Firefly Runs Jetson Inference in Lunar Orbit, Cuts Data Latency from Weeks to Minutes

When Firefly Aerospace’s Blue Ghost Mission 1 landed on the Moon in March 2025, it downlinked nearly 120 GB of raw imagery and video. Scientists are still processing that data over a year later. Blue Ghost Mission 2 — launching late 2026 — won’t make that mistake again.

Firefly’s Ocula moon imaging service will run NVIDIA Jetson edge AI inference directly in lunar orbit for the first time. Instead of dumping massive volumes of raw sensor data across constrained radio bandwidth, the Elytra spacecraft processes UV and visible spectrum imagery onboard, extracts the critical insights, and transmits only what customers actually need — in near real-time.

On-Orbit Inference Replaces the Slow Downlink Pipeline

Traditional space-based sensing follows a painful pipeline: collect, downlink over latent radio links, then process on terrestrial CPUs for days or weeks. Firefly’s Ocula service inverts that. The Jetson module, powered by solar panels and running Firefly’s AI software from its SciTec subsidiary, runs inference while still in lunar orbit. Data that once took months to yield results now arrives in minutes.

Jason Kim, Firefly’s CEO, put it bluntly: “We’re going to be able to do that for the first time in history.” The company is an NVIDIA Inception member, and this mission marks the first operational deployment of Jetson in lunar orbit.

What Ocula Will See and Why It Matters

Ocula’s high-resolution telescopes — built by Lawrence Livermore National Laboratory and embedded with a Jetson module — were fit-checked on Firefly’s Elytra spacecraft in April. The sensor suite covers ultraviolet and visible spectrum bands. Its on-orbit AI will:

• Map lunar landing sites with fine-grained surface detail for future human and robotic missions.
• Detect mineral compositions like ilmenite, a potential source for future energy applications.
• Provide situational awareness of infrastructure, vehicles, and operations on the Moon.
• Track objects and monitor operations in the cislunar domain — the vast region between Earth and the Moon.

Blue Ghost Mission 2 also carries a NASA-funded, UC Berkeley-led radio telescope destined for the far side of the Moon, tasked with detecting faint signals from the cosmic Dark Ages after the Big Bang. The lander separates and descends while Elytra continues orbiting for its five-year mission, running Ocula’s AI processing chain the whole time.

Building the Lunar AI Ecosystem

Firefly plans to fly Ocula sensors on subsequent Blue Ghost missions, iterating with each launch and adopting newer NVIDIA platforms — including the upcoming NVIDIA Space-1 Vera Rubin Module. Customers already span NASA, the U.S. Space Force, and commercial space, mining, and energy companies aiming to establish a permanent lunar presence.

With NASA planning roughly 30 robotic lander missions in the coming years, the cadence of on-orbit AI opportunities is accelerating. Kim’s closing line is worth remembering: “The way to predict the future is to manifest it.” Firefly is manifesting a lunar data infrastructure where all AI processing and sensing happens in space — and that changes how we build everything from landing pads to power grids on the Moon.

Source: Firefly Aerospace Operates NVIDIA Jetson in Lunar Orbit for the First Time
Domain: blogs.nvidia.com