Crypto can pool capital to train frontier models (~$50M for a Llama)
Why Decentralized AI Needs Cosmos: Greg Osuri of Akash Explains (The Interop)
“I think you can train a good Llama equivalent for about $50 million… there’s a lot of money in crypto that, if we all pull together, we can train the next trillion parameter model fairly well.” — 00:24:46
Context: Back-of-envelope: Llama 3.1 trained on ~30K H100s for ~2 months; at ~$1/GPU/hr that’s ~$30K/hour — affordable for token treasuries (he cites TAO’s ~$4B market cap).