This is true of public trading platforms, but for the earlier HFT apps that required speed, network was the bottleneck, and writing everything in C allowed them

This is true of public trading platforms, but for the earlier HFT apps that required speed, network was the bottleneck, and writing everything in C allowed them to hack the networking stack to get rid of all the bad assumptions of general-purpose computers that slow you down. Whether it's the BIOS itself or the kernel, or even if you're hacking the actual network appliances, the networking stack is almost certainly written in C, not giving you much of a choice if you have to change it. In more general quant and financial engineering applications, I think it was ecosystem inertia as much as anything. Many of the simulation and solver frameworks were based on physics libraries written in C++, and quantlib was built on top of Boost. C++ just became a de facto lingua franca for anyone doing that kind of work, and they weren't primarily programmers, so they weren't going to spend a lot of time learning other languages and ecosystems when they were already comfortable with and productive in C++.