THE ESSENTIALS IN 30 SECONDS

  • AI agents can control the Allen Telescope Array in Northern California, correct incoming data and run AI models on the live stream from one natural-language instruction, according to NVIDIA.
  • The pipeline processes 2.4 terabits per second on NVIDIA RTX PRO 6000 Blackwell GPUs, with inference running on site next to the instrument.
  • The team reports the first real-time AI detection of a pulsar in the Crab Nebula; the performance claims come from NVIDIA and its partners, not independent tests.

The search for signals from extraterrestrial technology produces far more radio data than people can inspect. Breakthrough Listen and the SETI Institute say they now hand much of the routine work to AI agents. In a post published on October 30, NVIDIA described a system at the Allen Telescope Array in Northern California in which agents steer observations and process data while it is still arriving.

How the system is built

According to NVIDIA, a researcher can give a single request in plain language, and agents then control the telescope, improve and correct the incoming data, and run AI models on the stream. Several software layers make this possible. NVIDIA's Holoscan platform handles the large sensor feeds, and its DAQIRI library moves data at terabits per second straight into GPU memory, so it can be analysed during the observation instead of being stored first. The SETI Institute's CyberEther framework performs signal processing using NVIDIA's math libraries. A web-based orchestrator called Nexus serves as the agent harness, giving a single view of instrument status, computing resources and pipelines.

The setup reaches 2.4 terabits per second on RTX PRO 6000 Blackwell GPUs, with inference running on premises beside the instrument. NVIDIA says agents can now run the daily search and compress work that would take people weeks or longer into hours.

What it has achieved so far

The team reports the first real-time AI detection of a pulsar in the Crab Nebula, a known natural radio source that serves as a test of the pipeline. Andrew Siemion, Breakthrough Listen's principal investigator, called the work "a breakthrough in translation". The researchers are also testing open models, including NVIDIA's Nemotron family, for tasks where low latency matters and for possible future observatories that operate offline.

Caveats

The throughput and speed claims come from NVIDIA and its collaborators rather than independent benchmarks, as Blockchain.News points out, and the project does not claim to have found any sign of extraterrestrial intelligence. The more practical significance is for other data-heavy instruments: the same pattern of agents orchestrating live sensor pipelines could apply wherever scientists struggle to keep up with the volume their equipment produces.