No. Not true at all. At around a billion records, you start getting into distributed/scale-out systems, and distributed indexes are performance death whether for OLTP or analytics. Even before you cross into distributed query processing, the random access nature of indexes levies a heavy tax. Indexes are a tiny piece of this issue. They address the problem of finding a single needle in a haystack, but if you need to aggregate lots of data, indexes are largely worthless.

What about different compression schemes for different sets of "column" values - with the idea that you execute against the compressed values, rarely ever decompressing. Even column store systems that allow you to specify a compression scheme aren't adaptive. Why the hell not? This is almost 2020.