This comment really hits the nail on the head. Want to write code that never hits the L2 cache? Write packet headers before you even know what's going inside? Do the math to see if Starlink provides a better latency based on distance between markets? The questions and answers are endless, difficult, and rewarding to solve.
Also, for people who love computers, HFT is very exciting. Imagine a job where no optimization is too premature, where everything must be done to get the fastest possible speeds and push machines and software to physical limits. Very challenging and very fun. And the rewards are great!