996 already became the new normal, it just wasn't evenly distributed. The perceived latency was mostly in the end-output, but you can see it happening as China became dominant not through scientific breakthroughs or discoveries, but through raw unopposed effort put in. Even if two people can't do the work of one, throwing five or ten people at the problem will work.

Now with LLMs which can effectively work nonstop, the only remaining bottleneck is whoever has to be around to approve or otherwise take control.