There are regional differences between the same language that can be pretty severe, all the way down to the pronoun level.
So, no they can't.
They can also get US data that doesn't follow EU privacy requirements. So Google has two source of data they can combine into a better product.
In the US the dialects are very similar and usually separated mostly by accent.
When dealing with different countries, entire groups of words do not mean the same thing because either (a) the language has changed, or (b) those things literally did not exist at the time the split occurred. Training an AI on Parisian French will result in an utterly broken experience for Haitian French and Cajun French.
Also, the biggest difference would be the pronunciation differences vs words being different. The word differences are easily modeled after any ASR, if the developer understands the differences and plans for it in their NLU models.
>Google can then use non-private data to adjust the model to the EU market.
Google can do everything a EU-only company can do. They can also do things a EU-only company cannot do. The reverse is not true. So at worst they'll be roughly as good and at best much better.