So why do they store indefinitely? If they stated a retention period openly they'd get much less criticism over privacy.

I'd assume for machine learning developments, specifically as training data and back testing. Build a new model with 3x the data you had before or be able to retrospectively see how a model would have performed over 3 years rather than 1.

Don't forget that they're also entirely complicit to NSA demands for live access to data as well, per the somewhat-recent leaks. That's another level of evil above regular profit motivations.