AI, Deep and Machine Learning Matter, But How Much For Typical Telecom Professional?

It is difficult to explain why deep learning, machine learning or artificial intelligence is relevant for nearly all who work in the telecom industry, beyond the ranks of data scientists working at or with chip, app and platform suppliers creating services and products used directly by network operators.

That is not terribly unusual. Machine learning already is an underpinning of most major consumer internet apps (content, social networking, search) as well as a growing range of business apps.

That is similar to the way consumers have learned to use cloud computing, without knowing how their apps are based on use of the cloud. But that “invisibility” also makes hard the challenge of figuring out whether “most in the industry” actually need to know very much about machine learning, deep learning or AI.

“I have very little to say that would be of interest to a telecom audience,” one data scientist working on machine learning and deep learning told me, for example. He probably is quite right.

People who work in outside plant can do their jobs quite well without understanding much of anything at all about marketing, billing systems, business strategy or industry trends.

Executives who work in the undersea networks segment of the business really do not need to know much at all about consumer demand for retail apps.

Most people who work for fixed network firms do not encounter any real need to understand mobility issues and vice versa.

That arguably is less true for C-level executives, though, who arguably might have to understand AI implications for their cost structures and capabilities.

At a practical level, a goodly number of people might already have job responsibilities where machine learning already is at work. That might already be operationally correct for people who work with software defined networks, network virtualization, customer service and analytics, for example.

That might be the “problem” with general purpose technologies: they are ubiquitous (like electricity, batteries, steel, airplanes, voice or other communications), and hence are invisible. People only need to know how to use product built on them. They rarely, if ever, need to know how they work.

For “C” title execs, AI might actually matter, in terms of understanding how AI might help create and change revenue models, operating costs, product value and “up the value chain” opportunities. That is what general purpose technologies generally enable: disruptions of whole economic systems. I,

Do Bananas Cost Too Much?

Even if some observers seem always to complain about how much internet access costs, that is a bit like complaining about bananas costing too much. Costs now have gotten so low (consumer or business internet access) that it is cumbersome to even bother using a metric such as cost per kilobit (kilobits per second).

We moved more than a decade ago to cost per megabit per second. Now we are moving to measuring in hundreds of megabits per second. In the future we will measure in terms of gigabits per second.

Around 1995, the cost of buying a business connection supporting a kilobit per second might have been US$1.50 to $1.75. In other words, a 56 kbps connection might have cost as much as $98 a month.

By about 2006, even consumer internet access costs had dropped to about two cents per kbps. So a 10 Mbps connection might then have cost the same as the 56 kbps connection of 1995. In 2017, U.S. 100 Mbps connections cost about the same as a 56 kbps connection of 1995.

As speed has grown and apps have evolved, consumers now use more data (megabytes), so the cost per consumed megabyte also has fallen, even as people use more data.

Source: Boston Consulting Group

While complaints about high prices never seem to stop, in developed markets as well as the United States, the percentage of disposable income spent on fixed network internet  access  is about 1.7 percent of gross national income per person.

New Role for Fixed Network: Backhaul and Feeder Functions

Going forward, communications infrastructure is going to take a perhaps-predictable path, many will argue.

At a high level, retail communications--globally--has become “mobile.” Though the role of fixed networks as a retail platform will remain more significant, dwindling retail revenue will force changes, including a huge need for lower-cost infrastructure, as revenue potential will be less robust.

source: Ericsson

Investment and revenue growth, on a global level, will shift to developing regions. The fixed network increasingly will take on the role of backhaul and trunking mechanism for mobile and wireless networks.

And despite the historic preference of fixed network operators for “smart” infrastructure, much of the global transport and access platforms will become simpler, with more intelligence at the edge, and less in the core.

For that reason, fixed networks will have to cost less.

At the same time, dwindling legacy revenues will force service providers to create big new revenue sources, sell themselves to larger providers or take other steps to match lower revenues with higher stranded asset profiles.

Critics of moves by AT&T to buy Time Warner, or Comcast’s past acquisition of NBC Universal argue that too much power or control results. In fact, we are likely to find that such acquisitions only help access providers reposition themselves for a market where traditional access services no longer support the core business model.

Will U.S. Fixed Wireless Access Revenue Be Bigger than U.S. Consumer Satellite Revenue by 2021?

Will the U.S. fixed wireless business generate more consumer service revenue than the U.S. satellite industry by about 2021? Possibly, if fixed wireless grows as much as some predict, and if one does not double count AT&T revenue.

Keep in mind that both fixed wireless and satellite are tails on a big telco dog. Total U.S. mobile service revenues are on the order of $256 billion annually, with fixed telco revenues perhaps $100 million and cable TV revenues roughly $60 billion, representing annual industry revenues of at least $455 billion. Others might prefer higher figures.

Any industry segment with annual revenues in the single-digit billions represents a fraction of a percent of total industry revenues.

U.S. fixed wireless revenue will double from $2.3 billion to more than $5.2 billion, according to the Carmel Group. Accounts also will grow from four million at the end of 2016 to eight million by 2021.

source: The Carmel Group

Depending on how one chooses to count revenue, the fixed wireless industry might then rival satellite service provider revenue. One big issue is whether “DirecTV” revenue should be counted as “satellite” revenue or “telco” revenue. In fact, both industries claim that revenue.

Assume DirecTV generates about $36 billion in annual revenue for AT&T. Assume all U.S. “satellite services revenue” is about $52 billion, including $40 billion of satellite TV revenue. Take out DirecTV and U.S. satellite revenue (booked in the telecom industry as well) is about $16 billion, with perhaps $7 billion in consumer service revenues.

In that scenario, the U.S. fixed wireless business would generate as much, if not more revenue, than does the “consumer satellite” business.

source: SIA

Why Price Anchoring and "Price Per User" Might Explain Antipathy to Linear Video

In the internet era, where the expected price is "free," prices north of zero might be one reason why many people complain about linear subscription video prices. There is a big difference between $10 a month for Netflix and $80 for linear video, even if the products are not full product substitutes.

It is hard to tell whether perceived value or price anchoring are bigger drivers of dissatisfaction with linear TV services, even if the stated reasons for unhappiness tend to center on “customer service” issues, in addition to price.

People often complain about price increases for cable TV services that go up more than the general level of inflation. Yes, that is an issue. But consumer prices have been climbing since 1970 or so.

source: Bureau of Labor Statistics

I cannot remember a time (over three to four decades) when consumers actually reported “liking” their cable TV services. Most often, they tend to rank them very low, when compared to most other consumer products.

These days, there are differences between providers. In recent years, telco TV and satellite TV services actually have produced more satisfaction than cable TV video services.

But the service as a whole still ranks quite low, along with internet access. Perhaps price anchoring explains some of the unhappiness. Most children growing up these days experienced internet access, mobility and video as “being free,” since their parents paid for those services.

So the anchoring price is “free.”

On the other hand, many adults who have been buying internet access for decades might have distinctly-different price anchoring experiences.

I recall paying AOL perhaps $15 to $20 a month for 56 kbps dial-up access using the voice line I already was buying.

I can remember paying more than $100 a month but less than $200 a month for a 756-kbps internet access connection. Recall that about 1996 a T-1 line supporting 1.5 Mbps cost as much as $2500 to $3000 a month.

Now I expect to pay $50 or so for 100 Mbps. So my price anchoring is $150 for 756 kbps in the past and now 100 Mbps for about $50 a month.

That is a different set of expectations from many others who have moved from “free in the past” to $50 a month now.

Perhaps the same experience now anchors pricing expectations for potential customers who are moving from “free” to $80 a month.

So I suspect, “value” explains the hostility many single people seem to feel about buying any linear video product (“cable TV”). In other words, the price-per-user is far lower, in a four-person family, than in a single-person household.

In other words, an $80 a month subscription “costs” $80 per viewer in a single-family household, but only $20 per viewer in a four-person household. Though use of streaming services is growing in all age brackets, that might be one reason why families, larger households or households headed by older persons tend to buy linear video streaming services at higher rates.

Also, despite the fact that linear video prices have risen faster than the background rate of inflation, lots of other products have risen at “above inflation rates” as well.

Consider the cost of college educations.

source: Mybudget360

source" Bureau of Labor Statistics

Medical care and housing are other products whose prices have increased more than the general level of inflation, as well.

source: BLS

IoT Networks Will Be Many-to-One

The architecture of coming internet of things networks is going to be radically different from the voice, consumer internet or business data network architectures. Those networks were built to support “point to point” communications from any node to any other node.

Irrespective of the access mechanism (mobile, Wi-Fi, other), the fundamental networking architecture of a sensor network (and therefore of  the internet of things) is “sensor to server,” not “point-to-point,” as was the architecture of the voice network or the early internet.

source: Aeris

That is going to have consequences. The IoT networks will be the inverse of the point-to-multipoint satellite TV, cable TV, radio and TV broadcast networks.

In contrast with point-to-point, any-to-any networks, these broadcast or multicast networks are optimized for one-to-many communications in the downstream direction.

IoT networks will be optimized for many-to-one communications upstream to servers. If there are scores of billions of devices using such networks, then such many-to-one networks will be optimized for collecting data from remote locations, not distributing information to many remote locations, and not for any-to-any communications.

That is new, really new.

Much of the upstream data will be simple: where are you; what is the temperature; what is the pressure; what is your speed. Some of the data might require more bandwidth, supporting camera feeds, for example. But almost none of the requirement is for downstream communications to sensors.

source: Cisco

How Big Will Satellite IoT Be?

All executives from mid-size to enterprise firms in agriculture, surveyed by Vanson Bourne on behalf of Inmarsat, believe they will be using  internet of things apps and services within about five years.

Some 85 percent of executives in transportation industries believe they will be using IoT apps and services within five years, and nearly that many in the energy business. In the mining business, perhaps 65 percent of respondents believe they will be using IoT in five years.

You are not likely surprised by such findings. Given the hype around internet of things, it would be a rare executive who thought it “sounded right” to project no use of such capabilities.

Neither are you likely surprised that Inmarsat might sponsor such research. No less than other segments of the communications industry, satellite firms argue they will be relevant for IoT, largely because of coverage capabilities. Basically, that is the newest form of the “no one platform is best for all scenarios” argument.

It remains to be seen which satellite implementations are workable for which IoT apps.

“Despite growing demand for IoT satellite services, the business case for IoT exclusive satellite constellations has yet to be proven, especially considering the exponential growth of LPWANs, LTE-M and NB-IoT terrestrial networks for IoT,” says Alan Weissberger of the IEEE ComSoc blog.

Obviously, geostationary platforms are going to have huge problems supporting ultra-low latency apps. Low earth orbit (LEO) constellations arguably will fit a wider range of missions. Low-bandwidth, non-time-sensitive apps might fit best, for any satellite platforms.

Satellite market researcher NSR predicts transportation apps will be the biggest application.

source: NSR

Broadcast (multicast) has been the geostationary satellite network’s fundamental architecture, so multicast apps will make sense, though that is perhaps not an IoT app at all. In other cases, some argue satellite has a role for connected car communications, possibly for multicast use cases.

Some will note that satellite presently serves a niche role in end user communications, serving ships at sea and remote areas where other networks do not reach. That is likely to be the case for most geostationary satellite future apps as well. It is not yet clear how big a role LEO constellations will have.

Expect to hear a lot of promotion from the satellite industry about that industry’s role in internet of things. It will be small, it is safe to say. Satellite researcher NSR projects satellite industry revenue from IoT at less than $3 billion globally by 2026. And that might be highly optimistic.

source: NSR

Some estimate the global IoT connectivity market will be an insignificant fraction of total IoT revenues by about 2025, NSR projects global satellite revenues from IoT might reach about $3.5 billion by 2026.

In that same year, total global satellite connectivity revenues might represent less than $20 billion in revenue. So NSR estimates IoT will represent as much as 17 percent of total satellite industry revenues.

By about 2025, some estimate mobile operator IoT connection revenue of $22 billion. Total global communications service provider revenues should be about US$1 trillion by then.

Also, projections for service provider enterprise connections must contend with the possible use of Wi-Fi or other access mechanisms which do not incrementally increase connection revenue in a direct sense, if at all.

The bottom line is that satellite IoT is at present as uncertain as mobile or fixed network IoT connection revenue might be.