(For the record, I've gotten an offer from Amazon before, but never from Google, and Google seems to make their interview process noticeably harder in terms of how much of your undergrad algorithms and data-structures curricula you have to remember to do the interview. Strangely, my lack of obsession with these subjects hasn't ever stopped me getting anything done in real life.)

Alternative hypothesis: these companies believe that the ability to answer undergrad exam questions about general algoriths-and-data-structures can act as a kind of intelligence quotient for programmers, giving a strong measure of general intelligence with low false-signal rates, and they are hiring for (what they believe to be) raw cleverness because they simply don't believe domain expertise or specialization to be real, exploitable phenomena. This alternative has the advantage of neatly explaining their allocation processes, too.

But on the other hand, believing these to be disjoint hypotheses requires me to assume that undergraduates have never actually specialized by the end of their degree.

If someone doesn't know how to evaluate a candidate, or doesn't feel safe or comfortable, well ... nobody gets fired for asking about algorithms.