NASA will accelerate scientists' interest in GPUs by facing a data stream from new telescopes

NASA Accelerates Launch of New Space Telescope

In September 2026—eight months ahead of schedule—NASA will launch the new “Nancy Grace Roman” telescope (RST) into Earth orbit. During its mission it will collect about 20,000 TB of data. This enormous volume, combined with streams from other observatories, will create a growing demand for graphics processors (GPUs) needed to process them.

How RST Fits Into the Modern Observation Ecosystem
Observatory | Operating Period | Daily Data Volume | RST | 2026–… | 20,000 TB (total) | James Webb | since 2021 | 57 GB per day | Vera C. Rubin Observatory | 2026 and beyond | ~20 TB per night | Hubble | up to 2000 h | 1–2 GB per day
*The comparison shows that RST will deliver almost a thousand times more data than “Hubble.” This confirms astronomers’ shift to GPU processing: manual analysis can no longer keep pace with the volumes.*

Key Figures and Their Contributions
- Brant Robertson – astrophysicist from the University of California, Santa Cruz.

- Has worked on mission data for over 15 years, collaborating with Nvidia.

- Transitioned from supernova simulations to developing tools for streaming data analysis.

- Ryan Haussen – former graduate student of Robertson.

- Together they created the deep‑learning model *Morpheus* for galaxy recognition in large datasets.

From Convolutional Networks to Transformers
*Morpheus originally used convolutional neural networks, but now moves to transformer architecture—the same technology underlying modern language models. This will allow processing several times more sky per pass.*

Generative AI and Enhancing Earth Observations
- Generative models trained on space data are being developed to correct atmospheric distortions in Earth images.

- Because launching an eight‑meter mirror into orbit remains technologically challenging, software processing becomes the best option.

Financial Challenges
*Robertson notes the growing demand for GPUs:*

> “People want to use AI and machine learning, and GPUs are the only tool for that. Universities are cautious about risks due to limited resources, but we need to show a direction for development.”

The National Science Foundation (NSF) funded the creation of a GPU cluster at the university, but equipment quickly becomes obsolete. Meanwhile, the U.S. President’s administration plans to cut NSF’s budget by 50%, which could hamper further research.

Conclusion:

Launching RST in September 2026 will be a pivotal event for astronomy, opening new horizons for data analysis. At the same time, AI models and GPU‑cloud solutions are evolving, but financial constraints require researchers to be creative and use resources efficiently.