Nvidia wants to run AI directly in space

« Space computing, the ultimate frontier, is now a reality », assures Jensen Huang, the CEO of Nvidia who has presented several new products designed to operate in orbit around our small planet. On the radar: space data centers that part of the tech industry sees as a solution to the ever-growing needs of AI. More pragmatically, the company also wants to equip satellites by giving them dedicated computing capabilities for artificial intelligence.

AI is entering orbit

Behind the big speeches, the idea is actually quite simple: instead of bringing all the data from the satellites back to Earth to analyze it, Nvidia wants to process it directly in space. Most satellites simply collect images or information and then send them to the ground, where they are analyzed in traditional data centers.

Nvidia's goal is to embed enough computing power on board the satellites to analyze the data immediately, without going through the "cow's belly" step. The key component of the manufacturer is the Space-1 Vera Rubin module, with capabilities comparable to a data center... but adapted to the harsh extraterrestrial conditions. Nvidia promises a chip up to 25 times more powerful for inference (AI model processing) than the previous generations used in space.

The company also offers more compact platforms (Jetson Orin, IGX Thor) to run AI models directly on board the satellites. They are capable of processing real-time image streams, data from sensors or navigation systems. In practice, this could be fire detection, real-time analysis of Earth images, or adjusting the satellite's trajectory without human intervention.

Beyond the "simple" satellites, the big issue preoccupying AI actors is the hypothesis of data centers in space. Elon Musk, never the last to promise the Moon (or Mars) to support the stock price of his companies, sees space as a solution to the energy limitations on Earth.

In orbit, these heavy infrastructures could in theory be powered by continuous solar energy, without weather constraints. The remaining "small" technical, logistical, and regulatory issues...

On the contrary, Sam Altman called the idea of "ridiculous." In OpenAI's eyes, the obstacles are considerable: launch costs still very high, logistical complexity, and above all the impossibility of easily repairing equipment in orbit. He estimates that "orbital data centers will not scale significantly this decade." Nvidia is already positioning itself, without waiting for the hypothetical industrialization of these space data centers.

👉🏻 Follow real-time tech news: add 01net to your Google sources, and subscribe to our WhatsApp channel.

Source : Nvidia