Nov 12, 2026 · 34 min
Satlyt moves AI processing into orbit
Running Google’s Gemma on Satlyt’s orbital AI (w/ CEO Rama Afullo)
Orbital computing could change how satellites handle urgent data, but its promise depends on launch costs, thermal limits, radiation, and reliable networking.
- 1Satlyt’s software coordinates workloads across satellites from different manufacturers, balancing power, storage, connectivity, and hardware constraints.
- 2Running Google Gemma models onboard could filter satellite data before downlink, reducing transmission demands while preserving the most useful information.
- 3Orbital data centers remain economically and technically difficult, requiring cheaper launches, durable components, radiation protection, and effective heat dissipation.
Don't miss
Rama Afullo explains how Google Gemma could analyze satellite data in orbit and send down only the useful information.
The brief
Rama Afullo traces his path through Tesla, Google, and SpaceX to Satlyt, where he is building software for coordinated computing across heterogeneous satellites.
The case for orbital compute rests on onboard decision-making: satellites could process wildfire, emergency, and other time-sensitive data closer to where it is collected.
Satlyt’s orchestration layer places services across different spacecraft while accounting for power, storage, telemetry, encryption, compression, and changing connectivity.
The clearest application is onboard filtering with Google Gemma: a small model can analyze satellite data in orbit and downlink only what matters.
The vision collides with physics and economics, from launch prices and radiation to vacuum cooling, packet loss, and the mass required for durable hardware.
The broader argument is that solving these constraints for space could produce terrestrial technologies, while making satellites active computing nodes rather than passive sensors.
What was said on this episode
16 statements · 12 positive · 1 negative · 3 neutral
Satlyt is building software that enables AI to run on satellites.
“He's building software that lets AI run on satellites”
Listen at 0:21
- Rama Afulloon Orbital edge computingPositive7:06
Processing data in space supports edge computing and faster emergency responses.
“you can process data in space”
Listen at 7:06
Orbital computing becomes more efficient after reaching critical mass.
“it's actually more efficient in the sort of long run as soon as you hit critical mass”
Listen at 8:12
- Rama Afulloon Orbital data centersPositive11:33
Orbital data centers can be positioned over any Earth location within 90 minutes.
“you can have this data center wherever you want at any point in Earth within 90 minutes”
Listen at 11:33
- Rama Afulloon Satellites in spacePositive13:30
“at least at 100 000 in the next three to five years”
Listen at 13:30
- Rama Afulloon Satlyt-NASA lunar video relayPositive13:44
Satlyt is working with NASA to relay 4K video from the Moon to Earth.
“we're doing is with nasa is relaying data from 4k streaming 4k video from the moon down to earth”
Listen at 13:44
- Rama Afulloon Direct-to-device satellite servicesNegative14:53
Direct-to-device satellite services could eliminate traditional telecommunications companies within ten years.
“it essentially eradicates the typical telco right um in the next 10 years”
Listen at 14:53
- Rama Afulloon Orbital computing launch costsNeutral16:39
Orbital computing needs launch costs near $100 per kilogram to match terrestrial data-center economics.
“in order to get parity with terrestrial data centers, you've got to get about $100 a kilogram”
Listen at 16:39
- Rama Afulloon Starship and New GlennPositive17:17
Starship and New Glenn need to become operational for orbital computing to scale.
“Starship needs to work. Blue Origin's New Glenn needs to work”
Listen at 17:17
- Rama Afulloon SatlytPositive19:52
Satlyt orchestrates software workloads across different satellites.
“a space version of that is part of what we do because we orchestrate across these different satellites”
Listen at 19:52
- Rama Afulloon Satlyt satellite workloadsNeutral20:15
Satlyt's demonstrated satellite workloads are constrained to 15 watts.
“we don't want to go above 15 watts”
Listen at 20:15
- Rama Afulloon Delayed-tolerant networkingPositive22:44
Delayed-tolerant networking can reduce packet loss and bit flips in space communications.
“you can actually stop sort of packet loss in space and bit flips”
Listen at 22:44
- Rama Afulloon Orbital computing innovationsPositive26:34
Orbital computing innovations will produce useful terrestrial technologies.
“we suspect that a similar thing will happen in this case”
Listen at 26:34
- Rama Afulloon Orbital computing thermal managementNeutral27:58
Large orbital computing systems require massive radiators to dissipate heat.
“you essentially need massive radiators”
Listen at 27:58
- Rama Afulloon Orbital computing deploymentsPositive28:20
New orbital computer deployments represent roughly ten times historical satellite compute.
“some of the computers being set up now is like 10x the total compute ever put across these 20,000 satellites”
Listen at 28:20
- Rama Afulloon Orbital electronicsPositive30:49
Longer-lived orbital electronics require radiation hardening or shielding.
“you do need to radiation harden or at least shield some of your electronic surfaces in order for them to operate a lot better”
Listen at 30:49
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.