The assumptions are that

* "better models" will remain so signficantly more profitable for firms that have access to them that that they're effectively a "must have" for big orgs, rather than a grossly overpriced marginal gain

* said better models will only be attainable by orgs in US jurisdiction, rather than by foreign alternatives that come to be either independently or through a legally clever "cleaving" of a US-jurisdiction business interest that wants access to an eager international market

If either of those are wrong, restricting Anthropic et al to only sell to the domestic market is effectively a poison pill that makes it much harder for them to meet growth and profitability objectives and could see them lose their market-leading position sooner and more thoroughly than if they retained access to a larger market and had more flexibility.

The dual risks of either a) accidentally pushing a foreign competitor into the lead and losing dominant status, or b) pushing the underlying companies hard enough that they decide to relocate.

a) is specifically the risk that the export controls push companies in other countries to prefer non-US models due to the lowered risk of getting cut off from a model. The increase in revenue for non-US AI providers combined with the drop in revenue for US AI providers allows non-US providers to double down on training and reach parity or exceed US SOTA models.

b) is sort of self-explanatory. Same model as above, but when the US AI providers start seeing the revenue drop they decide to relocate internationally instead. The US would probably try to stop that, no idea how successful they would be.

Because it would really increase the interest for Chinese/EU models and would even create real incentives to build models outside of the US.

risk