Startup Wants to Run AI Inference From Space

The framing of space-based inference as an energy play rather than a latency or sovereignty play is the detail worth holding onto. Orbital Inc. is not pitching faster inference or data residency compliance, it is pitching access to unmetered solar power, which repositions this as a utilities arbitrage thesis wearing a space startup's clothing.

The energy constraint angle connects directly to the broader inference scaling pressure visible across recent coverage. NVIDIA's persistent world-generation work from early May illustrates exactly the kind of stateful, long-horizon workload that compounds energy demand well beyond a single forward pass. As inference tasks grow more complex and continuous rather than transactional, the cost per useful output rises, and the grid pressure Orbital is responding to becomes more acute. That said, none of the recent Modelwire coverage addresses space infrastructure directly, so this sits at the edge of the archive rather than at its center.

Watch whether Andreessen Horowitz follows this seed position with a larger infrastructure fund commitment within 18 months. If other top-tier firms co-invest in a Series A, that signals the energy-as-constraint thesis has cleared basic due diligence. If Orbital cannot name a hyperscaler pilot customer by end of 2027, the thesis remains a paper arbitrage.