Firefly Aerospace tests on-orbit AI with NVIDIA Jetson for first lunar deployment

Firefly Aerospace has begun operating NVIDIA’s Jetson edge AI platform in lunar orbit for the first time, a step the companies say will reduce the time and bandwidth needed to turn lunar observations into usable insights.

NVIDIA’s account points to Firefly’s earlier lunar effort as a baseline. After Blue Ghost Mission 1 landed on the moon in March 2025, Firefly downlinked nearly 120 gigabytes of raw imagery and video from onboard cameras, with scientists still processing the data later.

For the next mission, Blue Ghost Mission 2, targeted for launch in late 2026, Firefly will fly Ocula, its moon imaging service designed to run AI directly in space. Instead of sending large volumes of unprocessed data back to Earth for later analysis, Ocula will use Jetson to extract relevant findings onboard and then transmit only the most important information based on what customers need.

NVIDIA says this approach addresses a long-standing constraint in space sensing. Sensors typically generate data that must be downlinked over limited radio bandwidth, which is slow and introduces latency. Then terrestrial processing on CPU-based systems can take days or weeks to produce valuable results. By running AI inference in lunar orbit, Firefly expects to compress that cycle, producing “near real time” insights.

Ocula’s target use in Blue Ghost Mission 2 includes a science payload intended for NASA-funded research led by UC Berkeley. The mission will separate a lunar lander and descend it to the moon’s far side, carrying instruments including a radio telescope. NVIDIA ties this work to efforts to detect faint indicates associated with the cosmic “Dark Ages” shortly after the Big Bang.

Alongside the lander phase, Firefly’s Elytra spacecraft is continuing a longer-running lunar orbit mission. NVIDIA says Elytra will operate for five years, running Ocula and the Jetson-based processing chain throughout that period.

The companies describe Ocula’s imaging approach as multi-spectral, collecting imagery across ultraviolet and visible spectrum bands. NVIDIA says that data will be rapidly processed onboard and autonomously transmitted back to Earth using Jetson, with power provided by solar panels and with AI software assembled from Firefly’s SciTec subsidiary.

NVIDIA also lays out possible applications that go beyond any single mission. It says Ocula can help map lunar landing sites using high-resolution imagery with fine-grained surface detail, detect lunar mineral compositions such as ilmenite that could matter for future energy applications, and provide situational awareness of infrastructure, vehicles, and operations as more countries and companies expand on the lunar surface.

For the cislunar environment between Earth and the moon, NVIDIA says Ocula can also support object tracking and monitoring of space operations, effectively turning onboard AI into a faster decision layer for what is captured and what gets sent home.

Firefly and NVIDIA did not provide additional performance metrics in the update, such as the percentage reduction in data volumes achieved in lunar orbit, the classification accuracy of on-orbit models, or how frequently onboard outputs are refreshed. The company also did not detail whether any specific Ocula outputs for Blue Ghost Mission 1 have already been reprocessed under the new onboard-AI workflow. Those details will likely come as flight data from Elytra and the upcoming Blue Ghost Mission 2 campaign are analyzed.