Juniper’s SiPh: Reductio ad Absurdum

Avatar photoPublished on
Blog 1554848528 260

Now Juniper Networks has joined Intel and Cisco Systems in offering 100G data communications transceivers (TXs), in which margins are awfully tough, if not unfeasible in many cases. What is next – Arista Networks being coerced by the hyperscale data center operators to purchase a Silicon Photonics (SiPh) supplier, such as Skorpios Technologies, which has appeared to be in limbo for a long time, but according to the optics supplier in 2017, has a CWDM4 solution? Even Mellanox Technologies (now in the process of being acquired by NVIDIA) entered the PSM4 space that same year, and it is still not totally evident to us whether it relinquished its problematic (SiPh) gear. Once again, the hyperscalers destroyed the optics ecosystem, which has resulted in Lumentum Holdings removing a principal optical TX supplier, along with varying levels of negative impacts on players, such as Kaiam and ColorChip. Separately, MACOM likely is still struggling to put its disparate piece-parts together to build a transceiver. Most vitally, there is the unknown future impact of the II-VI/Finisar merger. Thus, the cloud providers have been increasingly putting pressure on other major vendors to pick up the slack (including engaging in co-packaged optics publicity), despite their gear primarily concentrated in other technical areas. In the end, the long-term goal of these buyers is to use them to drive down prices on higher-capacity TXs to ridiculous levels, using similar, ruthless practices, as happened with 100G, including encouraging any artificial means to get that aim accomplished. Consequently, in the same way that Cisco’s top optical executive is put on the spot to explain the unexplainable, Juniper Network’s CTO has to endure the same painful experience.

Actually, we cannot rule out Juniper ultimately using conventional components for any kind of high-volume, 100G device shipments, as we have suspected Intel has been doing, as the former pointed out that the underlying, complicated SiPh technology, is similar to that of the latter. fibeReality is very skeptical of at least some of the improvement claims made by Juniper, particularly leading to a cost benefit, when it logically could not even come close to spending the resources that Intel did in the past, only for the chip giant to apparently keep on internally proving the inability to make this so-called, next generation of SiPh work.

Unsurprisingly, at Intel’s Datacentric Innovation Day, the firm mentioned SiPh only once, and there were no questions on it at all. At Juniper’s analyst briefing at OFC, Bikash Koley, declined to “quantity the size of [its SiPh] investment,” to provide “any volume numbers,” to get into cost differences with competitive offerings, or to really predict the timing for the arrival of 400-gig TXs. fibeReality also just picked up some intelligence that Juniper has had some technical difficulties, and unlike what was advertised at OFC, availability of the initial TXs could be delayed until the second half of this year.

At the same time, while there was the way-over-the-top rhetoric about “achiev[ing] the holy grail,” and that the present shift in activity happening at Juniper, in attempting to aggressively combine switches with optics, represents the biggest disruptor to pluggables, since the introduction of the VCSEL, there was no press release issued, and there was just a blog post, which was not well-advertised. If indeed Juniper has something that is “fully disruptive,” it probably would be making considerably more noise. It would have had the leverage even with the hyperscalers not to have to get into the 100G space at all — and just wait to leapfrog directly into the 400G space (putting aside the windfall of other opportunities that would have resulted) — in what could have been called from a PR standpoint, “a violation” of the laws of physics.

We are hearing pricing for a CWDM4 expected to be as low as $150 in late 2018, and so the ability for Juniper to compete on cost at 100G there, would seem to be a fantasy, to say the least. It would partially explain why Koley only discussed LR4s (also, this type of platform has been an area of focus for the ex-Aurrion engineers); however, unfortunately, a market which has declined dramatically (so, by definition, not “large volume”), and although it is at least conceivable to make some money in that space, if one presumes that the hyperscalers would be looking for a 50 percent discount on such components, we would still have doubts about this particular SiPh solution having achieved the necessary advancements to make that cost target a reality. Nevertheless, the good news for Juniper, as it has been in the past for Cisco, is that it is hard for a financial analyst to accurately estimate/disaggregate the manufacturing costs of optics for a large router corporation.

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]