When giant data center owners started promoting $1 per gigabit transport a while back, the idea seemed ludicrous. Well, at least for optical transceivers, pricing has reached a point in which that goal is actually in the ballpark. The willingness of vendors at the low end to continually drive down prices, such as a 10G SFP selling for no more than $50 at 10 kilometers (as low as $10, so a dollar per gig, to $20 for a multimode version at very short distances), has enabled room at the higher end to significantly reduce expenditures on 100G transceivers to perhaps under $3 per gigabit before the end of this year.
In fact, the 100G prices on transceivers have been dropping faster than for those devices at 40G, even before substantial shipments at the higher speed on the client side, as there are fewer competitors in the latter space. Therefore, based on our intelligence gathering, we find Lumentum’s Holdings’ statement during its quarterly earnings conference this month to be very credible: “I would say that the criteria for getting that [multimillion-dollar] order [for CWDM4 transceivers from a Web 2.0] was ability to give them product, and so I’d say some of our competitors are certainly good at giving price, but not necessarily delivering.” Nevertheless, it is probably only a matter of time before vendors that are aggressively pricing this equipment will be in a position to deliver in something resembling volume.
There has been discussion in the industry about such CWDM4 gear with a 2-kilometer capability, which appears to be selling no higher than $600, to be actually cut as much as in half by year-end 2016. There have also been indications in the industry that the 100G PSM4 (Parallel Singlemode 4 lane) with a reach of 500 meters is already shipping for as low as $300 (thus, it would obviously be $3/gig already) with the potential to drop by just over 15% before 2017.
There are at least a few people, especially at relatively new entrants, in the componentry business who appear to be in denial over the possibility of a precipitous fall in 100G CWDM4 prices because of the inability to be profitable. Yet, at the OFC 2016 panel, “The Promising Market of 100G and Beyond Pluggable Devices…,” there was a slide showing all of the form factors for 100-gig on one axis and the optical interfaces on the other with the clear implication that the volume of any one item is not big enough anyway to necessarily give a supplier a return on its development expense.
The most pivotal point is that the principal buyers of 100G QSFPs right now are the hyperscale operators, and as we have reported in the past, both Alphabet’s Google and Microsoft have invested in Kaiam (along with the public announcements of arrangements by the former with InnoLight Technology and the latter with Inphi). Therefore, the “captive” nature of these relationships would certainly facilitate an ultimately dramatic fall in 100G transceiver prices.
The big wild card in substantially slowing down the rate of decline on pricing for these 100G components in general are the signs of some recalibration of expectations in that that the large data center operators might be reducing their outlook on CAPEX for both this year and next because of the substantial downturn that has started to occur in the US economy. Three of the four such 100G customers that we know about include Google and Microsoft (which both recently reported unsatisfying earnings) as well as Amazon, which although lately had higher than expected earnings, one may wonder when the overall dismal signals in the retail space catches up to the company. While a fourth firm, Facebook is currently doing very well financially, it has been widely known for having its own “$300 CWDM4” in development.
To make matters worse, when it comes to the big users of 100G client-side transceivers, there can be additional customized requirements for a particular buyer, resulting in less of an advantage that comes with any kind of volume. For example, Microsoft insists on having pigtales in its modules, which has the added negative aspect of making it impossible to align directly to duplex singlemode LC ferrules.
Luxtera, which may possibly be the most out in front on its “silicon photonics (SPs)”-based 100G PSM4 is considered in the marketplace to be operating close to cost, perhaps making $50 per transceiver. In addition, we assume that Kaiam, which appears to have established a leadership position in CWDM4 for 40G and 100G, could be at least generating small margins with its MEMS techniques (MEMS costs have gone down substantially with volume use in cell phones) and packaging, as well as it cannot be ruled out that company is at least receiving indirect compensation for R&D efforts that could benefit Google (as well as others) in the future.
On a final note, at the Market Watch panel mentioned above, an executive from Facebook in the audience argued that “blaming the customer” is not legitimate (the speakers included representatives from Avago Technologies, Oclaro, Sumitomo Electric Device Innovations, and Cisco Systems). This writer later responded to the remark directly to the moderator, an architect from Google, that it was justified because his company, which gets down to working at the bare die level, was in effect directly competing with those suppliers on the stage through its financing of manufacturers. While, of course, only acknowledging InnoLight, he made the astounding retort that it was “Google Capital” (with the implication that it was a totally separate entity), which was responsible for the investment.
[written by Mark Lutkowitz]