Skip to content

Running Google’s Gemma on Satlyt’s orbital AI (w/ CEO Rama Afullo)

This Week in Startups

Hosted by Jason Calacanis

33m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

This episode of AI Basics explores the emerging field of orbital computing and artificial intelligence in space with Rama Afullo, the founder and CEO of Satlyt. The guest explains how his company builds software that orchestrates and runs AI workloads—such as Google’s Gemma—across heterogeneous satellite constellations, solving challenges like data compression, latency, and power management. The conversation covers why data centers in low earth orbit offer massive real estate, 24-hour solar access, and faster emergency response times compared to terrestrial infrastructure. Listeners will gain a concrete look at how space-based edge computing operates, how companies are tackling thermal dissipation and launch costs, and why edge orchestration in orbit is transitioning from science fiction to practical engineering.

  • Space-based edge computing can pre-process satellite imagery in orbit so that operators only download useful data, significantly reducing downlink costs. 21:38
  • To achieve cost parity with terrestrial data centers, launch costs need to drop from current rideshare rates down to roughly one hundred dollars per kilogram. 16:00
  • Containerization and Kubernetes-style orchestration can be adapted to manage compute tasks and software across dissimilar satellite architectures in orbit. 19:00
  • Delayed tolerant networking protocols developed by internet pioneers help prevent packet loss and bit flips during deep space and lunar data transmissions. 24:21

Chapters

  1. 0:00
    Podcast Intro and Host Greeting

    The host welcomes listeners and introduces the episode’s focus on satellite AI.

  2. 0:47
    Personal Background and Ikigai Discussion

    Rama shares his upbringing, family influence, and how the Japanese concept of ikigai shaped his career path.

  3. 5:55
    Genesis of Sat Lit and Early Career

    Rama recounts his early work at Tesla, Google, and SpaceX, and how those experiences sparked the idea for orbital compute.

  4. 6:33
    Why Put Compute in Space? Societal and Technical Rationale

    The conversation explores privacy, latency, and environmental concerns that motivate orbital data centers.

  5. 8:40
    Scaling to 100,000 Satellites and Cost Parity

    Rama discusses launch economics, the need for large launch vehicles, and the path to cost‑effective orbital compute.

  6. 9:51
    Technical Foundations: Power, Links, and Heterogeneous Systems

    Rama explains sun‑synchronous orbits, optical and RF links, and the need for cross‑satellite orchestration.

  7. 12:30
    Use Cases: Edge Compute, Direct‑to‑Cell, and Lunar Relay

    Examples include real‑time data filtering, direct cellular service, and relaying 4K video from the Moon to Earth.

  8. 18:25
    Demonstrating Satellite Software Stack

    A demo walkthrough shows containerization, Kubernetes‑style orchestration, and on‑board AI using Google Gemma.

  9. 27:39
    Thermal Management and Ruggedness in Space

    The host and guest debate heat dissipation, radiators, and the design of space‑grade processors.

  10. 31:38
    Wrap‑Up, Resources, and Ad Read

    The episode concludes with career opportunities, website info, and a Google partnership ad read.

Highlights

  • 2:50

    Family background and Tesla daysstory

    The guest explains his family background in electrical engineering and computer science, and how working as an early employee at Tesla and later at SpaceX led him to start Satlyt.

  • 3:53

    Naming the satellite startupinsight

    The guest shares how he came up with the name Satlyt a decade ago as a backronym for space and technology analytics before buying domain names.

  • 6:36

    Advantages of orbital data centersinsight

    The host and guest discuss how low earth orbit provides massive real estate, avoids public backlash against terrestrial data centers, and enables global coverage within ninety minutes.

  • 16:29

    Launch capacity economicsdata

    The guest details current launch economics, noting that Starship aims for one thousand dollars per kilogram while parity with terrestrial data centers requires one hundred dollars per kilogram.

  • 18:25

    Demonstrating orbital mission controlstrategy

    The guest walks through a mission operations demo showing how container registries, telemetry analysis, and power budgets are managed across an Nvidia Jetson Orin Nano cluster.

  • 21:00

    Running Gemma in spacedata

    The guest describes running Google Gemma as a small language model directly on satellites to analyze data onboard and filter out unnecessary downloads.

  • 27:39

    Debating thermal management in spacedebate

    The host asks about the contradiction between space being cold and chips overheating, and the guest explains that while space is a vacuum, massive radiators or ruggedized high-wattage chips are required to dissipate heat.

Quotes

  • “There's a Japanese terms called ikigai. Have you ever heard this term ikigai, Rama?”
    Jason Calacanis · 1:19
  • “If you can send only what’s useful, it’s cheaper for you, much more efficient.”
    Unnamed speaker · 22:16
  • satellite.ai 31:30–33:07
  • Satlit 33:43–33:48

About this episode

The TWiST AI Basics series is made possible by Google for Startups! Today’s show: Rama Afullo pitched orbital data centers while working at both Google and SpaceX. At the latter, he was told that the previous staffers who made the suggestion “are no longer at the company.” So he left and started Satlyt, designing software that allows satellites from different manufacturers to run AI onboard and pool their compute into an orbital network. Rama joins Jason Calacanis on a new AI Basics to explain why compute is headed to space, and what has to happen to make an orbital data center as cost-effective as one located firmly on the ground. The benefits of orbital compute are clear: 24/7 solar power in sun-synchronous orbit, avoiding local opposition to new data center construction, and providing needed compute to any spot on earth within just 90 minutes. Join us for an exclusive chat with one of the builders actually bringing this sci-fi dream to life. And get an answer to the ongoing debate: if space is so cold, why do chips in orbit still overheat? Guest Rama Afullo on LinkedIn: https://www.linkedin.com/in/rama-afullo Satlyt: https://satlyt.ai/ Book your software into orbit: https://satlyt.ai/book-orbit Related Links SpaceX and Google both said no. Now his software is riding a SpaceX rocket next to Google’s (TechCrunch): https://techcrunch.com/2026/10/13/satlyt-founded-by-a-former-google-and-spacex-product-manager-raises-8m-to-run-ai-on-satellites/ How Satlyt is building local AI for space with Gemma (DeepMind): https://deepmind.google/models/gemma/gemmaverse/satlyt/ Gemma 3 1B: the small model Satlyt has run in orbit: https://huggingface.co/google/gemma-3-1b-pt NASA Glenn's High-rate Delay Tolerant Networking (HDTN) software: https://software.nasa.gov/software/LEW-19897-2 Google Project Suncatcher: https://blog.google/innovation-and-ai/models-and-research/google-research/google-project-suncatcher-facts/ Planet (Planet Labs): https://www.planet.com/ Starcloud: https://www.starcloud.com/starcloud-1 SpaceX Starship Flight 15: https://www.spacelaunchschedule.com/news/starship-flight-15-launch-date-how-to-watch/ * Timestamps: 0:00 Rama's origin story 6:08 Why put compute in space? 14:12 Direct-to-cell and the end of classic telco 17:59 Demo: Satlyt's mission operations center 23:08 How will we get our data back from Mars? 27:03 The great overheated chips debate 29:51 Can off-the-shelf hardware survive a launch? Subscribe to our newsletters on Substack: TWiST: https://twistartups.substack.com/ TWiAI: https://www.thisweekinai.ai/ TWiVC: https://thisweekinvc.substack.com/ Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp Follow Jason: X: https://twitter.com/Jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis Check out all our partner offers: https://partners.launch.co/ Great TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland Follow TWiST: Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin Instagram: https://www.instagram.com/thisweekinstartups TikTok: https://www.tiktok.com/@thisweekinstartups Substack: https://twistartups.substack.com

Transcript

0:00
33:56

6,300 words · 2 speakers

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • Google
  • Starlink
  • Tesla
  • SpaceX
  • NASA
  • Starship
  • Kubernetes
  • Nvidia
  • Satcom
  • Google Gemma

Page updated · transcript and summary are machine-generated