As my post was the origin of this, I want to add my takeaways from the comments:

Datacenters are slower in adopting 100G than expected. Lots of manufacturers ramped up their 100G-production, so 100G is over-supplied. Prices are low for that reason, Applied Opto, Oclaro and  others seem to see a lumpy quarter coming.

CWDM is seen as the most promising 100G-transciever-solution on the market today. The reasons: Several wavelenths over one cable, no temperature issues, reasonable pricing. 4 x 25Gb/s over 1 optical cabel is the standard. Oclaro isn´t sure if they want to produce whole transcievers due to prices being down. Maybe they will just sell lasers to other producers.

SiliconPhotonics isn´t seen as competitor. It´s used for low-end-products only. Applied Opto has an own team for SiPh, but it seems to have a low priority.

It seems to me integration is not the main theme of producers. Maybe it´s because they are not able to deliver. Maybe that´s why datacenters wait until the dust settles - first movers may move wrong.

Please remember, Poets advantage (not produced yet) would be: Not CWDM but DWDM: More wavelengths = more bandwidth. But unike other DWDMs no temperature issues and low price. It seems Poet is not too late for the party as datacenters are slower than expected.

Please beware: These were not my comments. I just summarized what others wrote and added some own opinions. But most was said by companies like oclaro during their earnings calls.