Microsoft’s “Heat Check”

Avatar photoPublished on
Blog 1553878885 258

In basketball, when a player, who has been hot in a game in making several three-point shots in a row, which is within that person’s customary range, after, say, getting open by receiving the usual couple of screens from his teammates — but then later, for example, chucks up a somewhat longer shot, while closely guarded by an opponent, it is often referred to as a “heat check.” fibeReality believes that the same characterization can be made of Microsoft, which over the last several years, has been very aggressive in leaving its traditional, conventional comfort zone as representing a hybrid of an enterprise and carrier mindset, to strive to match, if not outright exceed, the too-often, heedless tactics of the other big-three, large, hyperscale data center operators – which, of course, resulted in the downfall of the optics ecosystem. Yet, at OFC 2019, along with Microsoft announcing a collaboration with Facebook in pushing for Co-Packaged Optics (CPO) standardization (although even if the concept becomes a reality, it would be in the very long-term), the former has seemingly taken on a cavalier attitude by expressing the outright dismissal of shorter-term, lower-cost options. In addition, there appears to be ignorance or denial of its precarious situation as a whole by certain executives on the optics side. In contrast, Google has evidently understood the need to devote more human resources to the actual hands-on development of its optical infrastructure needs.

If the number we heard at the “Will Disaggregation Drive Core Network Deployments in 2025?” session” from Mark Filer, Optical Network Architect, of only being in the low teens, involving people focused on optics at Microsoft is true (not including WAN folks or partner groups helping out with SDN pieces), then in our estimation, it would hardly be enough to deal effectively with the present situation. Filer also happened to make the flabbergasting remark at the OFC Analyst Luncheon that there is a “healthy ecosystem” in the market.

This writer confronted him on this statement, pointing out the impact on vendors, such as Lumentum Holdings. Kohichi Tamura, Senior Director, at the supplier (and most ironically, formerly with “Oclaro Japan”) happened to be one of the speakers at the event, and he elaborated on the vendor’s divestment. Filer pretty much responded that he was in the dark about the margin situation in the more mature, data communications componentry space.

While we also recognize that the optimization of Microsoft’s data center network is a big priority for Microsoft, we had thought that dealing with the matter of transport capacity in the most cost-effective manner was as much a vital part of that endeavor. At the same time, fibeReality knew that Jeff Cox, Partner Director Network Architecture, Microsoft Azure Networking was not exactly a fan of the use of MultiMode Fiber (MMF) combined with VCSELs for applications that are 30 meters long or less — for what we felt was an irrational prejudice.

When we asked him about it at a “Special Chairs’ Session,” we were stunned by what can only be described as a total rejection of the solution in favor of heavily leaning toward WDM, which he asserted would also be helpful in facilitating CPO. Part of our astonishment was based on both Google and Alibaba already moving noticeably in a MMF direction in general.

Moreover, Cox made it clear that he wants to focus on a 4x increase in capacity, as “intermediate steps are difficult to digest.” Nevertheless, with the technical difficulties in just achieving 400GbE, and taking into consideration that the industry engineers ,who we most respect, are anticipating that 8x100G will be necessary to achieve 800 gigabits (so, there will be added complexity alone in moving the market away from the standard four lanes), we have to give him the benefit of the doubt in this case, in that he had nothing to lose by asking for the moon and the stars at the “hyperbole-fest” that is OFC. Even in Cox’s slides, a viable 1.6 terabit ecosystem occurring until after 2024 was not ruled out.

Furthermore, we have no memory of either Cox or Filer mentioning “COBO” in their presentations. While it hardly means that Microsoft will completely abandon the mid-board approach, the idea that the customer is just nonchalantly moving from one form of vaporware, in which major participants in the industry were expected to invest substantial time and money (which caused “Oclaro’s” eloquent objection) to a superior type of ethereal model, is a vivid demonstration of how far removed Microsoft has become from its initial culture, which despite its faults, at least did not result in eliminating important, innovative suppliers. Even taking into consideration that Cox made it clear that the views he expressed were his opinion, he is the top optical architect at Microsoft, and we see little evidence of concern at the firm that the necessary technological advancements that need to be made will call for hefty amounts of development dollars by the cloud provider in order to achieve just incremental kinds of improvements in capacity.

Although it may be a situation in which Microsoft actually does not have much of a problem in remaining at a lower data rate for quite some time, it obviously cannot be that way forever. Moreover, we would not be surprised if Cox changes his mind on VCSELs. Finally, if there is indeed a high level of overconfidence at the customer that the increasingly  difficult R&D work will magically happen in the optics space, just by pressing for the usual standardization efforts, Microsoft may be well overdue for an internal “heat check.”

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]