I've been meh on RWA perps beyond indices + commodities (which make sense given existing demand)
BUT
My switching between Nebius, Lambda, Prime Intellect, etc. to minimize my GPU cost has made me really want a price per token or energy per token perp to lower cost volatility
Main economic lesson AI learns from crypto is there is only one equilibrium - Birkin Bag (high decentralization, moneyness, etc.) - Cost effective (lowest fees, MEV, etc.)
Crypto people need to learn from the devex of @vllm_project & @DeepSpeedAI; I have rarely used OSS that can go from 0 to 1 as easily for distributed HPC stuff
Half of the HPC C++ stuff I did in 2016 is 5 lines of Python now 🤯