At ECOC this week, the aspect that will almost assuredly not come across is that large optical component firms have to be asking themselves whether major development efforts on high-end optical components, especially on the client side, is logical anymore. So, regarding work on FR4, DR4, etc. modules — why would they not be clobbered on price to the same extent as happened with the QSFP28 on the CWDM devices? Making it even worse is that at least for the next three-years, the customers that need 400GbE will just be the big-four hyperscalers. Only two of them have demonstrated a willingness to make investments, which have been strategic at best, in an environment in which capital is hardly plentiful. A further indication of the slow uptake on the 400G client is that a number of the optical system suppliers are looking for a modification on the use of one of the standards. In adding in the usual clamor for unrealistic first-cost demands, the very difficult technological obstacles, and the continuing doubts of significant penetration soon on the public line side, there is an undeniable state of uncertainty, and it is difficult to foresee how, as well as when, a major jump to the next data rate will play out.
Both Facebook and Amazon have shown no evident desire to pay for innovations in optical networking themselves, and in our opinion, the latter could care less about future developments, as it is only interested in driving prices down with the apparent mentality that everything it needs can be eventually bought in China. Although Microsoft has provided some investment in outside companies, including at Inphi, it has been mainly focused on driving standards/solutions, such as ZR and COBO. While Google, as with the other three hyperscalers, has made pricing a challenge, and it has enabled a Chinese vendor, InnoLight Technology, fibeReality believes it has shown the greatest appreciation of the four buyers that such gear is hardly a pure commodity, and that meaningful, external funding in optical technology, albeit fairly targeted in nature so far, is necessary.
Concerning getting a higher amount of capacity up to the line, we look at the ZR standard, which was written to support 400GbE on the client, and of course, the 400-gig on the line side. Some of the system houses have approached the component community about expanding the standard, making sure that it is within the power envelope, to have 4x100GbE on the client muxed up to the 400-gig on the lines. There will need to be further investigation of the impact on power dissipation in adding the sub-rates.
Industry observers, hoping for the possibility of a greater impact of 400-gig on the network portion, will point to the decoupling between the market demand on the line and on the client sides. In other words, much of the line side is about aggregation of four 100-gig circuits.
On pricing, the knee-jerk reaction from customers (comparable to previous moves to higher data rates) is that 400G (once again, most notably client-facing) initially has to be less than 4x100G. We see no way to drive it any lower because it is not going to be a step-function down.
There is the possibility that with the next optical boom expected to be in support of 5G happening over the next several years, and as the component companies increasingly diversify into potentially more lucrative business sectors, that additional cash could be available to speed up progress on 400G market developments, particularly on client-side applications (assuming that times become so good that it leads to some cases of amnesia). However, it is also quite conceivable that the technical hurdles in place would not be resolved any faster.
As always, fibeReality does not recommend any securities, and this writer does not invest in any companies being analyzed by us.
To follow us on our totally separate, quick update company blog, which is exclusively on fibeReality’s LinkedIn page, please click here.
[written by Mark Lutkowitz]