Wall Street is about to start trading AI compute like it's oil. The last time someone tried this with bandwidth, it was Enron, and it didn't end well.

Reading through a few roundups this week (sources at end) on how compute itself is about to become a tradable commodity. CME is planning to launch two "compute futures" contracts on Oct 5, tied to the hourly rental price of Nvidia H100 and B200 GPUs, and the CFTC just kicked off its review this week.

My actual take: the logic is real. GPU rental prices are wildly volatile (H100 rates reportedly ranged from $2.78 to $7.18 per GPU-hour on the same day depending on provider), so companies genuinely want to hedge that cost the way an airline hedges jet fuel.

In 2000 however, Enron tried to do exactly this with bandwidth, turn internet capacity into a standardized, tradable commodity with an index and a futures market. It collapsed. And one of the main reasons is that bandwidth wasn't actually exchangeable: capacity on one route at one time wasn't interchangeable with another, so the standardized contract never plotted cleanly onto the real thing being traded.

Compute has the exact same problem. An H100-hour on one cloud, in one region, with one interconnect is not the same product as an H100-hour somewhere else. Even one of the people building a rival compute-futures market has openly admitted compute simply is not like a commodity.

Is compute enough like oil for this to work, or enough like bandwidth for it to break?

Oil futures succeeded because a barrel is a barrel.
Compute futures only work if an index can paper over how non-fungible compute actually is.

Sources:

* [Newsletter](https://blog.reserve.org/ai-supercycle-weekly-aug-17-2026-99fef4b3281b)
* [CNBC](https://www.cnbc.com/2026/08/11/ai-computing-power-becomes-a-tradable-asset-class-as-cme-starts-futures.html)
* [Axios](https://www.axios.com/2026/08/12/ai-futures-compute-cme)