS/ZCS topologies (Sine Amplitude Converter) at the upper end. Vicor's BCM3423 bus converter achieves 94% at the 100V→33V step vicor-newspace-dc-dc.2, and point-of-load VTMs achieve 91-93% for the final step down to GPU core voltage (0.42-4V) vicor-newspace-dc-dc.4. VPT's GaN-based SGRB12028S achieves 95% at 120V→28V with 100 krad hardening vpt-epc-gan-sgrb.1. NASA's SOA survey confirms the range: flight-proven PMAD systems span 86-98.5% efficiency, with most in the 93-97% range nasa-soa-power-2021.1. The range across existing space-qualified buck regulators is 67-95% vicor-newspace-dc-dc.1, with newer ZVS/ZCS and GaN modules at the upper end.

A compute satellite is unique: 95%+ of power goes to a single load type at a single voltage, enabling purpose-designed bus voltage matching and potentially eliminating unnecessary conversion stages. The system-level DC-DC conversion efficiency depends on the number of stages and the topology chosen:

These system efficiencies reflect the end-to-end conversion chain from solar array bus to GPU core voltage. The optimistic case assumes a purpose-designed bus voltage that enables a single conversion stage, drawing on the upper end of flight-proven PMAD efficiencies (93-98.5%) nasa-soa-power-2021.1. The central case reflects a well-designed but more conventional architecture using modern ZVS or GaN converters (Vicor PRM2919 at 96% vicor-newspace-dc-dc.3, VPT GaN at 95% vpt-epc-gan-sgrb.1). The conservative case assumes a multi-stage chain with legacy-topology modules (Vicor VTMs at 91-93% vicor-newspace-dc-dc.4) where stage efficiencies compound: e.g., two stages at 94% and 91% yield ~85.5% combined — but a purpose-designed compute satellite would avoid such inefficient chains; the ~93% conservative estimate reflects a realistic floor for a cost-optimized system.

Housekeeping power budget

For traditional spacecraft, bus subsystems consume a substantial fraction of total power: the NASA-affiliated subsystem power allocation guide shows attitude control at 11-28%, C&DH at 13-19%, and communications at 0-30% of total bus power depending on mission type uah-spacecraft-design-101.1. JWST's as-measured bus power confirms these magnitudes: ACS 184W, C&DH 140W, comms 170W, propulsion 118W — fixed-power subsystems totaling ~610W excluding thermal, harness, EPS, and deployment control jwst-design-2023.1. A compute satellite is fundamentally different — the IT payload dominates power (>95%), and these fixed bus subsystem power draws become a negligible fraction of total at 100 kW scale.

For a 100 kW_IT satellite in sun-synchronous LEO:

Combined PUE

Orbital PUE = (IT_load + DC-DC_loss + housekeeping) / IT_load

Power conversion dominates: at the central case, DC-DC losses account for 80% of total overhead. At the conservative case, the ~93% system conversion efficiency drives the majority of the PUE increase. Improving conversion efficiency from 96% to 97% reduces PUE from 1.05 to ~1.04, a larger effect than eliminating all thermal pumps.

Scale dependence

Housekeeping overhead as a fraction of IT load decreases with satellite power — a 50 kW_IT satellite would have PUE ~1.07, while a 200 kW_IT satellite would see ~1.04, because the same ADCS/comms hardware serves a larger compute payload. The model assumption of ~100 kW_IT is representative.

Validation

The 1.05 central estimate is a plausible component-level figure derived from converter datasheets and spacecraft bus power data. However, it remains unvalidated at system scale: no orbital datacenter has been built and measured, and the complete sourcing failure during the original version of this page (see Sourcing note below) should temper confidence in exact calibration. Integration losses, thermal derating, and degradation over mission life could shift the real value.

Sourcing note

Source accuracy reviews (2026-03-24) removed all seven original evidence items from this page as misattributed or fabricated. The page has been re-sourced with verifiable evidence from three independent DC-DC converter sources (Vicor [vicor-newspace-dc-dc.1-.4], VPT/EPC Space vpt-epc-gan-sgrb.1, and NASA's SOA PMAD survey nasa-soa-power-2021.1) plus two spacecraft power budget references (JWST as-measured data jwst-design-2023.1 and NASA-affiliated allocation guides uah-spacecraft-design-101.1). The replacement sources support the PUE range at the component level, but the complete sourcing failure of the original page warrants caution: the 1.05 central value is a plausible component-level estimate, not a confirmed system-level measurement. Confidence in exact calibration should remain moderate until validated by an integrated system design or flight data.

Evidence

DC-DC converter efficiency

1. Vicor reports that existing space-qualified buck regulators specify efficiencies in the range of 67-95%, and forward/flyback DC-DCs specify 47-87%. — vicor-newspace-dc-dc

2. Vicor's BCM3423PA0A35C0S, a 300W space-grade SAC bus converter (94-105V input, 31-35V output at 1/3 ratio), achieves 94% maximum ambient efficiency. Uses zero-voltage/zero-current switching topology. — vicor-newspace-dc-dc

3. Vicor's PRM2919P36B35B0S, a 200W space-grade non-isolating ZVS buck-boost regulator (30-36V input, 13.4-35V adjustable output), achieves 96% maximum ambient efficiency. — vicor-newspace-dc-dc

4. Vicor's VTM2919 step-down modules for point-of-load delivery: the VTM2919P32G0450S (16-32V to 2-4V) achieves 93% efficiency; the VTM2919P35K01A5S (13.4-35V to 0.42-1.1V, suitable for GPU core voltage) achieves 91% efficiency. — vicor-newspace-dc-dc

5. VPT's SGRB12028S DC-DC converter, using EPC Space GaN technology, achieves up to 95% efficiency at 120V input, 28V/400W output. Radiation hardened to 100 krad(Si) TID and 85 MeV/mg/cm2 SEE. VPT supplies high-reliability power solutions to organizations including NASA, ESA, Lockheed Martin, Boeing, BAE Systems, and Thales. — vpt-epc-gan-sgrb

6. NASA's State-of-the-Art Small Spacecraft Technology report (2021) lists flight-proven (TRL 9) PMAD systems with maximum efficiencies: Pumpkin EPSM 1 at 98.5%, AAC Clyde Space Starbuck Micro at 97%, GomSpace P31U at 96%, ISISPACE iEPS Type C at 95%, DHV EPS Module at 93%, EnduroSat EPS I at 86%. Notes that GaN-based field effect transistors yield "improvements in overall efficiency" due to lack of gate oxide layer, and enable higher switching rates and lower switching losses improving SWaP metrics — though the report cautions that GaN-based PMAD options have drawbacks including high complexity of control circuitry and lack of flight heritage. — nasa-soa-power-2021

Spacecraft subsystem power allocation

7. NASA-affiliated spacecraft design reference (UAH) provides a subsystem power allocation guide. For "Other" missions (non-comsat, non-metsat): thermal control 33%, attitude control 11%, power subsystem 2%, CDS (command & data) 15%, communications 30%, propulsion 4%, mechanisms 5% of total bus power. These allocations represent traditional spacecraft where the bus subsystems are the primary consumers; for a compute satellite where the payload (GPUs) dominates power, bus subsystem power is a small fraction of total. — uah-spacecraft-design-101

8. JWST's as-measured spacecraft bus power during normal operations: ACS 184W, C&DH 140W, Telecommunications 170W, Thermal Control 437W, Propulsion 118W, EPS 64W, Harness losses 227W, Deployment Control 29W. Total bus: 1,369W out of 2,029W observatory total, with 660W to science payload. These fixed bus subsystem power draws (totaling ~705W excluding thermal and harness) would represent <1% overhead on a 100 kW compute payload. — jwst-design-2023