As much as I personally wanted compute futures to work, I no longer see how compute is comparable to oil or electricity. You either use compute for something productive or it’s wasted. Unlike oil or electricity there’s no physical good to store, and further unlike oil or electricity compute usage can increase in efficiency overnight. For example - even if someone invents a better engine that runs on less oil it would take years and years to permeate through the economy and affect oil usage enough to move markets. This doesn’t work with compute in the same way, I could open source my inference engine that’s an OOM faster than sglang, or for arguments sake I could merge a pr to sglang and have that immediately distributed to a large amount of users of the compute and lead to far more efficient usage -> many more tokens much faster, but all with the same amount of compute that already exists. Would love your thoughts on this!

Auto mode on talis makes sure you get the best execution for all of your trades on Hyperliquid :)