While I'd have to agree at least somewhat with Kanter on his items 1 and 2, number 3 shows a heavy, deeply and fundamental lack of understanding. Let me quote Kanter here:

3. Silicon photonics are mostly likely to be done on a separate die and co-packaged, no point is adding to the cost of the silicon to add optical elements.

He is completely missing the point that this adding of optical elements allows just for that: adding of optical elements. This will be saving the huge costs of doing optics on a separate, co-packaged die. Instead of a multitude of dies (chips), you'll have just a single one and thus a lower bill of material.

Much more could be said about optoelectronic integration, digital/analog integration or the different optical capabilities of Si and GaAs, and Stephane Gagnon has been doing just that at the CIC. Read the video transcript starting at 04:55 ("Integrating electronics, optics, digital, and analog") and later from 12:21 ("Saving power using optics""), where Stephane is explaining that stuff.