Data Center Interconnect Space: Rife with Out-of-Whack Expectations

When it comes to projections in any optical space that has not reached a high level of maturity, a large percentage of stock and technology analysts in our industry are attracted to steep hockey sticks. In the overwhelming number of cases, the prognosticated numbers are continually pushed out to the right, and so despite those rather scant situations in which the outcomes wind up in the ballpark, the standard track record in being originally off by four or more years hardly results in any vendor feeling confident in building a legitimate business case a decade earlier. As to the general outlook for Data Center Interconnect (DCI) itself, it seems part of a subset of a currently pronounced desperation by several notable people purportedly conducting objective investigations to make the numbers work favorably for the data center technology sector as a whole.

There is no other way to explain the incredibly, unrealistic fast ramp that some analysts have been advocating, especially that about 50 percent of the new servers will be going into the cloud by 2018. Even more astounding, we believe that one of the more notable research firms apparently felt pressured this past summer to move its projections in a highly unusual “leftward” direction, asserting that cloud servers will represent half of the total of servers a year sooner. In our opinion, it would hardly be shocking under the most optimistic scenarios that the non-hyperscale buyers of servers did not even come close to representing less than 70 percent of the full sales in the foreseeable future.

The difference in perspectives is critical in looking at the future DCI potential because the biggest amount of transmission on the Internet is already represented by the replication of traffic across multiple sets of servers throughout the various networks around the world. Certainly, the degree to which smaller enterprises outsource their infrastructure to those giant content service providers will determine the amount of bandwidth per wavelength necessary at the edges of those large data centers.

As we have recently noted, with only a very small number of exceptions, the internal backbones (the top-of-rack to spine switch connections) at the hyperscale companies are still not operating above 10-gig – providing a realistic view of the present capacity needs at the high end of the market. In addition, we have also discussed that we are currently only aware of one these firms being especially aggressive on DCI build-out spending. While a few of the others are involved with extending these networks out, we know of at least three in which there are presently doubts about whether they will grow their inter-data center infrastructure at all.

Moreover, while an article in BloombergView, this past October, entitled “Web’s Big Spenders Take a Breather”, is rather optimistic about a bounce-back in CAPEX, at Google, Amazon, and Microsoft, the level of estimated dramatic declines over a two-year period has to be viewed as somewhat alarming. One of the items to perhaps keep in mind is that the emphasis on DCI “throwaways” may be exaggerated – and that overbuilding may be a means to keep capital costs down, as there would continue to be growth with the older generation deployments. For example, we have learned that certain transport gear installed about three years ago in Google’s network has not been removed, yet. Naturally, overlays would result in less purchasing of new devices.

Moreover, our industry in general tends to look at the prospects for optical growth, including DCI, in almost a total vacuum, with hardly taking into account macroeconomic and other considerations. For example, the adverse effects of the international movement to net neutrality are practically ignored.

Lastly, the capitulation by large device vendors away from closed systems resulting in losses of substantial revenue in the DCI space needs to once more be brought up. The “open line system” has become the big buzz-term, albeit in a comparatively quiet manner. There are least five hyperscale companies, which at a bare minimum have begun looking to shift toward this path.

[written by Mark Lutkowitz]