OTHER

Satlyt Raises $8M to Advance AI Technology for Satellites, Founded by Former Google and SpaceX Product Manager

Rama Afullo has envisioned a future where orbital data centers thrive.

The Kenyan American engineer, who made contributions to Google’s cloud computing sector and briefly worked with SpaceX’s Starlink communications team in 2024, believes that space offers untapped opportunities for computing.

“While at SpaceX, I tried to pitch this idea internally, but they turned it down. The same fate befell my proposal at Google,” he disclosed to TechCrunch.

After leaving SpaceX, Afullo co-founded Satlyt, a company with headquarters in Sunnyvale, California, and Nairobi. Satlyt aims to develop software that allows multiple satellites to work together on large-scale computing tasks in space. He informed TechCrunch that the company has raised an $8 million seed round.

The timing seems propitious for him. SpaceX has committed to advancing orbital data centers, and Satlyt’s software is set to be launched aboard a SpaceX rocket this Thursday, coinciding with the inaugural prototype of Google’s space data center project, Project Suncatcher.

In contrast to SpaceX, Google, and other emerging entities like Starcloud and Cowboy Space Company, Satlyt does not plan to create its own spacecraft. Instead, Afullo’s team focuses on developing software that enables AI models to run on satellites, comparing their approach to that of VMware and Snowflake — companies that allow users to run sophisticated software without worrying about the underlying infrastructure.

“While companies like SpaceX are akin to the iPhone of orbital data centers, we aspire to build the Android framework, creating a horizontally integrated, open ecosystem,” he remarked. Basically, Satlyt aims to offer software compatible across various satellites from different manufacturers.

Satlyt has successfully tested its software on two missions to date. Their main goal is to enhance operational efficiency for spacecraft, which typically depend on ground-based flight controllers to troubleshoot issues. Given the high costs and slow speeds associated with downlinking data to Earth, employing AI for onboard decision-making could result in substantial savings, whether in resolving anomalies or processing sensor data on the spacecraft itself.

Earlier this year, Satlyt utilized Google DeepMind’s Gemma AI model on a spacecraft operated by Momentus. This model decreased the amount of onboard software error transmission by over 60%, a level of efficiency that Afullo suggests could translate to savings of hundreds of thousands of dollars for each satellite annually.

The upcoming launch will deploy Satlyt’s software on a spacecraft created by TakeMe2Space, an Indian startup specializing in satellite computing hardware. This mission involves three clients: NASA, which is funding testing of cloud computing protocols in space; Stellerian, a startup focused on space surveillance aiming to trial image processing workloads; and TakeMe2Space, showcasing that their satellite can support software from various sources.

Currently, there are few high-powered GPUs in orbit. The main opportunity resides in the workloads Satlyt is targeting. However, the company’s next initiative aims to create a shared computing system — essentially a cloud — spanning two different satellites, with plans to attempt this next year. If successful, it will lay the groundwork for a genuine compute cloud in orbit, acting as a third-party provider for satellite manufacturers.

“You can think of it as turning your satellite into a revenue-generating managed service,” Afullo clarified.

Satlyt’s seed funding round was led by Non Sibi Ventures based in Houston, with partner Bernard Harris, a former NASA astronaut with over two decades of experience, contributing to the firm’s belief in this innovative, space-focused business.

The firm’s decision to invest was driven by the understanding that Satlyt does not depend on the ambitious vision of space data centers, which may face unpredictable economic realities and bottlenecks in launches. “Rama doesn’t need space data centers to achieve considerable success, right?” Non Sibi partner Kent Lucas noted to TechCrunch. “Success can simply arise from the growing number of satellites being launched.”

Afullo aims to have his technology operational on 20% of satellites by the end of the decade, foreseeing that nearly all spacecraft manufacturers will include GPUs and similar advanced processors in their satellite designs.

“If you’re launching a satellite without a GPU, at the very least, you’re forgoing a significant opportunity,” he stated.

Satlyt is one of many startups evaluated by TechCrunch and is taking part in Disrupt, as part of our larger Battlefield competition happening this week in downtown San Francisco. The event kicks off on Tuesday. Learn more here.

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.