I've seen only three ways to make decent money in a still-technical position:
1. management (deciding what gets done and supervising it), in this case being the interface between the pure managers with MBAs and the developers, like the gearing mechanism between the stick and the engine (this is pretty much where I sit);
2. specialization, your unix-bearded database sysadmins who can do wonders with the most constrained resources, think someone who actually worked on building Postgres/Ingres since the 80s, because some institutions like banks do need that kind of talent and they can name their price; I mention beards because when I did look for such a profile, every applicant with one exception was in his 40s or 50s and wore a large grey beard;
3. crossing over to "data science" (which is the bubble word of the day, and where a lot of the fast money is), i.e. picking up skills in statistical learning and applying them to company datasets. It's still 90% getting the data into the shape you want. I used to think this field was the preserve of the PhDs, but most companies I talked to at least in Asia have almost no capability on that side of things (not even for questioning the assumptions and significance of their financial models). I also saw that Andrew Ng said much the same thing in his Coursera lectures, something along the lines of "by the time you finish this course you'll be more competent at machine learning than most data scientists I've met in the Valley" so I suspect this extends to the US.
They're sort of related: you want to know about things like the innards of database engines anyway for anything beyond trivial select operations (2.); you need to be aware of what upstairs wants in order to do a good job as an analyst (1.) and you'll naturally start questioning the requests that come to you ("why are you asking me for this data, what problem are you trying to solve, what assumptions are you making") which is where 3. comes in. That's the most valuable part to the company, but also the hardest politically.
I think mixing these things in an SME setting where you can have a fast impact is where you have the highest ROI as an employee (and thus highest likelihood of good money). Big corps, despite being able to offer great salaries, are so slow and bureaucratic that it is very hard to make a meaningful impact - cue in stories of no company email account for months, datasets shared in Excel because database access can't be given for "security reasons", and the utter lack of accountability that comes from 7-layer middle management structures. Startups (pre-revenue/profitability) are too small to be able to afford, or get much return, on a data scientist type.