I like it because I have more control over the size and layout of my memory structures than many GC languages, and the goroutine/channel data flow design model lets me use all the cores pretty evenly without having to worry about mutexes or subtleties. Pretty easy to get into the 100k request per second performance regime without special tweaking. I tend to either write long lived servers where the performance per container directly affects the cost, or analytics sort of Calais where I want my laptop to use 1/2 or 3/4 of its cores and get a faster answer from scraping 10M whatevers.