The "Vacuum-Ready" Supply Chain: Scaling 800V–10kV SiC for Orbital AI Clusters

The material properties that make SiC superior to silicon in terrestrial data centers become existential advantages in orbit. SiC's 3x thermal conductivity over silicon means heat moves from junction to package surface faster — critical when the only thermal path is radiative dissipation. SiC's inherently smaller die sizes at equivalent power ratings reduce payload mass. And SiC's voltage stability at 800V+ eliminates the multi-stage conversion losses that consume power budget in silicon-based orbital power systems.

The radiation hardness argument is the one that separates terrestrial and orbital applications entirely. SiC's bandgap of 3.26 eV versus silicon's 1.12 eV translates directly to a higher threshold for Single Event Burnout (SEB) — the catastrophic failure mode triggered by cosmic ray particle strikes. In LEO, cosmic ray flux is 100–1,000x higher than at sea level. In GEO or deep space, it is orders of magnitude worse. Silicon power devices at 800V+ require heavy shielding or aggressive derating that negates their cost advantage. SiC devices can operate at rated voltage with standard space-grade derating protocols.

Wolfspeed's CPM3-10000-0300A — the industry's first commercially available 10kV SiC MOSFET, released March 2026 in bare die format — enables direct medium-voltage conversion in orbit. A single 10kV SiC device can convert solar array output directly to distribution bus voltage without the intermediate conversion stages that silicon topologies require. The TDDB lifetime of 158,000 years at continuous 20V gate bias provides the long-term reliability that 5–7 year orbital missions demand. Wolfspeed's resolution of bipolar degradation at 10kV — a failure mode that had blocked every prior 10kV SiC attempt — is the specific engineering breakthrough that opens this voltage tier for space applications.

On Earth, a 1.2% efficiency improvement in a 1GW data center saves money. In orbit, it saves mass. Every watt of waste heat that cannot be conducted or convected away must be radiated — and radiative cooling follows the Stefan-Boltzmann law (power radiated proportional to T⁴), which means the radiator infrastructure required scales non-linearly with thermal load. In a 1GW orbital cluster, 1.2% efficiency gain equals 12MW of thermal load eliminated — which translates to thousands of square meters of radiator panels not launched, not deployed, not maintained.

This is where SiC's efficiency advantage transitions from a cost argument to an existential one. Wolfspeed's SiC power stages achieve 99%+ conversion efficiency at rated load, compared to 94–96% for silicon IGBT equivalents at the same voltage tier. That 3–5 percentage point delta, which represents millions in opex savings on Earth, represents the difference between a thermally viable orbital architecture and one that collapses under its own waste heat. At 1GW scale with silicon power conversion at 95% efficiency, you generate 50MW of waste heat requiring approximately 125,000 m² of radiator surface at 400W/m². At 99% SiC efficiency, that drops to 10MW and 25,000 m². The radiator mass savings alone could exceed 500 metric tons.

The switching speed advantage compounds this. Wolfspeed's 10kV MOSFET switches at up to 10kHz with <10ns rise time — 16x faster than equivalent silicon IGBTs. Higher switching frequency reduces the size and mass of magnetic components (inductors, transformers) by approximately 300%, which in an orbital context directly converts to additional compute payload per launch.

In terrestrial data centers, the physical size and weight of power electronics is a facility design consideration. In orbital data centers, it is a launch cost multiplier. Every kilogram of power conversion hardware that rides a Starship to LEO displaces a kilogram of compute hardware — GPUs, TPUs, memory, networking. The SWaP-C (Size, Weight, and Power-Cost) optimization equation inverts the terrestrial value hierarchy: mass reduction in power electronics has a direct, linear relationship to compute capacity deployed per dollar of launch cost.

SiC's advantage here is quantifiable. At the same power rating, SiC power stages are 3–5x lighter than silicon equivalents. The mechanism is switching frequency: Wolfspeed's 10kHz+ switching (versus 600Hz for equivalent silicon IGBTs) reduces the required size of passive magnetic components — inductors and transformers — by approximately 300%. Magnetic components typically constitute 40–60% of power stage mass. Reducing them by 3x while maintaining the same power throughput creates a compounding mass advantage that scales with system size.

SpaceX's Starship can deliver approximately 100–150 metric tons to LEO per launch. At a projected cost of $200–500/kg by 2030 (current best estimates for fully reusable Starship operations), every kilogram saved on power electronics represents $200–500 of compute infrastructure that can be launched instead. For a 1GW orbital cluster requiring ~10MW of power conversion infrastructure, the mass difference between SiC and silicon topologies could exceed 50 metric tons — equivalent to $10–25 million in additional compute payload per constellation, or an entire additional Starship launch avoided.

An orbital data center is not a satellite. It is a distributed compute cluster connected by inter-satellite optical links operating at orbital velocities (>7 km/s relative to ground, 0–15 m/s relative to cluster neighbors). Google's Project Suncatcher architecture calls for clusters of approximately 81 satellites flying within 200m to 1km of each other, connected by optical inter-satellite links tested at 1.6 Tbps in the lab. The power and precision requirements for these links create a distinct SiC opportunity that has no terrestrial analog.

Each inter-satellite laser link requires precision power delivery with sub-nanosecond stability. Beam pointing accuracy at 1km range with 1.6 Tbps coherent optical modulation demands power supply ripple below millivolt levels — any power fluctuation translates to beam wander that degrades bit error rate. SiC's fast switching capability (<10ns rise time on Wolfspeed's 10kV MOSFET) enables the pulse shaping and ripple suppression needed for coherent optical communication power supplies. Silicon IGBTs, with rise times of 100–200ns, cannot achieve the power supply bandwidth required for next-generation optical links.

Marvell's Ara T — the first 1.6 Tbps PAM4 optical DSP using 8x200G transmit-retimed optics — represents the signal processing layer that must be powered by these SiC-driven supplies. While Marvell's terrestrial product targets data center switch-to-switch interconnects, the underlying DSP technology is directly applicable to space optical communications. The power efficiency of the DSP itself (~15 pJ/bit at 1.6 Tbps) means the power supply architecture around it becomes the determining factor in overall link power budget.

A satellite launched from the surface transitions from 1 atmosphere to hard vacuum in approximately 3 minutes. During this ascent, every electrical connection on the spacecraft passes through the pressure regime described by Paschen's law — where arcing voltage reaches its minimum at intermediate pressures (~1 Torr for most gas compositions). Mechanical switches, relays, and spark-gap protection devices that function reliably at sea level or in vacuum can arc catastrophically during ascent at pressures between 0.1 and 10 Torr. This is not a theoretical risk; it is a well-documented failure mode in launch vehicle avionics.

SiC solid-state switches eliminate this failure mode entirely. A solid-state power switching topology based on Wolfspeed's 10kV MOSFET has no physical contact gap, no gas-filled chamber, and no pressure-dependent breakdown characteristic. The device switches identically at 1 atm, at 1 Torr, and in hard vacuum. The <10ns rise time enables solid-state power switching that is impervious to pressure changes — the only reliable power path from ground through upper atmosphere to orbit.

Beyond launch survival, solid-state power switching using SiC is essential for orbital power management. Satellite solar arrays must be switched, regulated, and fault-isolated as they cycle between sunlight and eclipse every 90 minutes in LEO. Each eclipse transition requires the power bus to switch from solar to battery and back within milliseconds — a task that SiC solid-state switches handle without the contact erosion, bounce, and arc damage that degrade mechanical switches over thousands of cycles across a 5–7 year mission life.