You are correct that if you don't care how complex it is, or how ugly, you could create a new language pretty quickly. GitHub probably has 100 languages invented in the last year. But as for a simple general purpose language taking a lot of man-hours, the evidence from the Red, Eve, Luna, and Beads projects shows that thousands of man-hours are indeed input into such a project. To be general purpose, you have to test it against a diverse set of project types, from games, to business applications to scientific, all the while making adjustments to keep smoothing the rough spots, while trying to keep it simple. If you exclude data storage and graphical interface/event management, and imagine we are still in the 70's on terminals, you can probably crank out a simple language like Lua or Python easily enough, but i don't expect any new general purpose language to catch on if it can't do interaction and at least touch on client-server types of programs. We already have acceptable terminal-based languages that can't draw or interact, i don't think we need any more. Swift, which is probably the best designed language from the major players, is representative of the end of the road of OOP. Certainly the Apple development community has strongly embraced it, and i think it has a very good percentage of use in the Apple ecosystem, ramping up faster than hardly any other language. Another aspect of language design that sucks up the man-hours is the diverse range of platforms that are now extant: you have wristwatch computers, mobile phones, tablets, desktops, browser apps, game consoles, VR headsets, AR headsets, so many more platforms than before, and so many more devices to connect with. This is one of the big reasons for example Swift got taken up so fast: if you want an Apple Watch App, you are very likely going to be using XCode and one of only 2 languages. It is very hard for new languages to cover the full breadth of computing today. If you get down into the nitty-gritty of the GPU, you are talking tens of thousands of hours to master each of the commonly used GPU's. This is why people use Adobe AIR, and Unity, and Unreal Engine, because those systems did the really low level work of learning the hardware and making a reasonable layer on top of the hardware to make it feasible. You will note therefore that most of the projects listed on Steve Krouse's spreadsheet are 2D products, because a 3D platform is a truly massive undertaking, and a moving target because 3D hardware is iterating fast.

No arguments about the difficulty on creating generic languages, of course. That's gonna take them hours. But my core point is still about non-general-purpose languages, on domains where the target audience grows narrower. As for target platforms there are techniques, transpilation being the foremost that comes in mind. For example if the language innovation that one is pursuing is on the front-end side, then the only thing you need is to specify the runtime semantics be such that they can be reasonably efficiently expressed in C, JS or Java runtime models and you're... well not golden, but let's say "silver".