The activity around 5G has picked up considerable pace in terms of the standardization process, experimental testbeds, deployment plans, marketing campaigns, regulatory approvals, spectrum plans, and much more. However, one area that needs critical assessment and understanding is that of the 5G data demand that hasn’t received much attention. Network planners of the LTE era will remember the 2009-10-time period very well. iPhone and Android devices had just started coming on to the network especially in the US and we saw an incredible surge in traffic as a result. AT&T’s network became a living laboratory for the wireless ecosystem as it had the most number of “superphones.” With each new iPhone and Android devices, with each network upgrade, with each new blockbuster app, network traffic was taking quantum leaps. For the US, the traffic was roughly doubling every year with AT&T seeing unprecedented growth.
Most of the early network and economic models proved to be wrong primarily because the underlying assumptions were inaccurate. We are proud that our network forecast models turned out to be the most accurate in the industry. It closely tracked the nuances and the basic variables that made up the overall traffic models. The key early insight was that “superphones” behaved very differently than the regular smartphones (Blackberry, Nokia, Microsoft). We realized early on that consumption patterns were changing on the monthly basis which impacted the overall traffic growth. Our models were fairly granular and all bottoms-up based on actual operator data. This helped us create fairly accurate models and we constantly updated the model as new information became available on a monthly or quarterly basis. Other models were top-down and didn’t capture the furious shifts happening on the tectonic plates.
There were some incorrect assumptions based on the existing data at the time like the pace of LTE adoption, pace of smartphone adoption, growth in social media, data cards being the most dominant mobile data driver, new business models e.g. unlimited, tiered, tethering, etc. which if unaccounted for will lead to erroneous calculations.
Despite all of this, the LTE growth model was still straight-forward as there were limited number of variables one had to track (albeit painstakingly) and forecast. As we embark on understanding the 5G journey, the model is much more complicated. There are a number of new variables that add more uncertainty to the model. Additionally, we just don’t know how the various use-cases will pan out or the new ones will get created based on the capabilities or what new business models will be introduced that can have a drastic impact on the traffic growth. While we are fairly certain that video will play a dominant role but not sure where this video traffic will come from over the long run. We have some ideas about a dozen or so use cases that are likely to come but there is uncertainty around the growth and adoption trajectory. The business models and industry M&A will further impact the data growth curve. As such, one has to build a model that can adapt to the changing scenarios, evolving use cases, and morphing consumer behavior.
We started working on building the model in 2008 and published our first paper on the subject in 2009, “Managing Growth and Profits in the Yottabyte Era.” It was the first paper of its kind which we followed up with another edition in 2010. This paper is our attempt to start the dialog around the lessons from estimating demand from the last cycle and understanding 5G data demand going forward. Of course, the model will evolve as we adjust the framework and the model to the realities of 5G network evolution but we hope this paper lays the groundwork for studying the 5G data demand curves.