Astronomers are turning to the same GPU chips that power ChatGPT and Claude to hunt for galaxies — and they are making an already strained global chip supply even tighter. As next-generation space telescopes prepare to flood scientists with unprecedented volumes of data, the scientific community is discovering that its computational needs now compete directly with the AI industry for the world's most sought-after hardware.
A Tsunami of Space Data
NASA announced that it will launch the Nancy Grace Roman space telescope into orbit in September 2026, eight months ahead of schedule. The telescope is expected to deliver 20,000 terabytes of data over its lifetime. That torrent joins the 57 gigabytes of imagery the James Webb Space Telescope downloads every day and the 20 terabytes per night expected from the Vera C. Rubin Observatory in Chile, which begins its survey later this year.
For context, the Hubble Space Telescope — once the gold standard — delivers just 1 to 2 gigabytes of data daily. The next generation of observatories will produce orders of magnitude more data than humans could ever analyze manually. The only way to process it is with GPUs — the same chips that AI companies are spending hundreds of billions of dollars to secure.
From Galaxies to Transformers
UC Santa Cruz astrophysicist Brant Robertson has spent 15 years working with Nvidia to apply GPUs to space science. He and former graduate student Ryan Hausen developed a deep learning model called Morpheus that can scan massive datasets and identify galaxies automatically.
Their early AI analysis of Webb data identified a surprising number of disc galaxies, adding a new dimension to theories about the development of the universe. Now Robertson is upgrading Morpheus from convolutional neural networks to the transformer architecture — the same technology behind large language models like GPT and Claude. The switch will allow Morpheus to analyze several times more area than it currently can, dramatically speeding up galaxy identification.
Robertson is also building generative AI models trained on space telescope data to improve the quality of observations from ground-based telescopes, which are distorted by Earth's atmosphere. Since it remains extremely difficult to launch an 8-meter mirror into orbit, using AI to enhance ground-based imagery is the next best alternative.
The GPU Competition
But the same GPU demand that is driving billions in AI infrastructure investment is squeezing academic researchers. Robertson built a GPU cluster at UC Santa Cruz through National Science Foundation funding, but the hardware is already becoming outdated even as more researchers want to apply compute-intensive techniques to their work.
The pressure is intensifying from both sides. AI companies like Anthropic are securing 5 gigawatts of compute capacity through cloud deals worth billions. Google is launching new custom AI chips to supplement Nvidia hardware. And SpaceX just partnered with Cursor to use compute equivalent to a million H100 chips. Against that level of demand, a university research cluster struggles to compete.
Making matters worse, the Trump administration has proposed cutting the NSF's budget by 50 percent — a move that would further limit academic access to the computing resources that modern science increasingly depends on.
Why It Matters for AI
The GPU crunch facing astronomers illustrates a broader tension in the AI era. The same hardware that enables frontier AI models also powers scientific discovery, drug research, climate modeling, and national security applications. As AI companies absorb more and more of the world's GPU supply, every other field that depends on high-performance computing is feeling the squeeze.
Robertson described the challenge bluntly: universities are risk-averse with constrained resources, so researchers have to be entrepreneurial and make the case that GPU-powered analysis is where the field is going. The science is compelling — AI tools are discovering things in telescope data that humans would never find manually. But without access to the hardware, even the most sophisticated algorithms are useless.
The AI industry's appetite for GPUs shows no signs of slowing. Whether the scientific community can secure enough compute to keep pace with the data streaming down from the next generation of space telescopes is a question that goes beyond astronomy. It is a question about how society allocates its most powerful computing resources — and whether the commercial AI boom is crowding out the basic research that drives human understanding of the universe.
