# 2025 broke the "decentralized AI is impossible" assumption

- **Speaker:** Greg Osuri
- **Attribution:** attributed
- **Said:** 2025-07-01
- **Theme:** Decentralized AI — https://www.gregosuri.com/predictions/decentralized-ai/
- **Source:** Superintelligence Needs The Supercloud Why the AI Revolution... \| PMLS 2025 \| Day 3 \| Open Source AI (Blockworks)
- **Watch at:** 00:06:45 — https://www.youtube.com/watch?v=FBHsD1bgtvc&t=405s
- **Canonical:** https://www.gregosuri.com/predictions/decentralized-ai/2025-07-01-2025-broke-the-decentralized-ai-is-impossible-assu-y6sg/
- **Cite as:** Greg Osuri, "2025 broke the "decentralized AI is impossible" assumption," Superintelligence Needs The Supercloud Why the AI Revolution... \| PMLS 2025 \| Day 3 \| Open Source AI (Blockworks), 1 Jul 2025, 00:06:45. https://www.youtube.com/watch?v=FBHsD1bgtvc&t=405s

## Quote — verbatim

> Until recently, decentralized AI was just considered impossible — too slow, too fragmented. But in 2025, that changed. Pluralis introduced async swarm training... Gensyn trained a 72 billion parameter model on a decentralized testnet — no campus, no cloud, just thousands of volunteers with GPUs. Bagel used ZK-LoRA... to verify decentralized fine-tuning without needing to rerun it. Nous is training a 405B parameter open model using distributed GPUs.

## Context — site annotation, not spoken

Evidence roll for "models trained across continents with compute contributed from living rooms, research labs and micro data centers."