Tired of ads? Enjoy an ad-free experience by signing up.
👩‍🍳 How we use AI at Tech in Asia, thoughtfully and responsibly.
🧔‍♂️ A friendly human may check it before it goes live. More news here

Nvidia-backed Starcloud trains first AI model in space

Starcloud, a Washington-based startup backed by Nvidia, has trained an AI model in space for the first time, marking a step forward in the race to develop orbital data centers.

In November, Starcloud launched a satellite equipped with an Nvidia H100 GPU, which is significantly more powerful than previous chips sent to space.

The Starcloud-1 satellite is running Google’s open large language model Gemma in orbit, marking the first instance of such a model operating on a high-powered GPU off Earth.

Starcloud also trained NanoGPT, an AI model created by OpenAI founding member Andrej Karpathy, on the satellite.

The company says its technology could help address the rising energy use and environmental impact of terrestrial data centers by leveraging solar power in space.

Starcloud plans to add more Nvidia chips and integrate additional cloud platforms in its next satellite launch in October 2026.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

Space data centers need launch costs under $200/kg for a real business case

  • Starcloud’s November satellite launch with an Nvidia H100 GPU marks a milestone, yet the case needs launch costs below $200/kg by mid-2030s to match terrestrial data centers per-kilowatt/year 12.
  • Radiation may force hardware swaps every five to six years; thermal control requires radiators that add mass; inter-satellite links (connections between satellites) need tens of terabits per second with kilometer-scale or closer separations 12.
  • Google’s Project Suncatcher (an internal study into space-based solar power for compute) found that panels in dawn-dusk sun-synchronous orbits (paths that keep satellites in near-constant sunlight) reach up to 8 times Earth’s output, while environmental trade-offs remain unclear because rocket launches and reentry produce pollutants that harm the ozone layer 12.

Ground stations can win orbital compute traffic with high-throughput backhaul

  • AI tasks in orbit need high-throughput downlinks to clouds for data ingestion and model deployment, which opens room for ground-station-as-a-service networks (operators that provide shared satellite antennas with connectivity) such as KSAT (Kongsberg Satellite Services)’s 300+ antennas plus Dhruva Space (an India-based satellite company)’s Low Earth Orbit (LEO) tracking stations with S-band plus X-band radio capabilities 34.
  • Telesat Lightspeed (Telesat’s planned LEO broadband constellation)’s 198-satellite LEO system offers gigabit-per-second speeds with Metro Ethernet Forum (MEF)-compliant interfaces, which lets ground stations add caching and Machine Learning Operations (MLOps) layers for space workloads 5. Providers that offer flexible bandwidth with low-latency links comparable to fiber enable hybrid compute across orbital and terrestrial infrastructure 5.

Stay ahead in Asia’s tech landscape

You've reached your 2 free content limit for the month. Sign up for free to read the full story.

🏄 For casual readers / 👶 Free

Basic

US$0

Free forever

Get instant access to this article and more every month

0 premium content

Unlimited news briefs

5

5 articles

Ad-free reading experience

Just US$0 per day

⌛Sign up in 20s. No payment details needed.

📖 For learners / 👍 Starter

Lite

US$4.92/month

Billed annually at US$59/year

Get instant access to this article and more every month

4

4 premium content

Unlimited news briefs & articles

Ad-free reading experience

Just US$0.17 per day

Cancel anytime

Our subscriber community includes professionals from these companies:

Stay updated on the go with our mobile app.

Get latest insights with smoother, more personalized experience through TIA mobile app.