Compute trading, right now, feels a lot closer to the earlier part of the fracking boom (early 2000s) or pre-ETF BTC than a mature commodities market Clearly a real thing in the long run, but likely something that requires more supply chain stability before hedging costs are brutally optimized There are a lot of "irrational" or wrong-way premiums that cost many percentage points that far outweigh the 10s-100s of basis points you save with hedging These premiums can disappear quickly (as we see with both shale and BTC) but when they exist, it is hard to bootstrap a de novo hedging marketplace because participants are spending more on keeping supply alive [BTC basis trades, fracking land purchases/leverage on land] than on lowering risk from variable demand There are numerous persistent premiums that reflect the non-existence of a stable market clearing price in compute trading: - price of new powered land > land cost of an existing build - refurbished A100s/H100s being worth more than list - cost of refurbishing existing data centers to handle ASICs/Vera Rubin/etc. > cost of a pure new build - cost of origination + new rate for a new build > refinancing an existing data center loan In the fracking boom, there were many idiosyncratic, long-term wrong-way financing costs that persisted for years until 'Bakken' and 'Marcellus' were footnotes in a debt arrangement rather than in the name of the SPV In such a market, owning supply and order flow (proven shale deposits/fresh BTC/proven token demand) is usually better than owning the right to hedge. It is a bit like being a crypto exchange in 2013 — you would have a huge amount of demand if you survived for 4 years, but you were at the whim of your only persistent customer (miners) until that day arrived. And they would bleed you dry via 1000 cuts [remember quanto perps?]. The only thing I see accelerating the visceral need for hedging, much like there was for BTC, is the introduction of a new technology that moves the Pareto frontier and lowers borrowing/refi costs by an order of magnitude For BTC, that came from wrapped assets, DeFi, and the ETF complex (e.g. IBIT options) For AI, that seems to be coming from open source models, model routers, and inference providers Trillion(ish) Dollar Question: Should we expect the same microstructure that we see in DeFi in the matching algorithm token routing algorithms used to match inference capacity with open source demand?