Urgent.News

What's breaking now, across thousands of outlets.

AI

Google is sending its first AI data center to space

Tech leaders have come to wonder if orbital data centers can help solve the industry’s power crunch.

Google is sending its first AI data center to space

Google has successfully launched Project Suncatcher, an experimental satellite carrying artificial intelligence chips, into low-Earth orbit on October 1. The satellite, equipped with Google's Tensor Processing Units (TPUs), aims to evaluate how these chips perform in the challenging space environment. The trial TPUs have already been tested under simulated solar radiation and cosmic rays at the University of California, Davis.

Google claims the chips handled the radiation "remarkably well," suggesting they could withstand more ionizing radiation than expected over five years in space. Alongside the chips, Project Suncatcher will also test a novel cooling system, crucial in space where there's no airflow to dissipate heat from the TPUs. While the project could provide a solution to the resource-intensive and politically unpopular terrestrial data centers, it also faces numerous engineering and cost challenges.

Astronomers have also expressed concerns about potential light pollution from orbital data centers. Google plans to launch another experimental satellite in 2027 to test communication between potential data centers via lasers.

Written by urgent.news from Scientific American's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Every outlet that reported this, and when →

Read the original at marketwatch.com →

More in AI

OpenAPA: an old military idea to block prompt injection in agents (without new chips)

Classify a session as "secret" after accessing sensitive data and deterministically prevent it from ending up in less secure contexts. The real problem: agents do not distinguish "who's in charge".

  • OpenAPA uses military-grade security model for prompt injection prevention
  • Each resource/destination assigned classification level (public/private)
  • Action attempts to exfiltrate data blocked or require explicit approval

What GPT-6.1 Sol chose to build

A cutting layout can look convincing while leaving its cut order ambiguous. A document comparison can work while putting its navigation offscreen.

More from Thursday 1 October →