Nvidia’s AI agents now steer the Allen Telescope Array in the search for alien signals
Nvidia has integrated AI agents into the Allen Telescope Array, allowing researchers to steer the Northern California observatory with plain English instructions.
The Allen Telescope Array uses 42 antennas, and running them has traditionally required careful manual work, such as lining up instruments and cleaning up bad data.
Under the new setup, AI agents take on those daily chores, calibrating instruments, correcting errors, and applying AI models to live radio data.
Researchers can now give orders through natural-language prompts, and software agents handle the rest, a shift from offline analysis to in-flight processing.
Hardware and software
The system can handle up to 2.4 terabits per second of sensor data, with the muscle coming from Nvidia’s RTX PRO 6000 Blackwell GPUs.
Two software pieces make the pipeline work: the DAQIRI library moves high-bandwidth data streams directly into GPU memory, and the SETI Institute’s CyberEther framework relies on Nvidia’s CUDA-X libraries.
Importantly, inference happens on-premises, right next to the telescopes, so data does not need to be shipped off before analysis.
Real-time detection
In February 2025, the team achieved real-time detection of giant pulses from the Crab Pulsar, delivering a speed-up of approximately 600 times over prior benchmarks.
a breakthrough in translation.
The new agentic layer sits on top of that foundation, aiming to make it easier to operate the telescope while enhancing anomaly detection.
Related
