Here are some details:

We are also encouraged by the customer traction we are generating with our 100G products, specially our CWDM full MSA spec 100G transceivers. We secured 9 design wins in the quarter, including 3 100G products and we have approximately 40 100G and 200G qualification efforts underway with various customers, many of whom are outside of our core hyperscale customer base.

We continue to have ongoing discussions with this customer and based on those conversations, we believe the disruption in order flow is related to the ongoing transition from 40G to 100G and not specific to AOI.  We believe there was some inventory buildup during the transition and based on conversations with this customer, we believe that inventory conditions will normalize within the first half of next year.

100G datacenter products, which represented 56% of our datacenter revenue and 40G products, which generated 41% of our datacenter revenue

In the third quarter, we had 3 10% or greater customers in the datacenter business that contributed 37%, 24% and 10% of total revenue, respectively. Based on current orders and forecast from our customers, we continue to expect that we will have 3 hyperscale datacenter customers representing 10% or more of our revenue for the full year 2017.

So far so good..

Here is a crucial part, RJ analyst asking whether 100G was down sequentially. The answer:

Well, I think we saw an overall decline -- as we mentioned in our prepared remarks, we saw an overall decline in business from one customer. So that included both 40 gig and 100 gig. On the other hand, the other customers that we had, were actually up for 100 gig, so that didn't quite balance out the overall decline from the one customer.

Basically, Amazon demand overall was down, with 40G and 100G, and was not entirly made up with wins from others. One could argue that this weakens the Amazon tech transition story, but not necessarily. They might just be in a pause overall during such transition. The decline is in 100G PSM4, as 100G CWDM4 was up 21% sequentially.

And to put fearful minds at ease:

there's no customer that was -- let's put it this way, there was no customer that is sizable in Q3 that is, say, significantly lower in Q4 or lower to the point of nonexistence or something in Q4.

No Facebook or Microsoft bolting, then... Not that we expected that anyway. In fact they are diversifying their customer base strongly:

for the 3 quarters in early -- leading up to this, in other words, Q1 through Q3 of 2017, our datacenter customers, besides what had been our largest customer, are up 180% year-over-year.

That should also increase visibility at least a bit

There was an increase in inventory, in case you are worried that needs to be discounted, worry not:

When we got order from the customer for VOI, that's hub order, then the price is set. So we are not -- when the customer will pull the inventory into revenue, the price will not change. Okay?

Management expects this to come down to normal levels in the first two quarters next year. They have notched capacity utilization of their factories down a bit, but the impact on gross margin is pretty muted. And here is why that is not costing them much in terms of gross margins:

No. I mean, I think we've said pretty consistently, ever since, at least as far back as our analyst day last summer, that is the summer of 2016 that we've made great strides in adding automation, reducing costs, enhancing our flexibility of our manufacturing processes and being able to really adapt our process. Now in previous quarters, we've mainly been talking about it in the context of -- as revenue grows and as we increase and improve our outlook, how fast we can adapt. But it also works the same way on the downside, too, that we have a very flexible and adaptable manufacturing infrastructure that allows us to adapt up or down.