5G: if you build it, we will fill it


In early 2000, right at the top of the dotcom bubble, the mobile bubble and the broadband bubble, European mobile operators spent €110bn on licenses for 3G spectrum. Now, almost 20 years later, I’ve just got back from CES, and 5G is a Topic. Many of my friends at big companies tell me that ‘what is 5G?’ floats around a lot of corporate headquarters almost as much as ‘what is machine learning?’ does. 

There are a bunch of different ways to answer this. If I was still a telecoms analyst, I would be spending a lot of time thinking about spectrum, deployment schedules and capex - mobile operators around the world spend several hundred billion dollars a year on network capex, and 5G will become a big part of that. I’d talk about network efficiencies, refarming, vendors, Huawei, chipsets, and maybe NFV. But I’m not a telecoms analyst anymore - I work in Silicon Valley. So, seen from Silicon Valley, I think there are maybe four things to talk about.

First, what actual changes should we expect?

Without going into the technical details (any more than absolutely necessary), what do we actually get from this? A fatter pipe.

  • As with each previous generational change, 5G makes it cheaper and easier for mobile operators to build more capacity. So, they can continue to accommodate growing usage.

  • 5G will be deployed on existing cellular radio frequencies, but also lets operators address much higher radio frequencies (over 20 GHz, AKA millimeter wave or ‘mmWave’) that have never previously been usable for mobile services. (This will also require the installation of many short range base stations.)

  • Mainly because of this new spectrum, mobile 5G speeds in good conditions could be well over 100 megabits/sec and potentially several hundreds megabits/sec (mobile speeds of over a gigabit/sec are technically possible but unlikely in the real world).

  • However, deployment in this frequencies, and hence these eye-catching new speeds, will be possible only in pretty constrained areas and will happen pretty slowly. Signals at such frequencies have worse range and don’t go through walls ( to simplify hugely), so don’t expect these speeds in rural areas or indeed inside buildings (this is why they have not previously been used for mobile at all). 5G deployment on more conventional mobile spectrum will have speeds (and coverage) closer to 4G

  • 5G is promised to have much better latency than 4G - perhaps 20-30ms in the real world, down from 50-60ms for LTE (4G). It’s not clear how visible this will be to users.

  • Some people (eg Verizon) think that you can also use 5G in these higher frequencies for a home broadband service (which would mean an antennae on the outside of your house, or in a window), offering up to a gigabit/sec. There is also a fair bit of skepticism about both the economic and technical cases for this. Of course, the newest version of DOCSIS (cable internet) offer much the same - something around a third of the US population could have access to these speeds anyway.

So, mobile gets better latency and the mobile pipe keeps getting fatter. Fixed broadband will get more competition, in some places.

Second, what does it mean to have steadily fatter pipes?

The internet first took off with dial-up, and the first consumer mobile internet service to get mass adoption, NTT DoCoMo’s i-mode, had 2G data speeds - both of these gave us tens of Kbits/sec at best. DSL and the first deployments of 3G gave us a couple of hundred Kbits/sec. Then improvements to 3G (‘3.5G’) and then 4G gave us tens of Mbits/sec (and also much better latency) and improvements to DSL and DOCSIS gave us fixed home broadband speeds in the tens or low hundreds.

With each of these surges in speed, two things happen. First, the things we’re already doing get smoother and easier and quicker, and also get more capable (or bloated). Pages get more images and become more dynamic. Second, new things become possible. You could not have done Flickr or Google Maps on dialup, and you could not have done Netflix (or at least not well) on the broadband of 2003. In the following generation, Snapchat only worked when you could presume that all of your users can connect at tens of Mbits/sec (when they’re not on home WiFi, of course). That in turn means networks with the overall capacity to give that speed not just to one person at a time but to lots of people, and network infrastructure that can do that at a vaguely reasonable price. If you’d shown Snapchat to a mobile network executive in the early 2000s, their hair would have gone white - there was just no way the early 3G network could have supported that kind of load. 

In the same way, then, 5G speeds, and ever-faster home broadband, will mean that existing applications will get richer, and also that new applications will emerge - new Flickrs, YouTubes or Snapchats. We don’t know what yet, exactly, though we can make some early guesses, but the creativity of entrepreneurs and platforms and the choices of consumers will decide. This is the great thing about the decentralized, permissionless innovation of the internet - telcos don’t need to decide in advance what the use cases are, any more than Intel had to decide what the use cases for faster CPUs would be.

Third, AR and VR, and cars.

Having said that we don’t know what the use cases will be, there are a couple that do repeatedly come up in conversations around 5G; AR and VR, and autonomous cars. It’s worth spending a little time talking about each of these. 

I think of VR as fundamentally an indoor product - you will not use it walking down the street or pop it open for 20 seconds while you’re waiting for the bus. That means that the connectivity is whatever your home broadband is - DSL, fiber, cable or, perhaps, 5G, plus however you connect to that (i.e. WiFi, mostly). 5G here means two different possibilities. It might mean fixed 5G (with very limited coverage) at up to a gigabit/sec to an antenna outside your home and then WiFi to the headset, or else it means a cellular, ‘mobile’ modem in your device, in which case you will get speeds much closer to today’s 4G LTE (again, 20 GHz signals do not go through walls). You’ll also get 5G latency, which is better than 4G. It 5G better than the existing connection? Will you notice? Is that what VR is waiting for? For how many applications? 

5G seems rather more interesting for AR. To clarify first, ‘AR’ today is used to describe three different things:

  1. Waving your phone at something and seeing things on the screen

  2. A wearable heads-up display (Google Glass) with no awareness of the world around you,

  3. An transparent, immersive, fully 3D color display with a sensing suite that allows it to map the room around you and recognise things and people. A bunch of companies (including Magic Leap, in which a16z is an investor) are working on this - it’s still a few years away from being a mass-market consumer product.

The third of these seems much the most interesting to me. If you could put on a pair of reading glasses that could look at the world around you and show you things in response, that could be pretty useful, in much the same way that, say, having the internet in your pocket turned out to be useful, and to enable all sorts of new and unpredictable things (imagine pitching Snapchat when our only internet experience was on a PC over dialup). This would work on 4G, but continuous low power high speed low latency connections from 5G would make it a lot better. 

At the other extreme, I also hear a fair bit about autonomy as a 5G application. I’m not sure about this one. Autonomous cars will certainly use a great deal of data. They will be downloading ‘maps’ (really, very high definition 3D models of the streets they’re driving along) and also updating those maps with data from their own sensors, and they will be downloading updates to their driving systems and uploading more data about how real people drive. However, very little of this needs to happen in real time - it can happen every night or even every week. No mainstream autonomous car project requires continuous connectivity at all, let alone through 5G - the car has to work with no cellular service.

As we look further out, into a world with lots of autonomous cars, there is widespread interest in those cars being able to talk to each other, so that for example they can speed through a junction without slowing down. This might be done with 5G, or perhaps with another wireless technology. However, most of these use-cases only make sense in a world where there are no human drivers to worry about - this is, obviously, a long way away.

More imminent is remote operation as a back-stop for when a vehicle gets confused and stops, or when you shift from manually-driven to autonomous - for when autonomy doesn’t work. This is particularly relevant to trucking: 90% or so of the mileage of a long-haul truck in the USA happens on highways, and highways are much easier for autonomy than suburban streets, so a number of companies are exploring a model in which the truck drives itself on the freeway and a human takes over when it leaves - either by physically getting into the vehicle or by connecting remotely. This would be a good 5G use case. But to begin with, you can do it with 4G. It doesn’t presume 5G. The advantage of 5G comes partly with latency, but also with things like network slicing, which takes me onto the next section. 

Fourth, networking slicing and industrial uses.

One of the cooler features of 5G is that it lets you split out dedicated capacity for particular use cases - so-called ‘network slicing’. Today (to simplify hugely), although network operators try to do traffic management, all traffic in the cell is fundamentally using the same capacity. 5G lets you create dedicated private capacity in the radios network with specific characteristics. So, you could sell a truck operator dedicated capacity on the two miles between a specific freeway exit and a specific warehouse. Or, you could offer an IoT operator (or alarm company) much lower bandwidth but over a wider area.

Hence, you could theoretically customise any mix of data speed, coverage, quality, latency, and reliability, or even more narrow things like power consumption, and there will probably be a layer of resellers that emerges to aggregate and implement these kinds of services on behalf of MNOs. This seems interesting - it also seems likely to be an enterprise and vertical application story, not a consumer story. On this theme, we are already seeing a trend for large industrial organizations to use private 4G networks instead of WiFi or even Ethernet. These issues will also apply to 5G.

So, what’s the killer app for 5G?

In 2000 or so, when I was a baby telecoms analyst, it seemed as though every single telecoms investor was asking ‘what’s the killer app for 3G?’ People said ‘video calling’ a lot. But 3G video calls never happened, and it turned out that the killer app for having the internet in your pocket was, well, having the internet in your pocket. Over time, video turned out to be one part of that, but not as a telco service billed by the second. Equally, the killer app for 5G is probably, well, ‘faster 4G’. Over time, that will mean new Snapchats and New YouTubes - new ways to fill the pipe that wouldn’t work today, and new entrepreneurs. It probably isn’t a revolution - or rather, it means that the revolution that’s been going on since 1995 or so keeps going for another decade or more, until we get to 6G.