How Space-Based Data Will Drive the Digital Economy

By Andrew Cavalier | 4Q 2024 | IN-7630

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.

The space industry is undergoing rapid transformation as new reusable rockets and Artificial Intelligence (AI) tools for optimizing and enhancing space networks emerge. Alongside integrating these emerging technologies, which signify an inevitable future of edge-computing satellites on the horizon, the government and commercial sectors are showing a growing appetite for one of the most important commodities in our digital economy: space-based data.[1]

According to ABI Research's latest release of Satellite Constellations and Launch 2024 (MD-SATCC-103), the number of Earth Observation (EO) satellites is expected to grow from more than 900 in 2024 to more than 2,300 by 2032. With some industry estimates indicating that each EO satellite generates approximately 100 Terabytes (TB) of data daily, this would collectively amount to around 230 Petabytes (PB) of data per day by 2032. Enhancing this large volume of EO data is an ever-growing suite of powerful AI models and tools, often developed and trained by EO operators themselves and by third-party data analytic firms to help accelerate productivity and enhance analysis. And, unlike AI pure-play companies, they have more control of their training data. In this way, distinct clusters of AI companies, those focused on Internet-trained AI data (Large Language Models (LLMS)) and others focused on space-based unlabeled data (Foundation Models (FMs)) are set to evolve.

[1] Space-based data refer to information obtained from instruments like satellites, telescopes, and space probes and includes remote sensing data, space weather data, satellite imagery, and navigation and communications data.

• Map View Data for Autonomous Systems: Autonomous vehicles use a range of sophisticated sensors onboard to provide a comprehensive view of their surroundings; however, these integrated sensors and cameras are typically limited by range (around 250 meters) and can be limited by lighting and weather conditions. In space, the range of such systems can be significantly expanded, and architecture enables flexible data routing around storms. By leveraging FMs with a persistent view of dedicated locations on Earth, EO satellites could provide critical geospatial data analysis for autonomous vehicles, effectively extending the vehicle's data perception range, and aiding them in navigation, prediction, and even real-time decision-making. This synergy of sensors and AI models on Earth and space could also be applied to aerial vehicles, such as drones or air taxis, and provide advantages over radio Simultaneous Localization and Mapping (SLAM) or radar, especially in more dense urban areas that affect the waveform of these technologies.
• Support for Self-Healing Networks: Communication Service Providers (CSPs) and telcos can also leverage space-based data to enhance the predictive analysis and response times to critical outages of the network. As networks become increasingly complex and gain the ability to detect, diagnose, and repair the network autonomously, the need for more data to identify patterns and potential problems before they occur is paramount. In this scenario, space-based data can help support advanced self-healing capabilities such as AI-driven anomaly detection and repair or deployment of ad hoc connectivity infrastructure, such as High-Altitude Platform Stations (HAPSs), for network infrastructure that is offline due to cyberattacks or inclement weather. In these ways, space-based data can add a critical analytical support element to network resiliency efforts and help maintain network Service-Level Agreements (SLAs).
• Evolution of FM and LLMs Parameters: Major potential customers of space-based data are other AI models, particularly LLM companies such as OpenAI or Anthropic, which could use these data to improve their model performance and generalization. While this can help train image recognition models to learn more diverse patterns and scenarios and provide more robust predictions of real-world events, it also unlocks the potential for faster and higher-quality responses to geospatial data. From this perspective, future models trained on and analyzing multi-dimensional and even multi-modal data can synthesize and provide insights into complex events occurring online and in the physical world. Much like The 3rd Generation Partnership Project's (3GPP) Non-Terrestrial Network (NTN) standard seeks to integrate terrestrial and non-terrestrial network infrastructure, the evolution of AI models will inevitably integrate data from terrestrial and non-terrestrial sources.

Many organizations depend on satellite data for decision-making, but transmitting critical space-based data is often restricted to times when satellites pass over ground stations. In this way, true real-time EO satellite data are challenging to achieve; however, some systems are coming very close to providing near-real-time data, such as ICEEYE, Plant Labs, or MAXAR Technologies (with image collection every 20 to 30 minutes). To overcome time restriction challenges in orbit, many companies are beginning to adopt edge computing, processing data directly on the satellite, to help reduce the need to send large volumes of raw data back to Earth, optimizing the limited time window provided for analysis.

While edge computing in satellites is still early days, mass deployment of EO satellites for continuous coverage or integration into a larger mesh architecture may help provide customers with the continuous stream of data they may need. As satellite technology continues to improve and become more accessible, there will inevitably be greater integration of space-based data into our digital infrastructure. This convergence of space technology and digital capabilities will not only enhance existing services, but also unlock innovative solutions for a more connected and sustainable digital future. Indeed, with the projected synergies with AI and Machine Learning (ML), ABI Research estimates revenue generated from the sale of EO data will increase to over US$6 billion by 2028, reflecting a 2023 to 2028 Compound Annual Growth Rate (CAR) of 15%.