I totally agree with your first paragraph, though.
> because the trend in language design for 25 years has been to make slow languages
No one sets out to make a language to make them slow. The trend is to make higher level languages. Do you really think that there is no reason for it besides novelty and coolness factor?
He's instead saying that it's very much possible to build a language with a similar level of abstraction/ergonomics to say, Java, or Python, or C#, or whatever but with similar performance characteristics to a lower level language like C. And we are starting to see this - there are languages like Rust or D which are (at least to my eyes) much less arduous and foot-gun prone than languages like C or C++ while having similar (or better) performance.
Of course there's also Jai, but I think we should remain unbiased here :P
As an aside though - I think some of those orders of magnitude of performance gains could be had by just writing better code in your existing high level languages. (At least in my experience with enterprise software dev).
So far, reality seems to confirm my intuition.
Can I say "pretty much all of Linux userspace"? Or Java VM? Or Gnome?
A lot of optimizations left on the table have nothing to do with manual memory management, and have everything to do with "eh, let's just the query the database again, that'll shave a day off the schedule".
Answer that and I get the feeling you’ll understand why your current thinking is so misguided.