Peak Internet growth may have been a couple of years ago. For more than a decade, Internet traffic went up ~40% every year. Cisco's VNI, the most accurate numbers available, sees growth this year down to 27% on landlines and falling to 15-20% many places over the next few years. (Chart below) (Thank you, Arielle)

The result: bandwidth cost per month per subscriber will continue flat to down. For large carriers, that's been about $1/month since ~2003. Moore's Law has been reducing equipment costs at a similar rate. This is confirmed by the global carrier spending on service provider routers, which continues about flat per customer.

Mobile growth is staying higher. 40-50% worldwide. Fortunately, mobile technology is moving even faster. With today's level of capex, LTE networks can increase capacity 10x to 15x.

Cisco's projection of 7X growth in mobile data in five years does not justify a conclusion "For mobile carriers (and other network operators as well), a migration plan to 5G is extremely important." In the U.S., the growth will be less than 5X.  My guess is that "unlimited" plans will raise that a little more than historical data suggests.

Cisco's estimates for mobile traffic in the U.S. and Canada in 2020 is 4525 petabytes and in 2021 is 5883 petabytes. That's a 30% growth rate. Total consumer traffic in the U.S. and Canada they see as 48,224 petabytes and 56,470 petabytes in 2021. That's a 17% growth rate, lower on landlines. 

The current surge in 5G mmWave is not because the technology will be required to meet demand. Rather, it is inspired by costs coming down so fast the 5G networks will be a cheaper way to deliver the bits. In addition, Verizon sees a large opportunity to replace cable and other landlines. DOCOMO may be seeing similar, but would never say so publicly because many of those landlines belong to parent NTT.

The problem for most large carriers is that they can't sell the capacity they have, not that they can't keep up. 

A previous Cisco report saw mobile growth falling more rapidly. I'd guess the adjustment is due to the rapid spread of unlimited offerings. They bring down the effective cost per bit. Some people are staying on LTE more rather than switching to Wi-Fi. The available data is very limited and has many caveats, but the projections here are consistent with what I see.  

Moore's Law has at least three more generations, but they seem to be coming more slowly. Router and other equipment prices seem to be falling at a slightly lower pace. The fall in traffic growth in landlines is timely. The decline in the cost per bit is also coming down. The best guess is that decline in cost will remain about equal to the growth in demand, leaving the cost per month per subscriber roughly unchanged.

Cisco notes people are watching more TV over the net in evening prime time, so demand in those hours is going up somewhat faster than the daily average. This could be costly - networks have to be sized for highest demand - but is somewhat offset by the growth of content delivery networks, like Akamai and Netflix. (Google YouTube and increasingly Microsoft & Facebook have built their own.) CDNs eliminate the carrier cost of transit and backhaul. They deliver the bits to the appropriate segment of the carrier network, reducing network costs.

Spend a few hours with the Cisco data if you want to understand what''s happening in networks. Here's the crucial chart.  

  

Table 1.      Global IP traffic, 2016–2021