Latency, clouds, and how the telcos hope to take over
The 1 ms latency frequently touted will be available to less than 2%, probably much less. 300-600 million homes will be able to connect at 15-25 ms through carrier edge cloud networks. Most will be in China. (This and the other predictions are well-researched but of course not guaranteed.)
Deutsche Telekom is rapidly building a network to connect all Germans to servers within 20-25 milliseconds. Multi-player Pokemon has a great demo. DT is working with rivals. Timotheus Höttges and team have decided a project like this needs to reach the largest audience and are including rivals.
Carriers have monopolies on reaching their own customers. Many expect to charge a high price to use their edge clouds, which will be 10 ms to 30 ms faster than other choices such as CDNs. It is totally unclear whether the market for that kind of service will be large. I am more sceptical than most but very unsure.
They are already making deals, such as China Unicom with Tencent. Verizon is building game servers. The partnerships are logical, if and only if the terms work.
Telcos dream of a wireless umbilical cord between your phone or autonomous car and their servers
A modest, less expensive phone connected to a cloud server could be more powerful than even a top of the line phone. Your phone would then be able to run AR/VR apps that are far more demanding than a phone could accommodate. Two years ago, high-end games and autonomous cars were the primary arguments for 1-4 ms telco networks.
So far, all of the networks have decided to build 15-25 ms networks or higher because the faster networks are too expensive. Some expect that to change.
More about networks
Carrier costs will continue plummeting. Verizon's estimate is 40%/year, 90% from ~2017 to 2023. Some of that will pass through to consumers in the form of much higher bandwidth caps, many over 100 gigabytes/month. Carrier prices are mostly determined by competition, hard to predict in the future.
Many wireless customers will be able to connect at 100-400 megabits down, good 4G or 5G. Very few will be offered the gigabit because most telcos are not building the necessary millimetre wave networks. 80+% of "5G" will be low and midband, not the gigabits possible with mmWave.
Most cable customers will be able to download at nearly a gig. Some will be offered 100+ megabit uploads as full duplex rolls out. Many telco homes will be able to connect as a gigabit with fibre. China has 340 million fibre homes, Fibre in the rest of the world is growing 20%/year.
Carriers are so big most plans take three or four years to build. They tell investors and sometimes others what they plan, so you can predict the networks fairly well. How many will buy the service is a marketing decision, much less certain.
Gigabit nets with 1 ms latency exist in the labs. In 2017, the engineers in 5G thought they would be offering the gigabit widely. In Spring 2018, standard organization 3GPP decided to call LTE hardware with Qualcomm's NR Software "5G." It runs in low and mid bands but does not have the speed or reach. Most telcos have decided customers will be happy with 100-400 megabits. Verizon & Telefonica Germany are building gigabit millimetre wave networks but few others.
Every CTO wants to build the faster nets. If Verizon's better network succeeds in making money, many will follow. The build time will be years, however.
Edge clouds spreading
Edge clouds will probably spread rapidly if the early results are promising. Most will not be at the small cells because they are too expensive. They may be one step back, at the C-RAN that controls multiple cells (10-15 milliseconds) More likely, they will be further back, 15-25 milliseconds. That would require fewer servers and save money.
15-25 ms is the plan at Deutsche Telekom, the largest edge network building in the West. It's the tentative plan of the giant Chinese, who have been "directed" by Minister Miao Wei to cover 90% of China very quickly.
The technical people are excited about building much faster networks but are being held back until customers are identified
No one on the telco side is certain of what the market will want. Budgets are tight and delaying projects.
Google has hundreds of engineers designing multi-location clouds. It is already selling boxes to businesses that are part of Google's cloud but residing at customer sites.
Memory and processing power will come down in price but large networks of edge servers will remain expensive. Google and Amazon internally have multiple levels of servers. The most used data will be cached in the many nearby servers. Less used data might even be on another continent.
Multi-level cooperative networks at scale are extremely complicated. Huawei is the only telecom vendor with claims in public.