Brad is going around claiming that preemption is coming for local rules on data centers, with Monterey Park held up as the example.
This isn’t true, and the text is unambiguous.
Sec 121 preempts only a law “specifically regulating the development of any artificial intelligence model.” The findings say the same thing. The stated purpose is to preempt “State regulation specifically targeting artificial intelligence model development.” The bill defines “development” as the acts of a developer: determining training objectives, training, fine-tuning, modifying a model’s weights, and evaluating before deployment whether it clears safety or capability thresholds. Building, powering, or zoning a data center appears nowhere in that definition. A company operating a data center and leases out compute is not a developer under the bill, because the definition requires both setting the training objectives and directing the training.
On top of that, Sec 121(c)(1) expressly preserves “any State law or regulation of general applicability.” A data center moratorium is a land use rule. It applies to the building regardless of what runs inside, whether that is AI training, crypto mining, or ordinary cloud storage.
Now look at the measure Brad uses. The ballot language prohibits data centers citywide to protect air quality, water, and public health, and to prevent spikes in electricity and water rates. It doesn’t mention AI model development anywhere. It’s about as clean an example of a generally applicable local land use law as exists, and nothing in this bill preempts it. The voters who passed it would not being overridden by anything in here.
There’s one place a preemption argument might have footing. The word “specifically” is used but not defined. If a city wrote an ordinance that singled out data centers because they’re training AI, a developer would have an argument that the ordinance specifically regulates development. A neutral data center ban doesn’t get there. The general applicability carve-out and the developer definition both cut the other way. Monterey Park’s measure is neutral. It bans all data centers for environmental and infrastructure reasons. So this is the one example the argument cannot reach.
This is the second time today this account has steered the public toward a reading of this preemption that the text does not support.
The first was CSAM. Brad and ARI claim that scrubbing child sexual abuse material from training data a development-level mandate the bill would strip from states. But possessing that material is already a federal crime, which means no lab may lawfully hold it in a training set no matter what any state does or doesn’t require. A general criminal prohibition on that material is a law of general applicability, which is clearly left untouched. The protection they claim the bill removes is secured by federal criminal law that this bill does not touch.
Both arguments are built the same way. Each one stretches “specifically regulating the development of an AI model” to cover something that sits outside it, and each one only holds up if you leave out the general applicability carve-out written into the same section. The CSAM version leaves it out. The data center version leaves it out and then points at Monterey Park, which the carve-out squarely protects.
This is a pattern. This is an organization with lawyers. A general applicability carve-out is not buried, and it is not hard to understand. When a group steers the public toward reading a preemption clause to reach already-criminal conduct and a local land use vote, twice, while leaving out the one sentence in the text that answers both, the charitable explanation gets hard to hold. At that point the framing reads like an effort to be believed by people who have not opened the bill.
There are real questions about this bill. But the argument has to start with what the text says. Read Section 121, all of it, and then we can have the real fight.