This is semantics. The original slides, and the WaPo writeup, made it sound as if the NSA has "direct access", which to those of us who work in this area understand to mean "they can login to the servers and do whatever they want". But it is clear that what is really meant is that the government can access all of the personal data that these companies have, provided the "proper" clearance and assertions by the NSA analyst.

TheNextWeb's speculation that the change in wording means that the data is gathered "indirectly" doesn't make a lot of sense:

For one thing, it adds to the growing claim that the agency instead accessed the information indirectly. In such a case, the most likely method would be via ISPs or mobile operators, but that remains unconfirmed.

Given that the WaPo wording specifically mentions "company-controlled locations":

In another classified report obtained by The Post, the arrangement is described as allowing “collection managers [to send] content tasking instructions directly to equipment installed at company-controlled locations,” rather than directly to company servers.

What this more than likely means is that the NSA uses special-access "portals" provided by these companies to do their searching in. Meaning that the NSA accesses a special, locked down website provided and built by Google/Apple/Facebook etc to do their searching in, a site that runs within Google/Apple/Facebook's datacenters/infrastructe, a system that has access to all of the backend data.

"Direct access" means something different to those of us that work with software systems like this all day long than it does to a journalist. We understand that to mean a login and shell access on a server, while the journalists seem to be misunderstanding "direct" versus "indirect" to refer to where the data is obtained from.

edit: think about it this way: would the government want access to the raw bytes stored on Google's servers, which would require them to understand how to decode Google File System data and be kept 100% up to date on how all of Google's backend systems work? Or would they prefer if Google provided them with a private search engine to access data, which Google has to keep up to date to work with all of the other Google systems?