In a groundbreaking development, Axiom Space has announced that it will be launching the first-ever orbiting data center nodes in space by the end of this year. The data centers would use artificial intelligence (AI) and machine learning algorithms to process the data from Earth observation satellites. Considering the annual satellite launches worldwide, which have increased exponentially in the last 5 years, the sheer volume of data generated by Earth observation satellites warrants on-orbit AI data processing. Although it would be a first launch of the sort, it will only be a beginning, as the other countries are likely to follow suit because of several advantages of on-orbit data processing nodes.
The primary objective is to make the data processing efficient, secure, and fast. Sifting through large volumes of data generated by geosensing satellites, as they continue to exponentially grow in the Low Earth Orbit (LEO) in the last few years, is indeed an arduous task. There are algorithms and systems in place to analyze the data as it is downloaded from the satellites. However, the in-orbit data centers can communicate with other satellites, accumulating the data on multiple nodes in space where it would be processed, and consequently the end user has access to the requisite insight directly. It would allow the users direct access to the satellite imagery and supplementary data to fast-track the decision-making process. The feature in particular could be helpful in disaster response during emergencies like wildfires and hurricanes, particularly in the wake of the climate change-led drastic shift in the weather patterns lately, to help mitigate any humanitarian crisis.
One of the challenges associated with the data centers is the high energy cost for their functioning and cooling. More than 400 million terabytes of new data are generated by governments, businesses, and common citizens every day. While much of this data is locally stored on the devices, a significant chunk is backed up on cloud storage or uploaded to platforms like YouTube. The demand for cloud computing is only expected to grow significantly in the coming years. The energy demand for an AI model to crunch data or make an image is even more exorbitant. For a large language model (LLM) like ChatGPT to generate a 100-word email, it would take approximately 0.14 kWh of energy, enough to light 14 LED bulbs for an hour, and nearly half a liter of water to dissipate the heat. Considering the scale on which these platforms are used, the energy cost and environmental impact of high energy demand beg the question about the carbon footprint of the ‘efficiency.’ In-orbit data centers could be an answer to the questions, considering the space has an unlimited supply of solar energy and the cold temperature to tackle the overheating servers problem. Though, the thermal regulation to maintain the temperature could be a challenge, as radiating heat out in space is comparatively slower than convection cooling on earth.
Furthermore, it offers an additional layer of data protection, as the data transmitted through downlink by the satellites to the ground stations is susceptible to interception. On-orbit processing could help mitigate some of these concerns. The raw data is generally large in size, requiring higher bandwidth and encryption. Processing the data in space and transmitting only the encrypted summary data makes it harder to interpret even if intercepted. With multiple orbital nodes, data processing could be distributed across multiple satellites, reducing the reliance on ground-based storage and transmission, making a single interception unlikely to compromise the complete data set. It may also make the need for high-power, wide-beam data links obsolete—commonly used by satellites to transmit data and are easier to intercept. Narrow beam, low-power transmissions augment the data security protocols, making the prospects of on-orbit data processing even more viable.
The concern, however, regarding space-based data processing and analysis, especially pertaining to military satellite data, is the risk of miscalculation and inadvertent escalation. While instant insights and situational awareness could significantly shorten the OODA loop for military response, there are chances of misinterpretation of the on-ground situation by the on-orbit AI algorithms. For instance, large-scale military exercises mimic the actual war-like situation and, as is evident from the Russian mobilization, could at times take a turn for actual war. Similarly, in the case of India and Pakistan, long columns of troops or equipment for a regular redeployment or maintenance could be interpreted by the satellites as the onset of a military offensive. The latter example is especially pertinent, as both countries possess nuclear weapons, making the situation more complex. An escalation could take a turn for the worse within a short span, and multiple rungs of the escalation ladder could be skipped in rapid succession. Since the training data for an AI model contains an inherent bias, it could exacerbate a conflict situation. The Line of Control (LoC) between India and Pakistan is almost perpetually in a state of limited hot conflict. For any side to gain a competitive edge based on space surveillance and reconnaissance—particularly when operational commanders receive pre-analyzed insights—could lead to far more drastic consequences.
The advent of AI-powered on-orbit data processing marks a transformative shift in the way satellite data is handled, potentially offering unprecedented efficiency, energy solutions, and enhanced data security. However, as space becomes increasingly militarized, the risks of miscalculation and escalation, especially when AI is involved, cannot be ignored. To fully harness the benefits while minimizing the dangers, robust international norms and regulatory frameworks must evolve alongside this technological leap.
