Astrolab’s FLIP Rover Set to Run Large Language Models on the Moon in First-of-Its-Kind Lunar Mission

Large language models are poised to get their first taste of lunar dust. Astrolab’s next robotic mission to the Moon, built around its FLIP rover, is expected to carry a high-performance computer capable of running LLM-based software directly on the surface—something mission partners say has not been done before.

The rover, weighing about 480 kilograms, is named the FLEX Lunar Innovation Platform, or FLIP. It is the first in a lineup of lunar vehicles NASA has contracted Astrolab to provide, including future crewed “buggies” intended to transport Artemis astronauts across the Moon. FLIP is engineered for harsh conditions: it uses deformable wheels designed to handle rough, unpredictable terrain and is built to endure at least 150 hours during the bitter cold of lunar night.

Astrolab plans to send FLIP to the Moon as early as November 2026. The mission is also a science and technology showcase, with roughly 20 payloads onboard. Among them are a lidar system for 3D mapping and an instrument aimed at measuring concentrations of helium‑3—an isotope found in lunar regolith that is often discussed in the context of future energy concepts.

The computing centerpiece will come from Hewlett Packard Enterprise. HPE’s Spaceborne‑4 “edge” computer, slated to fly on FLIP, pairs a high-performance 64‑bit CPU with a GPU suited for parallel processing, bringing a style of compute architecture common in terrestrial AI systems to the lunar surface. The aim is not simply to run a chatbot in space, but to demonstrate that custom-built LLMs can operate dependably amid the Moon’s punishing environmental extremes.

Advocates argue the payoff is practical autonomy. Communication between Earth-based controllers and distant robots is constrained by latency and bandwidth, making it difficult to respond quickly and forcing spacecraft to downlink large volumes of raw data. “The time delay, for example, on Mars can be up to 20 minutes,” HPE chief scientist and principal investigator Mark Fernandez said during a press event announcing the project. While the Moon is much closer, he emphasized that even small delays can leave operators working with outdated information—and he pointed to historical cases in which time lag contributed to the loss of two Soviet lunar rovers due to operator mistakes.

On FLIP, specialized LLMs will be tested on tasks tied directly to field operations. They are expected to interpret images from onboard cameras and data from the lidar mapper to support safer navigation, and to summarize or extract insights from collected observations so the rover can transmit less data back to Earth.

Space agencies have already deployed autonomy, but generally at a more basic level. China’s Chang’e‑6 rover, which reached the Moon’s far side in 2024, used neural networks and other AI methods for navigation. NASA’s Curiosity and Perseverance rovers, meanwhile, have long driven semi-independently along preplanned routes—but the radiation-hardened computers they rely on are comparatively low-powered, closer in capability to late-1990s laptops.

HPE has been building experience in orbit. Earlier Spaceborne systems were flown to the International Space Station, and Spaceborne‑2 (2021) and Spaceborne‑3 (2024) later supported the “Space Llama” effort in 2025, running Meta’s Llama 3.2 aboard the ISS to help astronauts accelerate research work and decision-making. “As we venture further and further out into space, I believe that we’re going to rely more on AI for space exploration,” said Francisco Cordova, chief operating officer for the ISS National Laboratory, “but more importantly, working in those environments where humans simply cannot.”

Rather than waiting years for traditional radiation-hardened hardware development, HPE is leaning on protective software. The company uses monitoring tools designed to detect radiation-induced “upsets” from charged particles and, when conditions worsen, dynamically reduce power and speed to limit risk—an approach meant to keep the system stable when repairs and spare parts are not options.

Why it matters

If LLMs can function reliably on the Moon, future robots—and eventually crewed missions—could make more decisions locally, reducing dependence on slow links back to Earth. That could translate into safer navigation, faster scientific discovery, and leaner communications on lunar and deep-space missions. The next key test will be whether these models remain robust under lunar radiation, temperature extremes, and long operating periods away from human intervention.