From Dario Amodei, CEO of Anthropic. I thought this was fair and unbiased despite the source. It agrees with everything I've read elsewhere.
To summarize:
AI models are known to improve according to scaling laws. The more data you feed and the longer you train the better they get. The scaling starts to require exponentially more effort to make progress. You can shift the entire scaling curve through algorithmic innovation. These innovations have been happening constantly (if you've been paying attention AI costs are regularly dropping).
Everyone expected continued cost reductions, the surprising thing is that a Chinese company made a big leap first. But just because the curve shifted doesn't mean everyone will stop. They will now continue to climb the curve which will still require massive amounts of compute in the form of advanced chips.
We don't want an authoritarian government to get to super human intelligence first (assuming that's possible using current methods). China simply does not have enough advanced chips and is too far behind to get what they need to continue to advance. And given the massive number required, even smuggling a few here and there won't help. We have no obligation to sell American technology to a geopolitical rival, particularly when the stakes are this high.