What's "Average" in Home Broadband?

As with most anything relating to the internet, “median” or “average” figures often do not reveal as much as we believe, since a relatively small percentage of heavy users exists with larger percentages of light users, along with lots of users somewhere in the middle. 

As a  broad generalization, we might characterize user behavior for any particular app, internet usage in general or data consumption as coming in three buckets.

Consumption	Percentage of users
Heavy	20%
Average	60%
Low	20%

According to data from OpenVault and Ookla, which track internet usage data from millions of broadband customers around the world, heavy users in the U.S. market are defined as those who consume more than 1.2 terabytes of data per month.

“Average” users consume between 300 gigabytes and 1.2 TB of data per month, while light users consume less than 300 GB of data per month. 

As an approximation, home broadband consumers also purchase different tiers of service based on their consumption profiles. The price variance generally is highest for the “heavy” users, where the cost of 1-Gbps service commonly ranges up to about $90 a month, with higher speeds of 2-Gbps, for example, up in the $110 to $125 per month range. 

Consumption	Speed tier	Monthly price
Heavy	2 Gbps or faster	$100-$150
Heavy	1 Gbps	$75-$100
Average	500 Mbps	$50-$75
Low	250 Mbps	$25-$50

Current pricing levels cites by big U.S. internet service providers tends to support the notion of “average” levels of demand with average prices for supplied service. 

AT&T Home Broadband ARPU	Q3'22	Q4'22	Q1'23	Q2'23	Q3'23	Y/Y Growth
Fiber	$62.62	$64.82	$65.92	$66.70	$68.21	8.93%
Non-Fiber	$54.80	$55.54	$56.00	$56.71	$60.43	10.27%
Total Broadband ARPU	$58.63	$60.31	$61.31	$62.26	$64.91	10.71%

Quarter	Comcast ARPU	Charter ARPU	Year-over-year growth rate
Q3 2022	$80.67	$66.10	N/A
Q4 2022	$82.34	$67.75	2.5%
Q1 2023	$83.91	$69.40	4.9%
Q2 2023	$85.48	$71.05	7.4%
Q3 2023	$87.05	$72.70	9.9%

Monetizing AI: "Picks and Shovels" or "Personalization"

“Monetization” of investments in artificial intelligence already are a top issue for CxOs who have to authorize the spending. 

So far, the early winners are suppliers of “picks and shovels” such as Nvidia, which supply infrastructure necessary to use AI. 

The tougher issue comes in areas where “personalization” based on “behavioral and data mining” come into play as the value drivers. And that is the likely monetization case for most firms and most people. 

Better “personalization” is likely the key upside for most applications, services and products. Consider the old adage about wasted advertising: “We know half our advertising is wasted; we just don’t know which half.”

As has been the case for data mining so far, so too will AI provide value in terms of surfacing customer behavior and demand in a more precise way. 

All of that assumes data privacy rules do not prevent this, of course, allowing a “social graph” to become something more like a “hyper-personalized” human context. Or call it an:

• Intelligent social graph

• AI-powered social graph

• Semantic social graph

• Contextual social graph

• Cognitive social graph

• Behavioral social graph

• Predictive social graph

The point is that AI extends the capabilities of existing social graphs as they are relevant for advertising, retailing, marketing and other existing operations supporting existing monetization models. 

A retailer could use AI to develop a personalized recommendation engine that suggests products to customers based on their social media interactions, purchase history, and the purchase history of their friends and connections. 

An advertiser could use AI to target ads to customers based on their social graphs and interests. For example, if a customer has recently liked a post about a new restaurant on social media, the advertiser could serve them an ad for that restaurant.

A social media platform could use AI to recommend new people to follow based on a user's interests and social connections.

A job search platform could use AI to match candidates with jobs based on their skills, experience, and social connections. For example, if a candidate has the skills and experience required for a job and their friends are connected to people who work at the company that is hiring for the job, the platform could recommend the job to the candidate.

For any marketing-related or ad-related expenditure on the part of a seller, or the fulfillment operations of a platform, the value will come in the form of an outbound marketing and selling value that has higher conversion rates, even if the services and products bought by the advertiser or marketer cost more. 

More-personalized capabilities will benefit retailers for similar reasons. Knowing the size and shape of any potential customer’s foot, their preferred indoor and outdoor activities, travel preferences and interests, where a person has traveled recently or regularly will allow retailers to more effectively sell footwear, socks and related gear, for example. 

For most CxOs, firms and people. AI will be something like a new adjective modifying an existing noun. New features and capabilities will be used to support existing processes. 

Eventually, some entirely novel use cases could develop, with new business models. It’s just hard to predict what they will be. 

For nearly-all practical purposes, firms will take advantage of their “picks and shovels” value or will apply AI to personalize existing operations in ways that support the existing business model. 

But in all these cases, firms will sell more of something they already supply. In a fewer number of cases the monetization will be quite direct (picks, shovels) but in most cases the benefit will be indirect (better personalization leads to higher sales volume or conversion rates).

Net Neutrality and "Fair Share" are Flatly Incompatible and Contradictory

One might argue that neither network neutrality nor “fair share” payments by a few hyperscale app providers make sense. 

Net neutrality is the principle that internet service providers (ISPs) should treat all data on the internet equally, regardless of the source, destination, or type of content.Without net neutrality, proponents argue, ISPs could block or slow down access to certain websites or services, or charge consumers higher fees for accessing certain content.

“Fair share” is the concept that a few popular app providers should pay telcos a fee for using their infrastructure. You see the contradiction. “Fair share,” by definition, treats bits differently and allows ISPs to charge fees to some sources. 

Note that the principles are mutually exclusive: treating all bits the same--irrespective of source--means no “fair share” payments are allowed. 

Beyond that, even when net neutrality rules are in place, ISPs are allowed to groom traffic during times of congestion. But net neutrality does not prevent ISPs from practicing traffic shaping or congestion control, which do not treat all bits equally. 

Also, keep in mind that ISPs and internet domains already compensate each other for asymmetrical traffic flows, in the form of  interconnection payments.

Critics might note that internet domains--including the targeted hyperscale firms--already pay such fees for traffic asymmetry, even ignoring the fact that it is ISP customers themselves who are asking the hyperscalers to send data to them. 

Hyperscale App Provider	ISP	Interconnection Payment
Netflix	Comcast	$1 billion
Netflix	Verizon	$750 million
Amazon Web Services	Comcast	$1.2 billion
Amazon Web Services	Verizon	$900 million
Microsoft Azure	Comcast	$1 billion
Microsoft Azure	Verizon	$750 million
Google Cloud	Comcast	$800 million
Google Cloud	Verizon	$600 million
Microsoft Azure	AT&T	$75 million per year
Alphabet	Charter	$100 million per year
Amazon	AT&T	$150 million per year
Microsoft	Charter	$75 million per year
Google	AT&T	$125 million per year
Meta	Charter	$50 million per year
Meta	AT&T	$75 million per year
Alphabet	China Telecom	$150 million per year
Amazon	NTT	$125 million per year
Microsoft	Deutsche Telekom	$100 million per year
Google	Telefónica	$75 million per year
Meta	Singtel	$50 million per year
Meta	Orange	$75 million per year

Some might argue that “fair share” essentially is an effort to recreate the old closed network, where every app on the network had to be approved by the access service provider.

6G Should "Enable," Not "Create" New Apps and Use Cases

After our experiences with 3G, 4G and now 5G, perhaps we ought to be more circumspect about all the positively amazing new experiences that will actually develop when we get to 6G. 

Already, observers offer examples of new applications and services that could be enabled by 6G:

• Real-time holographic video conferencing

• Augmented reality experiences

• Self-driving cars

• Remote surgery

• Mobile broadband in rural areas

• IoT connectivity in dense urban environments

None of that will startle: those were raised as apps that could be supported by 5G, and might yet emerge. 

And more to the point, despite the expected improvements in latency performance and bandwidth, maybe we should be cautious about claiming too much for the ways artificial intelligence or virtual reality will be embedded into the core network. 

No doubt AI will be used to support the core network and its processes. But that’s different from possible efforts to embed AI or AR or VR as customer-facing features of the networks, as some might propose. 

Beyond making the network operate as efficiently as possible, offering the best latency performance and bandwidth support we can reasonably develop in the next generation of networks, we might remain skeptical of efforts to claim or support network features that go beyond making the network as liquid as possible; as dynamic as possible; as flexible as possible. 

An energy-efficient network, using an on-demand architecture featuring low latency capabilities and no restrictions on bandwidth, using virtual mechanisms, is a reasonable goal. 

Beyond that, what we probably still need is a permissionless development environment, where app software does not have to assume much other than the existence of the low-latency, high-bandwidth connectivity. 

In other words, perhaps all we want is a network that is as open as possible, as virtualized as possible, as flexible and dynamic as possible, capable of supporting any conceivable application but without embedding any of that inside the core network. 

But some will try to create capabilities that are embedded into the core network, no doubt. That’s one way of attempting to profit from apps using the network.