Astronomers are adding to demand for scarce GPU chips

In short: Space researchers are increasingly using GPU chips to sift through massive telescope data, adding pressure to an already tight supply.

What's going on

Astronomy is entering a phase where telescopes send back far more information than people can review by hand. NASA says it will launch the Nancy Grace Roman Space Telescope in September 2026, earlier than planned, and it is expected to deliver about 20,000 terabytes of data over its lifetime.

Other major projects are also producing huge streams of data. The James Webb Space Telescope sends down about 57 gigabytes of images each day, and the Vera C. Rubin Observatory is expected to gather around 20 terabytes each night once its survey begins later this year. For comparison, the older Hubble Space Telescope produces about 1 to 2 gigabytes a day.

To cope, astronomers are turning to GPUs, which are specialized chips that can do many calculations at once (like adding more checkout lanes so a store can serve many customers at the same time). Brant Robertson, an astrophysicist at UC Santa Cruz, told TechCrunch he has worked with Nvidia for years and helped build Morpheus, an AI model that scans telescope images and identifies galaxies.

Robertson is now updating Morpheus to use “transformers,” the same general approach used in many modern chatbots. He also said he is working on AI that can help clean up ground telescope images blurred by Earth’s atmosphere.

What to watch

More science teams using GPUs means more competition for the same pool of chips used by AI companies and data centers. Robertson said his university GPU cluster is aging and demand is rising, and he pointed to proposed cuts to the US National Science Foundation budget as a potential headwind for upgrades.

Source: TechCrunch AI