AI agents join the search for extraterrestrial signals with NVIDIA Holoscan at the Allen Telescope Array

The article, published on NVIDIA's blog on October 30, 2026, describes how Breakthrough Listen and the SETI Institute are bringing agentic AI into the day-to-day search for extraterrestrial intelligence. The setting is the Allen Telescope Array in Northern California. According to the post, agents there can point the instrument, enhance and correct incoming data, and apply AI models to the data stream, all from a single natural-language request. The NVIDIA Holoscan platform processes the large sensor data streams and gives the agents live data to act on.

The approach has an almost poetic side: the search increasingly centers on human-built AI agents listening for possible alien-built AI agents. The article notes that some scientists have suggested that, if extraterrestrial civilizations exist, they could be billions of years older than humanity. They would then have technologies far beyond ours, including forms of AI. Andrew Siemion, Bernard M. Oliver Chair for SETI at the SETI Institute and principal investigator of Breakthrough Listen, says that whatever these new agentic observatories find may have more in common with them than with the humans who built them. He adds that this is nothing to fear, but rather an advance in translation.

The technical challenge is data volume. Radio telescopes generate enormous amounts of it, and scientists need to identify potentially interesting signals while the observation is still underway. To do so, the system combines several components:

• DAQIRI, a library in the Holoscan platform, lets the telescope stream terabits per second of sensor data directly into GPU memory. Observations are thus processed on the fly instead of filling up hard drives.

• CyberEther, a signal-processing framework developed at the SETI Institute, performs the computations using NVIDIA CUDA-X libraries such as cuFFT and cuBLAS.

• Nexus, a web-based observatory orchestrator, serves as the agents' harness. It gives them a single source of truth on the state of the instrument, compute resources and pipelines. Researchers can monitor packet loss, read live spectrograms and rewrite signal-processing blocks, all through natural-language instructions.

According to the article, the system sustains 2.4 terabits per second on NVIDIA RTX PRO 6000 Blackwell GPUs, with inference running locally, right next to the instrument. Altogether, an agent can point the telescope's antennas, calibrate and focus it to obtain a clean signal, detect and correct errors in the data, and plot the result. NVIDIA says the agents compress into hours what would take humans weeks or years.

As proof that the system works, the post cites the first real-time AI detection of a pulsar, in the Crab Nebula. A pulsar is a rapidly spinning stellar remnant that emits pulses like a cosmic lighthouse. As a next step, the team is evaluating open models, including NVIDIA Nemotron. The use cases are those where data locality and response time matter, and observatories that could one day operate offline on the far side of the Moon.

The article should be read with caution: it is NVIDIA's own, promotional in tone, and provides no independent data or methodological details. The claims about speed relative to humans and about the pulsar detection come from the blog itself, with no further context on how they were measured. Even so, it illustrates a clear trend: AI agents are moving beyond software applications to control physical scientific instruments, with data processing close to the sensor.