My interpretation is that the NSA basically have two main forms of collection: data directly from fibre intercepts, and data obtained (via voluntary agreement, court order, or otherwise) from private companies. This slide [1] would certainly suggest such an arrangement.

The fibre intercepts would fairly easily give access to HTTP traffic, and Facebook/Google/etc. would probably 'come online' at about the same time (there will likely be some differences as it appears there is a need to code a plug-in/processing engine for each major source to pull out usernames etc.[2])

What exactly the dates in the PRISM slide mean is somewhat unclear without more information. It could be, for example the date that the first court order is made, or the date when the company provides to the NSA a more automated way to query the data. I doubt that those dates are related to the fibre intercepts though.

[1] https://image.guim.co.uk/sys-images/Guardian/Pix/pictures/20...

[2] https://image.guim.co.uk/sys-images/Guardian/Pix/audio/video...