Data Center Domination
Impact of DCI on the Optical Market
Data Center Interconnect (DCI) having a significant impact on the optical market, both from a demand standpoint and from a requirements standpoint. While still a relatively small part of the overall market, DCI is growing quickly and has allowed exciting new players like Facebook, Google, and Microsoft to get involved in the optical market. Estimates are that DCI represents 30% of the total DWDM market at the moment. Understandably, DCI can be hard to predict as a percentage of the market. For example, how much of the optical equipment that Verizon is buying is being used for DCI? Still, no one seems to be arguing with the 30% number and everyone involved agrees that data center demands are a huge driver in the industry.
In parallel with the discussions of data center interconnect are discussions on the data center operators’ needs for short-reach, high speed optical connections inside the data center. Facebook, Google, Microsoft, and others have been pushing a variety of high speed optical modulation speeds and physical formats in an effort to get the cost per Gbps down on short-reach, high speed optics. This is certainly an area where things are happening quickly and the ultimate winners are very hard to predict.
Facebook has emerged as a leader in the short-reach optics community, and has very prominently been looking for a way to get their internal optics costs down. In 2016, the head of their optical strategy group said famously that they were looking for $1 per Gbps pricing (yes - $100 for a 100Gps pluggable). This year, Facebook has shown again how they plan to get to low cost optics – by changing the environmental specs (55C max), by changing the lifetime specs (5 years max), and by picking a universal standard for packaging (QSFP-DD – ironically a different standard than the one supported by other DCI titans). However, the statements about $1 per Gbps pricing have very noticeably disappeared from the conversation. Either Facebook is now getting their $100 100Gbps optics (not likely) or they are learning to live in a more rational marketplace where lack of inventory is making pricing less elastic than most would like it to be. Either way, it is certainly safe to say that the efforts led by Facebook, Google, and others have greatly contributed to the astounding drop in 100Gbps short-reach optics prices.
At the Executive Forum prior to this year’s OFC, Facebook presented their overall data center strategy, a truly impressive array of figures and statistics that dwarfed the deployments of most traditional carriers. Facebook is pushing their Voyager platform (part of the TIP) as the way that they believe they will be able to continue to build out their network to scale. An interesting observation was that Facebook (and others) are now almost entirely unable to predict when and where demand spikes will occur. Whereas in the past, a significant perturbation in data traffic might indicate that a virus was spreading, these days a significant unexpected news event can cause enormous spikes in online activity that cannot be predicted. It makes planning a network like Facebook’s very difficult, especially as most of Facebook’s traffic is machine-to-machine (backup, distribution, etc. of all of that new content).
Facebook is one of only about 5 or 6 internet content providers with that level of data center infrastructure. The scale of these datacenter builds dwarfs even the new builds by China Mobile by orders of magnitude. The question that is being asked in the industry as Facebook and others make demands on their optical networking vendors is, if there are only 4 or 5 mega-datacenter operators in the world that can take advantage of the scale that Facebook requires, how can vendors expect to make a profitable business catering to their needs? It’s certainly a large business, but it is incredibly specialized, which makes it very risky. This is a question that continues to come up in these types of forums, but there is no clear answer. For now, the companies who can make those bets continue to make them. The future, with open-source optical systems such as Facebook’s TIP, is less clear.
A cautionary tale presented in a Market Watch panel at OFC this year compared the cash flow positions of traditional carriers with that of the new mega data center operators. The traditional network operators do not have the cash or low debt necessary to do much in the way of innovation in their networks compared to the internet content providers, and there was a warning that the ICP market is driven by such a small number of players that it can change its requirements and demands quickly. That makes optical systems development and optical components development for the data center market risky, as the major customers are willing to very quickly innovate with the latest and greatest, making fast pivots towards and away from technologies in ways that were never seen when the optical world was dominated by traditional service providers. The market will be huge and there will be great innovation, but some vendors are likely going to be left holding a bag full of great technology solutions without a market.
If nothing else, the data center optical market promises to be exciting, if not entirely predictable. And that’s without even mentioning the impact of software, which adds even more interesting dimensions to the discussion. To learn more about the ECI soltuions for data center interconnect and cloud providers in general, click here.