Time to adjust priors.

Crypto projects worth 10s of millions or billions often lack the utility of a single DeepSeek model.

Trained on A100/H100 GPUs at a fraction (<$10M) of the cost of GPT-4, DeepSeek outperforms in reasoning, coding, and crypto-native tasks.

Yet, valuations don’t reflect this. Not in equities and definitely not in crypto.

Base models > hype. Think most of the crypto AI agent, frameworks are all fluff and 99% needs to get cleaned out in both valuation and actual product usage.