Nothing replaces the ability to use historical data to research something. Acquired data is historical information and the cost to re-acquire it (which is a falsity anyways) increases each time you throw it away.
It would be better to focus on forward looking data storage systems rather than re-acquisition.
...meaning that sometimes, we're better off keeping the data properly normalized.
Take that, NoSQL guys! :)
His quote, "Storage is pretty cheap but data acquisition and computation are even cheaper" seems a little out there. Data storage has gotten extremely cheep over the past decade. Computation on the other hand only will get much cheaper if it is considerably parallel.
This also means that compression for on-line files is enjoying a bit of a renaissance.