A few thoughts:

* These articles always say that hardware is amazing but software sucks. Let's not forget that hardware has its problems. Intel's management engine is a pile of complexity: https://www.zdnet.com/article/minix-intels-hidden-in-chip-op.... The x86_64 instruction set is hardly inspiring, and I imagine we lose a pile of performance because it fails to adequately represent the underlying hardware. (E.g. there are hundreds of registers on modern CPUs, but you can't access them directly and just have to hope the hardware does a good job of register allocation.)

* Languages unlock performance for the masses. Javascript will never be truly fast because it doesn't represent the machine. E.g. it doesn't have distinct integer and floating point types. Rust represents the machine and is fast, but is not as ergonomic as it could be. OxCaml is inspiring me lately as it's an ergonomic high-level language that also represents the machine. (Scala 3 is also getting there with capture checking, but that is still experimental.) If we want more performance we have to give a way to efficiently write code that can be turned into efficient code.

> hardware

Sure x86 is an absolute mess, but I don't think it's a primary bottleneck. High end x86 cpus still beat high end ARM cpus by a significant margin on raw performance. Even supposing x86/ARM are bottlenecks... yeah a bottleneck at double digit billion ops per second.

> Languages unlock performance for the masses. Javascript will never be truly fast because it doesn't represent the machine.

C# and Go are already really fast (https://github.com/ixy-languages/ixy-languages) languages for the masses and at this point you can compile most things to WASM to get them run in the browser.