AOI: Lesson of Hyperscale Operators Trap

Although Applied Optoelectronics’ inevitable fall from grace was at least partially self-inflicted, the actual moral of the story has to do with the potential hazards of getting too far ensconced in the hyperscale data center space. The opportunity in general has produced a great deal of fascination for high-level executives at both optical component and system vendors, and they could not resist penetrating the market with the hope of a rewarding return on their investments. They wanted to show off that their companies associated with customers, which have been among the biggest movers and shakers in the US economy. Even in ADVA Optical Networking’s situation, in which the leadership was apparently hesitant to go too deep into the abyss, the enticement was too great to overcome.

The other side of the coin has tended to be ignored, which are irregularities in order rates mainly because of the project-based nature of their network investments. Moreover, outside of China, only the “big four” including Google, Microsoft, Facebook, and Amazon really move the needle in the space. The tier-two types of operators and lower have a much smaller impact on optics expenditures in the sector.

With such a tiny number of large spenders, which all have varying requirements and timelines for deployment, along with occasional budget constraints, it was unescapable that there have often been high levels of concentration by various manufacturers with only one of these buyers, such as AOI with Amazon. So, an interruption at the online retailer was bound to have extremely negative consequences. In addition, our understanding is that InnoLight Technology did not go public because of its large percentage of business with Google.

With pluggables representing by far the largest portion of the optical componentry business with these hyperscale customers, the need to adequately scale is not only important, the necessity for sufficient R&D resources to produce the next-generation of technology within no more than three years is critical as well. Of course, in the case of AOI, it has been even worse with low-entry barriers for the simplified types of transceivers it tended to sell, and with most of its principal direct competitors privately held — it was more problematic to be certain about the extent of the competitive threats.

The dynamics of this marketplace also include other negative aspects that we have discussed in the past including making slashing prices into an art form, and relatively little vendor loyalty. Another factor that has increasingly become clear to us is that apparently these buyers are willing to blindside suppliers, including ones in which they have close ties, with a sudden pause in purchases without any notice, as is apparently the case with more than just Amazon right now for intra-data center gear.

Nevertheless, certain suppliers and advocates have strived to mythologize the space in asserting it was vastly more favorable than the traditional Internet service providers with attempts to make it seem like the large bandwidth requirements of these hyperscale players result in a lack of any lumpiness (in comparison to the cyclical tendencies of the incumbent carriers), and to even make it appear that the cost pressures are “no different than the tier 1 carrier market” (despite, for example, the public network operators not insisting on disaggregated optics in a significant way). In short, just as with the 3D-sensing prospects, selling to these data centers, with only some limited exceptions, is not for the faint of heart.

To follow us on our separate, daily company blog, which is exclusively on fibeReality’s LinkedIn page, please click here. 

[written by Mark Lutkowitz]