Satlyt has raised $8 million to build software that runs artificial-intelligence workloads on satellites owned by other companies. The startup is avoiding the capital cost of building spacecraft, but its business still depends on proving that onboard computing can save operators enough money to justify a new software layer.
From downlink savings to orbital computing
Satellites often send raw data to Earth for processing, a step known as downlink. That can be slow and expensive. Satlyt's software moves selected processing onto the spacecraft, allowing an AI model to compress, classify or act on information before transmission.
The company has flown software on two demonstration missions. On a Momentus spacecraft, Google's Gemma model reduced the size of a transmission about software errors by more than 60%, according to Satlyt's account to TechCrunch. Founder Rama Afullo said savings from faster anomaly handling and reduced transmission could reach hundreds of thousands of dollars per satellite each year, but that figure remains a company estimate rather than audited customer economics.
The next mission broadens the test
A planned deployment on a satellite built by India's TakeMe2Space includes three customers. NASA is paying to test cloud-computing protocols in space, Stellerian plans to test image processing for space surveillance, and TakeMe2Space wants to demonstrate that its hardware can host third-party software.
That mix is important because it tests more than compression. It asks whether one software platform can support government research, commercial imagery and infrastructure services on the same spacecraft. A successful mission would still be a technical demonstration, not proof of repeatable subscription revenue.
An asset-light strategy with hard constraints
Satlyt compares its role with software platforms such as VMware and Snowflake: customers use common tools without managing every detail of the underlying hardware. The analogy is useful, but satellites have stricter power, radiation, communications and reliability limits than terrestrial servers. Hardware differences can make a cross-vendor platform difficult to maintain.
The company plans to connect computing across two satellites in 2027. That would be a more meaningful step toward an orbital cloud because jobs would have to move across separate spacecraft. Launch schedules, limited access to powerful graphics processors in orbit and uncertain space-data-center economics remain constraints.
The $8 million round, led by Non Sibi Ventures, gives Satlyt resources to expand its team and deployments. The next evidence is operational: successful customer workloads on the TakeMe2Space mission, a two-satellite demonstration and disclosed paying deployments. Afullo's goal of running on 20% of satellites by 2030 is an ambition, not a forecast investors should treat as established demand.