Late in December, Cisco announced that it would acquire Luxtera a maker of silicon photonics parts. Luxtera is a cool company. They make ‘silicon photonics’, networking gear that combines the Moore’s Law-driven processing miracle of silicon with the literally-speed-of-light data rates of optical connections. Most people we spoke with agree that Luxtera has made some important advances in delivering this long-promised field. That they sold rather than embark on their long-rumored IPO, speaks volumes about the hard reality of the optical industry. For the company and its shareholders, this was a great outcome, but foregoing a public offering carries a message for the networking industry more broadly.
First, some background. Anyone who was around for the last Bubble in the 1990’s should remember all the excitement of the optical industry. The idea was that we could achieve incredibly fast data connections by sending communication signals along optical fiber. If you don’t remember that, trust us, there was more excitement about optical then than there is for Machine Learning today. Most of that excitement evaporated with the 2001 crash. The industry contracted sharply. The few companies that remained radically scaled back their ambitions and marketing. For the most part they ended up as suppliers to the telecom industry providing various optical products for long-range networks – across cities and oceans. And supplying telcos is never fun.
Still the dream persisted. Luxtera emerged a few years ago as a company that could embed optical elements in standard silicon semiconductors. In doing so, they opened up the market for use of optical components in shorter-range applications – notably data centers. And as we all know, over the past decade the Internet majors have gone on a massive data center building spree. Luxtera was able to tap into the market, growing nicely, but at heart optical technologies have limitations which constrain the size of their market. Luxtera was likely looking at a future as a public company constantly whip sawed by large orders from a very small pool of customers.
Here is the problem. Electrical signals – on cables or inside chips – are easy to manipulate. They can be routed and directed and made to do useful things by semiconductors. Optical signals (aka light pulses) are much harder to manage. When you send an electric signal down a wire, they contain not only your data but also instructions on how to route and make use of those signals. To manipulate optical signals largely requires translation from photons to electrons and then back again. That translation slows things down considerably and adds greatly to the cost of the networking gear involved. So long range transmissions work well as there is less need for incremental translation. Having optical signals wired directly into server boards sounds great on paper but usually ends up being impractical for many applications.
The whole optical industry is a study in frustration. It makes use of some of the most advanced physics and engineering in the world, but is ultimately grounded in the very messy process required to actually build it. We have always viewed Optical as the Airline Industry in tech, prone to wild swings in (minimal) profitability. The ratio of PhDs to manual laborers in the average optical company is punishing. It requires companies to have a massive and expensive R&D fixed cost base and also a large factory work force essentially building products by hand.
Luxtera’s true expertise lay in its ability to manufacture optical components in a smart new way. They could bond optical fibers into silicon or motherboards very rapidly (for optical). When people speak of silicon photonics the dream is to manufacture optical components using the same automated bulk processes used to churn out semiconductors. The reality is that there is still a healthy dose of manual labor involved.
This leaves optical companies stuck between products which are expensive to build but only of interest to a small pool of customers. The public optical companies tend to have modest operating margins that fluctuate unpredictably as those few customers accelerate then cancel orders. To make matters worse, the networking industry treats optical suppliers poorly. Speak to a networking professional about optical and they will likely be to quick to complain that the optical components they buy often cost more than the switches and networking gear they get plugged into. Buyers hate the high prices they have to pay for optical components but those prices are the result of high manufacturing costs not the unabashed greed of the suppliers. In reality, the buyers have far more power than the optical suppliers. Something like a dozen companies today consume over 50% of the networking industry’s output, and the next 40% goes to a few dozen more telcos. Rock -> optical companies <- hard place.
For Luxtera there is good news. They have found a good home in Cisco. Cisco has the channel to broadly deploy Luxtera products, reaching customers a small company could not. (For a good read on Cisco’s motivation check out the highly knowledgeable Greg Ferro’s quick post on the deal). Eventually, we will see a lot more optical going into the racks of data centers, connecting directly to individual servers. For the rest of the optical market the future is less clear. There is likely room for further consolidation, although not that much. The data center building boom continues. And the move to 5G in the cellular world will further build demand. That being said, the underlying structure of the optical industry remains challenged.