On 17 September, the US House of Representatives passed by 417 votes to 3 a bill amending a 1978 law on the regulation of public utilities. It requires very large consumers — data centres first among them — to pay the full cost of the grid upgrades they cause, instead of spreading the burden across households' bills. Three votes against, in a chamber divided on just about everything.
We have documented this issue from several angles: the local agreement imposed in Pennsylvania, the flight off the grid, the locking-up of nuclear capacity. Here is the legislative response, and its vote count is the main piece of information.
What the bill fixes
The mechanism being fixed is technical, and its effects are very concrete.
When a very large consumer connects, the grid has to be reinforced: new lines, transformers, sometimes additional generation capacity. These investments are traditionally shared, that is, spread across all ratepayers through the tariff.
That sharing made sense when new connections were a factory or a neighbourhood: everyone ended up benefiting. It makes far less sense when the installation consumes the equivalent of an entire city for a use from which local residents get nothing.
The bill makes whoever causes the cost pay it. It is the most banal principle there is in public economics, and it was not being applied.
Why 417 to 3
That score is worth pausing on, because it is exceptional in an assembly where almost nothing brings together such a majority.
The reason is that this subject has ceased to be a debate about technology and become a debate about the electricity bill. On Pennsylvania, we noted that the opponents' most powerful argument concerned the burden placed on the grid and its effect on household tariffs.
An elected official, whatever their position on AI, has constituents who pay a bill. That is what turns a divisive subject into a consensus.
This bill risks accelerating precisely what we described ten days ago: leaving the grid. If connecting becomes costly, generating your own electricity becomes comparatively more attractive. Yet a self-contained site escapes not only the cost, but also energy planning and public statistics. A measure that protects households could therefore make the sector even less measurable. This is not an argument against the bill, it is an effect to watch.
What it signals elsewhere
This vote is a signal for every jurisdiction facing the same question.
The principle adopted — the one who causes the cost pays for it — is transferable, and it has the advantage of not requiring anyone to take a position on AI. You can be enthusiastic or worried and still consider that an industrial facility should finance its own connection.
That is probably why this type of measure will advance faster than texts dealing with model capabilities: it falls within a domain where administrations know how to act, with tried-and-tested tools and a public debate that is comprehensible.
Key takeaways
There is a broader lesson in that score of 417 to 3.
The divisive questions — pace, safety, employment — remain stuck because they rest on projections that no one can verify. The questions that get settled are those that touch a bill, a plot of land, a measurable nuisance.
This suggests that effective regulation of this industry will come less from texts aimed at artificial intelligence than from those aimed at electricity, water, land and taxation. It is less spectacular, and it is what gets through.