The demand for co-packaged optics (CPO) will be driven by many factors. Today, latency-sensitive traffic such as AI and ML is getting a lot of attention. However, this is mostly because AI and ML is new and exciting, while video is old hat. Nonetheless, CIR believes that the video revolution is far from over and its bandwidth demands will sell a lot of CPO units before AI, ML and the like take over as primary CPO market drivers.
CPO is just moving beyond the trial phase and as it starts to be deployed in commercial quantities in the next 2-3 years, we believe that it will be video that will be the driver. Video streaming continues to consume more and more bandwidth every year. Yes, it is expected to constitute over 80% of all Internet traffic by the end of this year, but this still gives plenty of room for growth for video and consequently for CPO.
Phase 1: Zoom will never die: Although for many years the video revolution was supposed to be brought about by fanciful telemedicine applications, the reality was that video really took off as an application when video entertainment (i.e., a mass market application) moved to the Web and then never looked back. Netflix in particular proved a key enabler for video on the net.
Video entertainment continues to grow fast, but we think that something like the video entertainment story is about to be repeated with videoconferencing. There is some evidence that a significant proportion of the workforce like working from home (although admittedly another group hate it!) and these people will never go back to working in an office if they can possibly avoid it. Similarly, on-line events/conferences will never go back to being fully in-person affairs; hybrid events are the way of the future; at least that’s what is currently the received wisdom in the event industry. With a Zoom-like virtual component, event organizers can pull in more attendees and better speakers from wider geographies.
At this point, perhaps we should be honest and confess that the last time a crisis was supposed to launch a videoconferencing boom with the associated rapid deployment of optical infrastructure, it didn’t happen. In the wake of the 9/11attacks a lot of businesses in the United States, at least, switched to videoconferencing and away from travel. Then bit by bit these businesses switched right back.
What has changed this time is that the cost of videoconferencing has declined, and its familiarity has grown with business users. This has made it more natural to keep the video service in place and make use of it once the crisis has passed. But at least as important is the fact that the gear is now here to provide the infrastructure ubiquitous videoconferencing services. Indeed, CPO will be an important part of this infrastructure, with an immediate need in video data centers and in video-saturated core.
Phase II: Moving towards the server: Once the immediate need for CPO in interbuilding and machine-to-machine is fulfilled, and most likely before, we will start to give way to CPO closer to the rack, perhaps even in the rack. That could occur five years from now or perhaps even sooner. This will be a partly video driven deployment, but we think that this is where it will get interesting and where latency- sensitive traffic will start to make its presence felt.
Phase III: A new role for CPO: After five years or so, we see CPO taking on a new role in the network as AI, ML and other latency sensitive traffic begins to be more important as a factor shaping networking infrastructure. Under this scenario, CPO and other high data rate interfaces are being deployed not so much for the data rate per se, but rather to “make space” for this new kind of traffic.
That is to say that there must be enough data capacity in the pipe that the traffic isn’t blocked in any way. This can be a matter of life and death in some networks. Consider the edge networks that will provide control for autonomous vehicles. These must be able to respond instantly to obstacles and dangers, such as children or animals in the road, crashed vehicles, poor weather conditions. This is where high data rates and CPO products may go some way to providing them.
Last, and most speculatively, it is possible to imagine a communications environment where both latency and data rate matters a lot. [Note: For regular digital video, latency is important but not as important as it is for newer application such as AI, ML and autonomous vehicles.] Still, just imagine what will be required from the network when video and AI are combined into a single futuristic service. This may be when CPO may come into its own, supplying both the bandwidth and latency at performance levels that cannot be matched by any other interface technology.