HPE introduces a computer for lunar AI experiments aboard Astrolab’s rover

HPE is taking its space-computing program beyond the relative shelter of the International Space Station. On October 20, 2026, it introduced Spaceborne Computer-4, a computer planned for Astrolab’s FLIP lunar rover. The late-2026 mission will test whether onboard AI can interpret data and support operations in a much harsher setting.

Voyager Technologies’ Griffin-1 lander is scheduled to carry the rover near the lunar south pole under NASA’s Commercial Lunar Payload Services initiative. HPE’s announcement sets out planned experiments, rather than results: language models, image analysis and laser-based mapping will be tested alongside high-performance computing benchmarks.

An answer onboard, not another download to Earth

The practical goal is to make more use of information where the rover collects it. MIT Technology Review reports that custom-built large language models will analyze camera and lidar data, help with safe navigation and distill findings. That could reduce how much information must travel back to Earth, where communication delays and limited bandwidth constrain remote operations.

HPE describes several distinct jobs, not a single general-purpose AI pilot. Its planned language-model workloads include producing insights, analyzing visual information and supporting autonomous monitoring. Separate lidar processing will turn laser measurements into material for mapping, navigation and terrain analysis. NASA’s contributions add document search and coordination between AI models to the test program:

• Onboard documents: NASA plans to test a tuned language model that can help search and interpret documents carried on the mission.
• Coordinated models: NASA’s FLUID experiment will use a neural network trained on data from Earth, the Moon and the space station to demonstrate coordination among three AI models.
• AI deployment: An AI services container—a software package for running applications—will test reliable deployment of scalable AI services on the Moon.

Hardware survival and useful answers are different tests

HPE has already run language models in orbit. Spaceborne-2 and Spaceborne-3 ran Meta’s Llama 3.2 aboard the ISS in 2025, helping with research tasks and crew decisions. Those missions offer relevant experience, but the lunar surface removes the comparatively controlled conditions in which the earlier systems operated.

Spaceborne Computer-4 is a lightweight, self-contained, fanless module with a low-power central processor and a graphics processor. HPE says it is designed for the lunar south-pole environment. It uses solid-state storage, while the rover’s radiators will carry away excess heat—a different cooling arrangement from a conventional fan-cooled computer.

The protection strategy also relies on software rather than years spent developing specialized radiation-hardened hardware. HPE’s software watches for faults caused by charged particles. When radiation spikes, it can reduce power and speed to lower the risk to the electronics. The lunar computer retains that software-protected architecture.

Keeping the machine running does not settle whether its answers are dependable. Roser Almenar, an AI and space law researcher at the University of Valencia, told MIT Technology Review that language models have substantial potential for distant missions. She also warned that inaccurate or misleading outputs could have serious consequences in space, even when an error appears small.

A short mission after a long ground test

FLIP’s planned surface mission lasts 10 days. HPE’s Scott Wiest told MIT Technology Review that the company expects the computer to operate for that full period. The experiments are intended to establish whether the technology can function reliably in lunar conditions; that reliability is a goal of the flight, not an achieved milestone.

The lander has cleared an important preparation step. NASA says Griffin-1 completed nearly three months of environmental testing at the Jet Propulsion Laboratory in late September, including thermal-vacuum, acoustic and vibration trials. It then returned to Voyager’s Pittsburgh facility for final assembly. NASA and HPE both place launch in late 2026, with NASA identifying SpaceX’s Falcon Heavy as the rocket.