Virtual Data Centers in Space: The Software Revolution Transforming Satellites into Supercomputers
Satlyt has successfully demonstrated in-orbit supercomputing by distributing processing tasks across satellites, promising unprecedented improvements in data transmission speeds back to Earth.
When humanity launched 'Sputnik 1' on October 4, 1957, space was merely a place for observation. But today, thousands of satellites are orbiting the Earth. These satellites continuously collect billions of gigabytes of raw data, which must rely on a 'downlink' (the process of transmitting data from a satellite to Earth) to be sent back to Earth. However, due to limited 'communication windows' (the time of contact between Earth and a satellite), this process proves to be extremely slow. To solve this complex problem, technology startup Satlyt has taken a major step toward building a 'virtual data center' (a software-defined computing network) capable of processing data in orbit.
Founded in 2024 by former Google and SpaceX employee Rama Afullo, the company Satlyt is developing software that can connect existing satellites to create a space data processing network. Through this orbital data center concept, 'artificial intelligence' (AI) and other computational applications can be run directly on satellites. Because of this technology, data will be analyzed near where it is collected, reducing the reliance on analysis performed by humans on Earth.
| Country | Number of satellites |
|---|---|
| United States (USA) | 8,530 |
| Russia | 1,559 |
| China | 908 |
| Total global number | 12,952 |
13 months of hard work and a successful test: Rama Afullo's experiment
Satlyt has recently achieved a significant breakthrough in the direction of establishing a data center in space. After nearly 13 months of hard work, the company completed a successful test in orbit during the first fifteen days of October 2026. This experiment was conducted under a research project being run in collaboration with the University of Houston and with funding from 'NASA'. According to Afullo, users will only need to set permissions, priorities, and operating limits, while the software will make decisions on its own within those constraints.
The company's software has been successfully installed on 3 satellites currently in orbit. NASA missions also face similar limitations, including limited computing resources, intermittent communication, and the need to quickly deliver useful information to people on Earth. In this orbital experiment, Satlyt's software demonstrated that it could distribute tasks among various satellites based on available computing capacity and connectivity.
ઉપગ્રહો
How does this system work? Simulation and reality
During this test, the company created a simulated network consisting of 4 computers, 4 users, and 8 services. The software evaluated the available 'compute' (computing processing capacity) and communication facilities to create an appropriate task allocation plan. This software was run on only 1 physical computer on the spacecraft, while the remaining 3 computers were merely in simulation form. Finally, the results of this processing were sent back to Earth.
This experiment has proven that Satlyt's software can create processing plans on hardware in space. For example, if a spacecraft detects a wildfire on Earth, this software can quickly send only the most important information to Earth instead of sending the entire raw dataset. This allows for the optimal use of limited communication windows.
- Data collectionSpacecraft collects raw data or images of Earth events such as wildfires or natural disasters.
- Network assessmentSoftware checks the computing capacity and communication connectivity available in orbit.
- Task distributionSoftware distributes computing tasks among various satellites according to defined permissions and priorities.
- Fast downlinkInstead of sending huge raw data, only processed useful information is sent back to Earth.
Some interesting findings also emerged from the success of this technology. During the test, one approach worked the best but required more computing power than other approaches. Apart from this, a technical error regarding time was also observed in the software. The software was stamping the latency as 'arrival time' instead of 'elapsed delay' (the time taken for data to arrive). However, according to Afullo, this did not cause the experiment to fail, as the software development, onboard execution, and output retrieval operations were completely successful.
History of Space and Satellite Statistics
Starting with 'Sputnik 1', the first man-made satellite launched by humanity, the space sector has undergone a massive transformation today. According to data as of June 2025, a total of 12,952 satellites are orbiting the Earth. Of these, the United States has the highest number, with 8,530 satellites. Russia has 1,559 and China has 908 satellites active in orbit. Approximately 90% of satellites are located in 'low Earth orbit' or geostationary orbit.
In 2020, SpaceX's Starlink network, with over 1,000 active satellites, created a large satellite internet constellation. Today, due to the increasing number of satellites, the volume of data is also growing at an unprecedented rate. While early satellites only sent radio signals or weather images, current Earth observation and navigation satellites collect billions of bytes of data.
Limitations and Future Challenges: Next Steps
Satlyt has achieved a major milestone in the direction of sharing computing operations, but much work remains. The software has yet to be tested across various workloads, hardware, and connectivity conditions. Furthermore, development is still required in areas such as obtaining permissions from different spacecraft operators, ensuring secure data transfer, resource scheduling, and recovery when connections are interrupted.
According to the company, the next technical steps will involve comprehensive validation and integration. Subsequently, the ability to share computing operations between different spacecraft will be demonstrated. Rama Afullu's long-term vision is for developers to be able to build applications directly for space without creating new infrastructure for every mission, which will effectively create a virtual AI data center in space.
