NVIDIA’s Jetson chips are heading to lunar orbit

When Firefly Aerospace touched down on the moon with its Blue Ghost Mission 1 in March 2025, it beamed back roughly 120 gigabytes of raw files. Scientists back on Earth are still chewing through that imagery and footage today.

For its Blue Ghost Mission 2, targeted for late 2026, the company is skipping the data backlog entirely. Firefly is sending an NVIDIA Jetson edge AI module straight into lunar orbit, marking the first time the platform has operated around the moon.

Edge computing goes to the moon

Deep space exploration has always run into the same brick wall: transmitting raw data back to Earth over crawl-slow radio links takes massive amounts of bandwidth and weeks of terrestrial processing.

Blue Ghost 2 is taking a split approach. A lander will detach and head down to the far side of the moon, delivering instruments including a radio telescope for a NASA-backed, UC Berkeley-led study looking for faint signals from the cosmic Dark Ages.

Meanwhile, Firefly’s Elytra spacecraft will stay up in lunar orbit for a five-year mission. Powered by solar panels, Elytra will host the Ocula imaging service using a Jetson module built by the Lawrence Livermore National Laboratory.

Instead of dumping every raw pixel back to Earth, Ocula will collect ultraviolet and visible spectrum data, run onboard AI models from Firefly's SciTec subsidiary, and beam down only the relevant takeaways in near real time.

The system is designed to map out prospective landing zones for future robotic and crewed missions, scout for lunar minerals like ilmenite, track cislunar orbital activity, and monitor ground infrastructure as more players head to the moon.

Firefly CEO Jason Kim noted that NASA is preparing about 30 robotic lander missions in the coming years. Firefly plans to upgrade future Blue Ghost flights with newer hardware, including NVIDIA's Space-1 Vera Rubin module, catering to NASA, the U.S. Space Force, and commercial resource ventures.

Why it matters

Shifting AI inference off the ground and straight into orbit fundamentally rewrites deep-space reconnaissance.

For scientists and lunar prospectors, cutting latency from months to minutes changes everything. You no longer need to wait on a massive data dump just to find out whether a snapshot contains valuable mineral signatures or clean terrain for a lander. Elytra can spot what matters autonomously.

From an industry perspective, communications bandwidth between Earth and the moon will never scale fast enough to match the flood of high-resolution sensor data coming online. The only practical solution is to stop transmitting raw noise and start computing at the collection point.

We think Firefly's strategy of packing a proven commercial edge platform into lunar orbit is a very smart shortcut, allowing them to tap into rich software ecosystems without starting from scratch. That said, the elephant in the room is long-term radiation. Whether standard Jetson architecture, even packaged by a national lab, can survive five punishing years in deep space remains the major test to watch.