BLOG

100G PSM4 Pricing at $2/Gig and Lower

March, 2017

At the Rump Session at OFC 2017, the topic for debate will be “Sub $0.25/Gbps Optics; How and When will Fiber Finally Kill Copper Cable Interconnects in the Data Center (DC)?” One of the questions for discussion will be “What happened to the $1/Gbps optics cost target for switch interconnects?” As it turns out, our intelligence gathering indicates that the prices being offered by a couple of vendors on PSM4 are even more aggressive than we indicated in a blog article in May of last year. Purchasers of our Clash of Optical Component Vendors & Technologies in Data Center Networks report have already received our detailed explanation as to Intel’s pricing and actual product strategy with their “silicon photonics,” which points to as low as $190 a unit for a high-volume purchase. We also understand that Luxtera has dropped the price of its PSM4s to $200.

We believe that despite any excessive rhetoric to the contrary, including from the head of the silicon photonics effort at Intel, or in the very near future from a director at the company at the OIDA Workshop on Manufacturing Trends for Integrated Photonics, that the firm decided that it was willing to take a loss, and we think it is likely that the transceiver module group there is not long for this world. This maneuver by Intel is not really unprecedented in that its former Optical Platform Division was selling at least one of its components at no more than cost.

Regarding Luxtera, it is a mystery whether it can make money on its PSM4s at such a low price. It is our understanding that the supplier has not accounted for the capital equipment needed for elegant wafer-scale packaging. There are suspicions in the industry that Luxtera needed to purchase new alignment devices to accommodate increased volume. All in all, it would be difficult to believe it can remain profitable.

Regarding the Rump Session, followers of our daily blog read this piece about a month ago: “For both organizers, evidently the last two questions will be the most crucial. ‘Are low-power technical solutions like VCSELs being arbitrarily excluded?’ The answer is affirmative. ‘Is Silicon Photonics the answer to sub $0.25/Gbps or $1/Gbps optics?’ The answer is negative." Both gentleman are heavily VCSEL-centric and justifiably anti-silicon photonics, despite what Finisar may say as a corporation publicly. In fact, it was rather refreshing for this writer to get a honest appraisal of Si photonics from the Singapore division of the manufacturer this past January.

When it comes to the Vice President, Advanced Development, at Finisar, obviously the higher volumes from 3D-sensing applications will lower the unit costs of VCSELs even more. In addition, there is quite a bit of development work in using these lasers on singlemode fiber.

Although we have no doubt that the “provocateurs” will get a fair hearing, and everyone in the room will likely have a lot of fun as is usually the case at this annual event, the fix on the underlying message is in place. The killing of copper aspect of the discussion is really a diversion. There is not likely to be a compelling cost case to rip out cable any more than there is to do so with multimode fiber.

Finally, there is no doubt that Facebook coming up with the dollar per gig in general was not based on any technological principals. It simply sounded good from a public relations perspective – and there was nothing to lose in promoting the idea.

Please find details on our Clash of Metro 100G Optical Vendors with Shifting Network Paradigm here.

[written by Mark Lutkowitz]

Share Article

Add Comment

 

Back

 

Read Comments

Mark Lutkowitz
March, 2017
Respectfully, I do not believe that any issues involving patents is very important in that although VCSELs are "cheap and dirty," they are not a piece of cake to produce in high volume and at low cost.
Andy Sun
March, 2017
With some key high speed VCSEL patents are about to expire, do you think the VCSEL space will quickly become super crowded with both existing vendors that have been suppressed by alliance of these patents holders and emerging companies (I heard quite a few) with experienced teams and maybe years of development in stealth mode. After all, VCSEL is a discrete and not very complicated device to make, with the convenience of wafer-level testing and without all the nasty things to deal with a waveguide facet.