src=”https://techcrunch.com/wp-content/uploads/2026/06/coast_ground_postprocess.png” />
AI Autonomy Achieved in Earth Observation Orbit
A significant technological advancement has occurred in the domain of Earth observation, with an orbital satellite autonomously identifying targets for the first time without direct human intervention from ground-based analysts. This operational milestone, achieved in April, represents the inaugural deployment of a vision-language model (VLM) in space. It signals a transformative potential for AI to redefine the capabilities and economic value of space-based sensing platforms.
Traditional Data Processing vs. Onboard AI
Historically, the process for analyzing data from Earth observation satellites has involved transmitting vast quantities of raw data back to terrestrial control centers. There, human analysts or conventional machine learning algorithms scrutinize this information to discern relevant patterns and events. However, the YAM-9 spacecraft, developed by Loft Orbital, has demonstrated a paradigm shift. Equipped with a sophisticated onboard software package from NASA’s Jet Propulsion Laboratory (JPL), the satellite processed sensor data and identified points of interest in direct response to natural language prompts.
The Role of Vision-Language Models
The core of this breakthrough is Google DeepMind’s Gemma 3, a vision-language model specifically engineered for edge computing applications. This means it is optimized to function efficiently on hardware with limited resources, situated far from centralized data centers. VLMs uniquely integrate the contextual comprehension capabilities of large language models with the analytical power to interpret visual data. In the YAM-9 demonstration, researchers posed queries such as classifying sensor readings at the interface of natural landscapes and human infrastructure, or identifying key components within railway hubs, all of which the VLM successfully addressed.
Strategic Implications of Orbital AI
Near-Term Benefits: Enhanced Data Triage
This demonstration holds considerable significance, primarily through its potential to dramatically increase the utility of space-based sensors in the immediate future. By performing initial data triage and analysis directly in orbit, VLMs can substantially reduce the overwhelming volume of raw data that currently requires extensive processing by ground-based analysts. This efficiency gain allows for quicker identification and dissemination of critical information.
Long-Term Vision: Scalable Space AI Infrastructure
Looking ahead, this achievement serves as a crucial proof of concept for deploying more extensive AI infrastructure in space. Paul Lasserre, Loft Orbital’s Head of AI, highlighted the potential for “always-on, patrol layers in space.” He elaborated, “If you have a VLM, you can have logic—like ‘monitor this border for me, and let me know when something is suspicious,’ and interact back and forth with the satellites.” This enables proactive monitoring and responsive space asset utilization.
Loft Orbital’s Infrastructure-as-a-Service Model
Loft Orbital’s business strategy centers on providing satellite platforms for third-party clients, essentially operating as an infrastructure-as-a-service provider rather than a traditional satellite manufacturer. A recent agreement with EarthDaily exemplifies this model, involving the construction, launch, and operation of six satellites for the analysis and marketing of collected data. The YAM-9 spacecraft, launched in late 2025, acted as a vital pathfinder for Loft’s orbital AI initiatives and is equipped with an Nvidia Jetson Orin AGX GPU, a leading processor for space-based computing.
NAVI-Orbital: The Software Foundation
The software enabling this VLM integration, known as NAVI-Orbital, was developed under the leadership of Juan Delfa Victoria, a technical lead within NASA JPL’s AI group. While Gemma 3 is an off-the-shelf model, the JPL team undertook significant software engineering efforts to streamline NAVI-Orbital, minimizing its library dependencies and memory footprint for optimal performance on the spacecraft’s hardware.
Industry-Wide Momentum
While this marks the first publicly reported instance of a VLM operating in orbit, it is anticipated that other industry players will follow suit. Planet Labs, which utilizes Jetson Orin processors on its satellites, is currently employing them for simpler object detection tasks but is actively researching VLM applications. Kepler Communications, operating the largest cluster of GPUs in space, indicated that “several undisclosed use cases of our compute environment” have been active since their spacecraft launched in January, though specifics remain confidential due to partner agreements.
Future Constellation Development and Scientific Tools
“Now that we’ve proven the concept, that’s really the direction of travel,” Lasserre affirmed. The objective is to scale up to a constellation of 50 to 100 satellites like YAM-9 to achieve real-time global coverage. Lessons learned from deploying these advanced AI models in orbit will be instrumental in developing larger-scale computing infrastructure in space, particularly concerning power and memory management. Furthermore, this technology could foster novel scientific tools, potentially extending to advanced digital assistants for astronauts on lunar or Martian exploration missions, as conceptualized by JPL researchers.
Business Style Takeaway: The successful deployment of vision-language models on Earth observation satellites represents a pivotal shift towards autonomous space-based intelligence, promising enhanced efficiency and novel applications. This advancement underscores the strategic imperative for businesses and investors to recognize the burgeoning value of onboard AI in space, which could redefine data acquisition, analysis, and the overall economic landscape of the space sector.
Based on materials from : techcrunch.com