# ChatGPT needs one H100 per concurrent user

- **Speaker:** Greg Osuri
- **Attribution:** attributed
- **Said:** 2023-10-26
- **Theme:** GPU Economics — https://www.gregosuri.com/predictions/gpu-economics/
- **Source:** How Akash Network is Democratizing Access to Cloud GPUs \| Greg Osuri (0xResearch)
- **Watch at:** 00:16:43 — https://www.youtube.com/watch?v=CRA0yN28Dw8&t=1003s
- **Canonical:** https://www.gregosuri.com/predictions/gpu-economics/2023-10-26-chatgpt-needs-one-h100-per-concurrent-user-nitf/
- **Cite as:** Greg Osuri, "ChatGPT needs one H100 per concurrent user," How Akash Network is Democratizing Access to Cloud GPUs \| Greg Osuri (0xResearch), 26 Oct 2023, 00:16:43. https://www.youtube.com/watch?v=CRA0yN28Dw8&t=1003s

## Quote — verbatim

> Every time you hit a prompt in ChatGPT, the time it takes to come back requires one H100 per each concurrent user for that time. Now do the math: 100 million users... that's talking about 100 million H100s. That's impossible — they don't exist.

## Context — site annotation, not spoken

Explaining why inference scales with users and why OpenAI throttles concurrency; frames the structural GPU shortage.