Inphi Could Not Resist Hyperscale Vortex

Avatar photoPublished on
Blog 1537308585 223

fibeReality first used the term, “hyperscale vortex” relating to ADVA Optical Networking, as it was one of the examples of foreshadowing of the complete destruction of the optical ecosystem. Although ADVA’s CEO had reservations about getting too involved with the hyperscale data center operators, the vendor eventually succumbed to getting deeper into this quagmire. When it came to Inphi, the allure of joint development and marketing work with Microsoft when the supplier was still an up-and-comer had to be even more overwhelming. In our recent article, “Inphi’s ColorZ: New Buyer Constrictions,” we made the case that the product has fallen way short of expectations of deployment even at Microsoft. While we have tended to admire Inphi as a vendor in the past, from the beginning, anyone listening to its messaging would have received the distinct impression that for distances of 80 kilometers and shorter in general, ColorZ fit like a glove. In our most recent discussion with the company, it now calls that a bit of an over-simplification. Inphi also acknowledged to us the balancing act in playing on both sides of the court, as it sells DSP gear, PAM chipsets, and coherent TIA to system suppliers. Supposedly, it would be one of the reasons for Inphi claiming it has been trying to stay under the radar with ColorZ, which we would describe as strange, given the past fanfare about the product.

While we were willing to accept Inphi’s assertion that our skepticism about being able to get all of the wavelengths to work has not been the case, our statement about the need for some laser development then being irrelevant did not necessarily follow. While the vendor maintained that there was no need for any outside help with the channels, it did admit that the tunable version of ColorZ announced at OFC required working with an alternative laser, which additionally, of course, provided further confirmation on the inventory problems associated with the solution.

Moreover, while we can appreciate not wanting to be too specific about the network of its individual customer, Microsoft, the initial responses to our questions on why the apps for the solution may have been more targeted in nature, seemed evasive to us. Its objection to ColorZ being “shockingly characterized by at least a few industry observers as a solution looking for a problem” was rather perplexing because the focus was not really on the great need for a direct-detect technology in the DCI space at 100G, but the relatively large number of 10-gig DWDM SFPs still being shipped today, along with future 400G ZR modules expected to get greater adoption. After a little pressing, in our opinion, Inphi indirectly confirmed our contention that the main application for ColorZ has been very targeted in nature at Microsoft.

We understand that just because two large data centers are close in proximity, ColorZ does not make sense. Evidently, it is cheaper to use Inphi’s gear in a distributed data center arrangement because it is cheaper than the use of coherent technology. Multiple data centers within 70 kilometers are ganged together as a big data center.

Inphi agreed that 70 kilometers is the sweet spot in combination of being sufficiently outside the blast radius, keeping latency to a minimum, and being geographically attractive overall within a city. While Inphi also mentioned that Microsoft has said publicly that it is more cost-effective to buy smaller data centers inside a metropolitan area, and connect them together to make it work as one, the impression left appeared to be this movement was rather aggressive in nature. We think that Microsoft has actually been very selective in moving down this distributed path, which has accounted for a small market for ColorZ in terms of units.

Furthermore, we estimate that at about $1,000 per module, Inphi likely has done at a minimum, a passable amount of business with ColorZ. It was potentially better than selling a chip to another company in the same space.

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]