It's incredibly difficult to truly de-identify data and it's incredibly easy to re-identify it. Making data truly anonymous usually also neuters the data for any kind of predictive models.
23andMe is selling either straight up raw data (which is totally allowed under the investment terms) or tightly grouped 'summary' data which can be trivially re-identified. If they were selling actually anonymous data it would be worth nothing because it would yield no predictions.
Disclaimer: I work in ML/predictive analytics. 'anonymous data' is a marketing term for trivially reidentifiable data. If the data is truly anonymous then its predictive power tends towards 0 as does its market price. You don't think GSK dropped 300 mil on mostly useless data do you?