Our goal is to do the best we can to track the actual amount of GPU compute available to degree program CS students in America. By measuring this tightly, we can shed light on the severity of the "GPU famine" that most CS students deal with everyday, which if not fixed will lead to a "brain drain" in academia. And since industry doesnt publish research anymore in America, this will lead to a severe reduction in the quality and quantity in academic research in America. We are committed to transparency and accuracy, and we welcome all feedback from the community.

About Us

Why GPUsPerStudent.org?

After a spirited debate following my first-order analysis (see post), I wanted to do a deeper analysis and incorporate all the feedback into a second-order attempt. This website tracks a metric and leaderboard for GPU access at universities.

Shedding light on a problem often fixes the problem. Most people do not know how poor the situation is, so as a PhD student paying for my own compute out of pocket, I thought it would be a good idea to provide hard data on the problem.

It is critical for students to have access to GPUs: (1) to learn and (2) to conduct research. If we want more of both, we need more GPUs for students in America!

No effort online has made a real attempt to show the severity of the current GPU drought. So this is my best attempt. All feedback is welcome. We use GitHub for all PRs and issue tracking.

Calculation

Data Collection (3 LLM Ensemble)

Each university is researched by 3 AI models in parallel:

  • OpenAI GPT-5.2 with web search
  • Anthropic Claude Opus 4.5 with web search
  • Google Gemini 3 Pro with grounding

Results are aggregated using max values to ensure completeness.

Validation Pass

A separate Gemini pass filters out shared/national resources:

  • OSC (Ohio Supercomputer Center)
  • NCAR-Wyoming Supercomputing Center
  • TACC (Texas Advanced Computing Center)
  • DOE National Labs (NERSC, ORNL, etc.)

What We Exclude

  • Compute behind grant application walls
  • Resources the university doesn't own/operate
  • Exception: Free cloud credits available without application

H100-Equiv. Calculation

Total GPU Value = Σ (GPU count × market price)
H100-Equiv = Total Value ÷ $35,000

Student Weighting

  • Undergrad CS: 0.45× (basic coursework)
  • Master's CS: 0.70× (research projects)
  • PhD CS: 0.90× (heavy research computing)

Final Metric

GPUs/Student = H100-Equiv ÷ Weighted Students

Disclaimer

We dont claim to be perfect. Not from a strategy perspective nor an execution standpoint. We accept any and all constructive criticism. If you think we made an error or you believe you have better data or a better way of doing things we would love to hear about it. You can do so via the Dispute PR which will bring you to our GitHub. You can submit a PR and we will review.

Dispute via PR Raw Data

H100-Equiv. GPUs Per CS Student

Leaderboard

Rank University Weighted CS Students* H100-Equiv. Available* GPUs/Student*

Data Sources