The AI-Powered Quest to Explore the Galaxy Exacerbates Global GPU Shortage

The Nancy Grace Roman space telescope, slated for a September 2026 launch, eight months ahead of schedule, is expected to generate 20,000 terabytes of data over its lifetime. This substantial influx of data will further strain the global GPU supply, which is already under pressure from the data-intensive operations of other space telescopes. The James Webb Space Telescope, operational since 2021, downlinks 57 gigabytes of breathtaking imagery daily, while the Vera C.

Rubin Observatory in Chile is set to gather 20 terabytes of data nightly. For context, the Hubble Space Telescope, once the gold standard in space observation, delivers a mere 1 to 2 gigabytes of sensor readings daily. Brant Robertson, an astrophysicist at UC Santa Cruz, has been at the forefront of applying GPU technology to astronomical data analysis.

He has spent 15 years collaborating with Nvidia to develop GPU-accelerated tools for understanding space. Initially, this involved advanced simulations of supernova explosions, and now it entails creating tools to analyze the vast amounts of data from new observatories. Robertson notes a significant evolution in the field: from manually analyzing a few objects to CPU-based analyses on large scales, and now to GPU-accelerated versions of those analyses.

His deep learning model, Morpheus, which he developed with then-graduate student Ryan Hausen, can sift through large data sets to identify galaxies. Early AI analysis of Webb data using Morpheus revealed a surprising number of a specific type of disc galaxy, adding a new layer to theories about the universe's development. Morpheus is now being updated, switching from convolutional neural networks to transformers, the architecture behind large language models.

This upgrade will enable the model to analyze several times the area it currently can, speeding up its work. Additionally, Robertson is working on generative AI models trained on space telescope data to enhance the quality of observations collected by ground telescopes, which are distorted by Earth's atmosphere. However, Robertson faces challenges due to the global demand for GPU access.

Despite building a GPU cluster at UC Santa Cruz with National Science Foundation funding, it is becoming outdated as more researchers seek to apply compute-intensive techniques to their work. The Trump administration's proposal to cut the NSF's budget by 50% in its current budget request adds to the uncertainty. Robertson emphasized the need for entrepreneurship in securing resources, especially when working at the cutting edge of technology.

'People want to do these AI, ML analyses, and GPUs are really the way to do that,' he said. 'You have to be entrepreneurial…especially when you're working kind of at the edge of where the technology is.'