Although Microsoft’s major push to try to create a market for Inphi’s COLORZ solution was memorable, it was unsuccessful, as the former has remained the only customer so far. In fibeReality’s opinion, the lack of traction by the module is more than Microsoft’s unique optical networking proclivity, and so, although we are inclined to believe it will likely be a buyer of the 400G version as well, the of the product, in general, have resulted in significantly less purchasing than would have been expected by the champion/development partner. Clearly, Inphi has wanted to give the impression for a while that there would be at least one other major purchaser of the device, and we thought that Amazon, given its positioning of data centers, most resembled Microsoft, and thus, would be in the running. The problem is that there may be a good reason why Amazon has not been vocal about its future optics infrastructure, as well as seemingly the loop on current trends. It appears it has practically little to no interest in getting involved in new designs, which would not only make it potentially problematic in being willing to adopt a new concept like COLORZ, but it also seems quite likely to us that, especially compared with the other three major hyperscale operators, that Amazon probably could care less about playing a major role in reconstructing the optical ecosystem. In addition, Inphi’s inability to make meaningful penetration into the market with COLORZ could be the final nail in the coffin on any serious messaging that Ciena is really interested in selling its DSPs as , particularly with its new large focus on the stack to layer 3.
We understand that although Microsoft did deploy COLORZ, and it worked, ironically, there had been a lot of trouble with the link engineering budget. As a result, the hyperscale operator ended up buying purpose-built DCI for many of these point-to-point applications. COLORZ has even been shockingly characterized by at least a few industry observers as a solution looking for a problem.
More specifically, we think that all of the wavelengths did not work on ColorZ. We suspect that the channels at the edges of the band probably lacked enough power to close the links, and so the link budgets were failing. There has even been some speculation that at least one large component vendor was approached to help the situation, such as with developing some laser modifications.
Another obstacle is that despite the HPC folks being obsessed with lower costs, they will go out of their way to avoid the worry in managing too many part numbers. For instance, a large majority of the time, there was no active copper being utilized, within their data centers, even though its use could be a little cheaper. So, they tended to just stick to passive copper and active optical connections.
In this last example, we are only talking about a couple, while in the case of COLORZ, they would certainly not be happy about having 40 sets of part numbers. Most of the companies that might sell such equipment also did not like this problem, which was caused by the device not being tunable.
It would be logical that these adversities led to Inphi trying to salvage some of its investment by introducing its COLORZ-Lite product. Otherwise, an attempt to offer a low-end version, in what the vendor characterized early on as a “niche” opportunity, along with the fact that it had no direction competitors, would not make sense.
Moving onto Ciena, the impact of the lack of penetration by Inphi's COLORZ offering could just have the effect of further moving the optical system player away from the story about it getting into the module business. In the case of Ciena, we would argue it was set up to be a complete failure. Although the product was good, the costs of the 400-gig module was just too expensive.
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 .
[written by Mark Lutkowitz]