# GPU demand doubling every 3.5 months; GPT-4 at a trillion parameters

- **Speaker:** Greg Osuri
- **Attribution:** attributed
- **Said:** 2023-02-25
- **Theme:** GPU Economics — https://www.gregosuri.com/predictions/gpu-economics/
- **Source:** Akash Network Live with Greg Osuri: Akash's progress in 2023, open community development, and more (Akash Network)
- **Watch at:** 00:21:35 — https://www.youtube.com/watch?v=Oz-Oara8i78&t=1295s
- **Canonical:** https://www.gregosuri.com/predictions/gpu-economics/2023-02-25-gpu-demand-doubling-every-3-5-months-gpt-4-at-a-tr-i2lg/
- **Cite as:** Greg Osuri, "GPU demand doubling every 3.5 months; GPT-4 at a trillion parameters," Akash Network Live with Greg Osuri: Akash's progress in 2023, open community development, and more (Akash Network), 25 Feb 2023, 00:21:35. https://www.youtube.com/watch?v=Oz-Oara8i78&t=1295s

## Quote — verbatim

> GPUs are the fuel for AI... the need for GPUs is doubling every 3.5 months as these LLM models get more advanced... GPT-3.5 is 175 billion parameters and GPT-4 is going to be a trillion parameters — that means you need that much of compute.

## Context — site annotation, not spoken

Adds that OpenAI was proposing ~$250,000/year for a dedicated ChatGPT instance ([00:22:17](https://www.youtube.com/watch?v=Oz-Oara8i78&t=1337s)) and that ~90% of LLM cost goes to GPUs ([00:35:11](https://www.youtube.com/watch?v=Oz-Oara8i78&t=2111s)), positioning Akash as the low-cost GPU vendor for the AI wave.