# Inference is better at the edge

- **Speaker:** Greg Osuri
- **Attribution:** attributed
- **Said:** 2025-10-10
- **Theme:** Local Compute — https://www.gregosuri.com/predictions/local-compute/
- **Source:** DePIN: Hype or the Next Trillion-Dollar Market? - TOKEN2049 Singapore 2025 (TOKEN2049)
- **Watch at:** 00:16:13 — https://www.youtube.com/watch?v=xr6ZLHwuo2w&t=973s
- **Canonical:** https://www.gregosuri.com/predictions/local-compute/2025-10-10-inference-is-better-at-the-edge-5zr5/
- **Cite as:** Greg Osuri, "Inference is better at the edge," DePIN: Hype or the Next Trillion-Dollar Market? - TOKEN2049 Singapore 2025 (TOKEN2049), 10 Oct 2025, 00:16:13. https://www.youtube.com/watch?v=xr6ZLHwuo2w&t=973s

## Quote — verbatim

> Inference wasn't very centralized. Inference is a lot more flexible when it comes to deployment. I would argue that it's better when it's at the edge. So getting access to edge resources with a common set of rules in a protocol is significantly better than having a centralized coordination layer.

## Context — site annotation, not spoken

Rebutting the moderator's premise that compute inherently centralizes; he separates inference (edge-friendly) from training (communication-heavy, fault-intolerant, serialized).