Home / Technology / AI Trained in Space: A New Frontier
AI Trained in Space: A New Frontier
11 Dec
Summary
- AI model trained in orbit using an Nvidia GPU on a test satellite.
- NanoGPT trained on Shakespeare; Gemma chatbot created in space.
- Space data centers offer potential for abundant solar energy.
A startup has achieved a significant milestone by training an AI model in Earth's orbit. Utilizing an Nvidia H100 enterprise GPU aboard a test satellite launched on November 2, the company successfully trained an open-source AI model named NanoGPT. This model was trained on the complete works of William Shakespeare, demonstrating advanced AI capabilities beyond our planet.
Further advancements include running inference on the trained model and operating Google's Gemma AI model, effectively creating a chatbot in space. This satellite even sent a unique greeting back to Earth. The development is seen as a critical precursor to establishing orbital data centers, a concept that major tech players are exploring.
This move towards off-world data centers aims to alleviate the environmental strain and energy demands posed by terrestrial AI infrastructure, while harnessing the vast solar energy potential in space. Despite technical hurdles like heat management, the startup plans a second satellite with more GPUs, potentially offering services to customers.
