On the future of the Internet and everything

According to ITU data, Internet usage reached about 2.2% penetration in the US (2.2 users per 100 residents) in 1993. The figure in 2012 was 81%. The history of penetration is shown in the following graphs.

Screen Shot 2014-01-03 at 1-3-12.17.41 PM

Similar graphs can be drawn for other countries (data is available for 193 countries/territories.) I chose the following set of countries arbitrarily:

Screen Shot 2014-01-03 at 1-3-12.26.25 PM

When looking at all countries, the patterns are fairly self-evident. They nearly all behave the same way with respect to the impulse to move online. The pattern is approximately logistic (observed as linearity in the right-side graphs).[1] The main difference is one of timing.

The global picture should therefore not be surprising. In absolute terms, the penetration was 35% at the end of 2012 or 2.5 billion individuals. The graph of global population divided into Internet consumers and Internet non-consumers is shown below followed by penetration.

Screen Shot 2014-01-02 at 1-2-3.44.11 PM

Screen Shot 2014-01-03 at 1-3-12.55.17 PM

The linearity in the blue line above is consistent after 9% penetration is reached (P/(1-P) ~= 0.1) suggesting logistic growth (far more smoothly than individual country data). Note that I also took the step of projecting the logistic curve that describes the internet’s growth forward a few years.

Based on this history, it becomes possible to project that 2013 penetration reached around 39% or 2.78 billion individual users.

It also suggests that “saturation” could come between 2030 and 2035.[2]

If we look at growth,[3] the inflection point comes in 2016. Thereafter internet user growth will slow.

Screen Shot 2014-01-03 at 1-3-8.33.34 PM

Overall, the growth pulse of the Internet will be approximately 40 years long.[4] We are therefore nearly half-way through.

It might seem an interesting but benign fact, however I must caution that as we’ve seen with smartphones, once the point of inflection is reached, investors tend to flee the sector. That in itself may not be a bad thing but investment flight has deep effects on “ecosystem” participants.

Venture investments tend to feed entrants disproportionately and they are remarkably sensitive to the valuations of public companies (and IPOs). These investment might therefore be in danger in two years. Plan accordingly.

  1. Those that don’t are anomalies which should be inspected individually and might provide insights into the constraints for systemic state health. []
  2. The curve is defined as having a half-period of 19.7 years from 1996 []
  3. Derivative of logistic curve []
  4. More likely 35 years as saturation dynamics will be erratic. []
  • stabba

    Nice made-up numbers in your forecast!

    • marcoselmalo

      What do you mean by “made up numbers”? Are you a non-native English speaker that doesn’t know the meaning of “forecast”? It means a prediction. Horace looks at historical numbers to spot trends. Then he projects forward in time based on those trends to arrive at his forecast. It’s really quite simple.

      • rational2

        Wow easy on non-native English speakers. You can be a native English speaker and not know the difference between a forecast and ” made up” numbers.

    • Sharon Sharalike

      The actual numbers are not yet available, being in the future and all.

  • Walt French

    Not quite sure about this ~ 4 “basis points” (per year, I presume) growth. In my world, a basis point is 0.01% and the maximum growth rate looks to be quite a bit more than 1% every 25+ years.

    • Horace Dediu

      Foiled yet again by changes in axis labels without changing the title. The title should read “percentage points of penetration growth).

  • Walt French

    “…once the point of inflection is reached, investors tend to flee the sector.”
    I’d love to see any of the data underneath this claim. It’s obviously terribly important to many of us.
    I would think the underlying mechanics were that in previous penetrations/saturations, by the inflection point, companies had more or less carved out their competitive roles/positions, and only incremental innovations occurred.
    But in smartphones, I’m of the belief that we will use them in substantially different ways in 5 years, and that only a couple of the current major brands will still be around; they will have had to follow the sharp shifts in usage, much of which will come thru the Next Great Service.
    (I’d support anybody who wanted to claim that we’ve more-or-less been-there-done-that in apps, however.)

    • poke

      Anecdotally, it does seem that investors flee in the face of slowing growth. I think the bigger issue is how to interpret “sector”. Flee what and flee where?

      • Horace Dediu

        Flee technology and buy gold or porkbellies or the S&P 500.

    • handleym

      I’d put the issue slightly differently.

      Having “internet” is not a single thing. Even after the population is hooked up with some form of slow wired connection, there is still plenty of scope for speeding up that wired connection, making it locally wireless (i.e. WiFi) and generally wireless (i.e. cellular data).

      There’s also scope for the big thing I’ve been hoping for for a while, from either Apple or Google (or even MS — innovation from MS, wouldn’t that be something! or Yahoo reliving its glory days as an ISP) which would be federating WiFI base stations so that for “members” of the network you could simply move from one federated base station to another, across cities and even across borders, and everything would work as smoothly as moving from one cell tower to another.

      There obviously WILL come a time when global internet infrastructure growth slows. But I don’t think it will be as soon as Horace says because “penetration” is not the end of the story.

    • Sharon Sharalike

      “I’d support anybody who wanted to claim that we’ve more-or-less been-there-done-that in apps, however.”

      In the early 1980′s, when the Apple II and Atari 800 roamed the Earth, I was told “All the programs have already been written. Programming is a dead-end job.”

      • Walt French

        Point taken. But I’d still say that the innovation in smartphone apps is occuring at a declining rate.

        Of course, I have no metric for this.

      • handleym

        Compute. Communication. Context.
        Compute is basically solved.
        Communicate is solved (if you stick to one country — it’s still a major PITA the more you cross borders).
        Context is only just getting started with things like Google Now (and I assume soon enough an Apple rejoinder).

        Or to put it differently, yes, there are apps you can run on ONLY your phone, and that’s useful, but like you say, innovation is slowing down. But apps that are linked to your home computer, your online life, others devices (like health monitoring) are only just getting started. Things like Witness or Dropcam in home security (or similar sorts of things as baby monitors) work, but are clearly first gen products — using them you are constantly reminded of how much smarter they could be and how much more you want them to do.
        Likewise linking to autos is only just getting started.
        Likewise a new payments infrastructure.
        Likewise microlocation to augment GPS.

  • normm

    Growth of new internet applications isn’t very sensitive to overall growth of the internet: it’s the existing internet user base that is being colonized by new ideas. We see the same thing in smartphone growth, which has been very fast even though the overall cell phone subscriber base has hardly grown at all.

  • Nigel Hietala

    Makes you wonder if internet penetration might happen faster than a standard logistic s-curve. Unlike the roll out of electricity or fridges we have seen both the delivery infrastructure drop dramatically in cost as well as its ability to be deployed over huge areas with ease (the move from physical cables to wireless networks), while also making use of internet services has become drastically easier (move from clunky desktop browsers to smartphones and tablets that don’t just have easier to use browsers, but also frictionless applications that utilise the internet).

    • Walt French

      The logistic function can be modified to go from zero to any maximum, and to do so at any rate — eg, 3 weeks from 10% to 90%, or 25 years from 1% to 5%.

      No question that the revolution won’t be televised because television is no longer relevant to the revolution.

  • charly

    Internet == webbrowsers

    They didn’t exist in 1990 so why start the graph in 1990 and not in 1992

    • Sam

      The Internet started before web browsers with Usenet, Prodigy, Compuserve etc…

      I think you meant the WWW not the Internet? They aren’t the same, check your Wikipedia.

      Some of us here remember the days before Netscape and Internet Explorer.

      Some people communicated, and engaged in online commerce before web browsers.

      I bought airplane tickets on Prodigy years before Netscape or Internet Explorer existed.

      • marcoselmalo

        In my youth I used to arrange to buy drugs on a local BBS chat room in San Francisco. Same day delivery!

      • charly

        Public internet is not so much older and i have used mosaic (june 1992) and even used gopher (late spring 1991) so i’m starting to feel old. But i think that using 1992 as the starting year for internet/the web is correct

        ps. dates are first release dates according to wikipedia

      • Steven Noyes

        Mosaic was not the first Web Browser.

      • charly

        It was the first widely used web browser and the first one i used.

    • Steven Noyes

      WorldWideWeb was introduced in 1990 but even then, there was internet before WWW.

      • macyourday

        You guys cannot resist “debating” with the parallel dimensions. Reality and logic are different there.

    • rational2

      Web=web browsers. Before that, we used usenet to access the internet. As a matter if fact, I found my first job on the usenet and emailed my resume. That was in 1991.

  • charly

    What is the big drop (5%) between 2008 and 2010

    • Steven Noyes

      The obvious answer is the “Great Recession”. One cause might be the internet was a “luxury” to those that lost their jobs and was a monthly service that was let go to conserve money.

      • charly

        A 6% drop is a lot for such a necessity (you can’t find work without Internet)

        ps. 5% in penatration but 6% drop in people who use it

      • Sacto_Joe

        It’s not a drop in use. It’s a drop in growth. Big difference.

      • charly

        Sorry but i think it is a 6% drop in usage, not growth

      • Sacto_Joe

        The chart is clearly labelled “Internet Growth Forecast (percentage points of penetration growth”.

      • charly

        I was looking at the internet penetration chart of the US.

      • Sacto_Joe

        Got it. Apple’s and oranges! And that chart does register a drop coincident with the Great Recession. Clearly, all those layoffs reduced access to the internet. The computer I used is still not back in use, as far as I can tell.

      • Steven Noyes

        I have NEVER used the Internet to find work and I am in software.

      • charly

        Your network is the most important way to find work for most people but if you need to economize to such an extent that you need to cut your internet connection then you should use more than purely your network which coincidently needs internet for connectivity

      • Sacto_Joe

        I think you’re on to something there. Couple that “slowdown” in growth with what appears to be a linear increase in population, and the peak growth year may be farther out than two years.

        But note also that this growth in internet mapping doesn’t differentiate download speeds. So right behind the general use of the internet is coming a second wave of wireless, and right behind that is a third wave of much faster wireless, which is the wave that Apple has synched into for its products and the more sophisticated jobs they do. IOW, the difference between feature phone and true smartphone (or, in Apple’s case, geniusphone) mirrors the growth of this third wave of internet access.

        It’s also worth mentioning that, with fast wireless and proliferation of the Cloud, “thin” devices like the Chromebook start to look more attractive. Jobs are served as needed, reducing the cost of the actual physical device. (This cost reducer could be extremely important for reaching people who might not otherwise be able to afford to jump on board.) Apple is clearly aware of that, which is why they’re moving to increase their Cloud capacity.

        Hmm. It would be interesting to map the growth in worldwide Cloud capacity as well.

  • Accent_Sweden

    I feel it wrong to view Internet usage growth potentially limiting venture investment potential, unless you are exclusively talking about companies like Comcast that provide Internet service. After all, few people or companies make their money from Internet infrastructure or number of users online; it’s the possibilities that this infrastructure opens that make money and profitable investments. New innovations will continue even if usage reaches 100% saturation. Internet usage saturation allows more people to use innovations but doesn’t limit the potential for new innovations to gain marketshare or make lots of money. Saturation isn’t a limiting factor, it is a maximizing factor. It maximizes potential, it doesn’t limit it. A fine line, but important. The Internet isn’t a product like refrigerators or buggy whips. Infrastructure in place, access to the infrastructure mean the largest possible audience is available for innovations and those innovations will be what rise and fade as business opportunities. Smart phones aren’t the Internet, they are a way of accessing it and interacting with it. Access methods will change over time, but the Internet will remain and the potential it offers will continue undiminished.

    • Walt French

      These are good questions, but it’s quite possible that when adoption of “the internet” is well below saturation, capitalists will deem that the best opportunities for huge upsides — the “moonshots” that Andreesen mentioned in an interview recently — will have all been considered, and either discarded or tried.

      No, this is not ironclad logic, but it’s not hard to see that VCs might well use it as a proxy, explicitly or implicitly.

      I still want to see the info on it. If this has happened before, the data exists. If IPOs in tech space are slowing dramatically, that data exists, too.

      • Accent_Sweden

        Agreed, they may do so but I still think it illogical. Do we stop investing in companies because the Interstate road network is fully completed or that the power grid has reached 100% of a population? There also has to be an alternate place of investment for all that money before it drains from the Internet economy. I feel it far fetched and the anology tenuous but am open to examining the evidence.

      • Tatil_S

        Wel, there are very few car start-ups nowadays, compared to the much more numerous ones during the early 20th century, when there was not a good interstate highway network.

      • Accent_Sweden

        I think the car analogy fits better with internet providers than startups in general. Cars allow us to access the resources along the roads (restaurants, offices, recreational facilities, movie theaters, etc.) and to take us from A to B. But investors don’t avoid investing in restaurants, companies using offices or recreational facilities because the road network is complete, at least not that I’m aware of.

      • Tatil_S

        If you expand the analogy from highways to all roads and every physical good or service that depends on roads for something or another, it becomes fairly useless. This is akin to saying GDP of the US did not stagnate after we built a lot of roads.

        If you keep the analogy to the highway network and businesses that directly depend on the highway bringing customers, such as gas stations or pit stops on the highways outside major towns, you’ll see that the activity in business creation in those fields have stagnated. For example, almost all gas stations are part of nationwide franchise networks and I know of no new chains that started in recent years.

      • charly

        The introduction of unmanned gas-stations allowed the formation of new franchises but shifts in gas station rarely happen

    • SubstrateUndertow

      What job is the internet hired to do ?

      I think the internet is mainly hired to provide hitherto unattainable levels of far flung, organic, distributively-networked, synchronization. These new complex organic levels of commercial/social synchronization are just beginning to squeeze new levels of efficiency and attendant profits from our antiquated, abysmally inefficient, system of global social-commerce.

      What do I mean by abysmally inefficient system of global commerce.

      For starters, what percentage of potenial productive power sits idle because we have yet to synchronize an effective cyclical homeostatic relationship between productive-potenial, effective-demand(demand with money not just credit in its pocket) and valid money/credit systems that properly track and map paper assets on to their underlying goods and services in near realtime.

      Then there is the build-not-to-last global boondoggle that sucks up about 80% of all industrial productivity into wasted human and natural resources.
      Why build one fridge/car when 5 are myopically profitable for the few at the global expense of the many and the planet?

      Sure most would say I’m conflating more efficiently synchronized industrial productivity with unrelated, long standing, intractable political realities.

      But thats my point here !
      The job the internet is hired to do is bring complex organically-interdependent synchronicities previously unattainable to bear across the whole integated spectum of human endeavour.

      That emergent organically-interdependent synchronicity between newly networked political structures/values/institutions and reformulated win-win organic capitalist commercial productivity seem unstoppable given the immense disruptive-profits available to be synchronized into existence by progressive players.

      • Walt French

        Let’s contrast the internet to the “agriculture” technology. It’s gone thru several revolutions (and the practice is wildly different around the globe, suggesting more are to come as proven technologies are adapted to very different locales).

        Agriculture in the US has gone from being the dominant occupation, to being a minor contributor to the dining value chain. (Apologies if this sounds hackneyed; I’m not really schooled in this disruptive technology stuff.) There are indeed investments made into finding new, “natural” ingredients such as Stevia, and more into GMO, etc. But probably many, many times more R&D—emphasis on the Development—of food combinations, packaging & shipping, a whole host of activity/effort that builds upon the basic agriculture, which seemingly doesn’t need capital-market based ownership or R&D.

        Doesn’t seem like a bad future if the internet becomes so ubiquitous and stable (like water or electric utilities in developed nations) that all the energy goes into value-adding activities atop it.

  • Walt French

    Maybe others know the practical aspects of the logistic curve better than I. But recalling the Central Limit Theorem (a relative of the Law of Large Numbers), I looked at how various mechanisms might produce a logistics curve.

    My first try at a model said that a certain percentage of non-users see a user and adopt the tech—penetration “p” at time T is the previous period’s, plus X times the previous p times (1-p). I suppose a certain percentage of previous users could drop out or other changes could be worked in. But the curve fits superbly well.

    There must be other mechanisms that would also work well. (Maybe, if I were Gauss, I could prove it; maybe at the limit of time increments instead of the 90 steps I did up in Excel, it’s exact.)

    But for those who are looking for an easy rationalization of “why this goofball exponential/logistic function?” questions, this ought to serve well. Once dreamers produce an idea (the function needs to start with a non-zero number!), it’s taken up by the simplest possible mechanism.

    • Ian Ollmann

      If there is some probability Pc that a non-user will become a user per time quantum provided that he has seen someone else use the technology and liked it, then we can write a rate equation:

      dFc/dt = (Pc * Fn) * Fc;
      Fn = fraction of non-consumers
      Fc = fraction of consumers

      For simplicity, we’ll assert that once you are a consumer of a technology you don’t go back.

      We also claim that you are either a user or a non-user, so there is a mass balance equation:

      Fn + Fc = 1.0

      Some algebra:
      We substitute our mass balance equation into the earlier rate equation, to remove the Nn term:

      dFc/dt = Pc * Fc * (1-Fc)

      …which has the form of a logistics function. Its integral over time to get the instantaneous fraction of consumers has the form f(x) = 1/(1+exp(-x)).

      Intuitively it should have this form because in the initial growth stage, adoption of a technology catalyzes other users to switch. This causes exponential growth. As the technology matures, the number of remaining non-consumers gets very small, so the number of conversions per time unit drops off and the curve flattens out again.

      • Walt French

        Thanks for confirming my hunch that I didn’t have time to develop before I jumped on a plane (and I only have a few minutes before I leave NRT).

        Just to underscore the point, the model is sensible although obviously simplified. The chance that a non-user will contact/see a user and ALSO adopt the technology can’t really be a constant function of their numbers, but apparently the world is full of people who are actually quite homogeneous in goals, enough so that despite being hetero regards income, access to networks and other measures of ability to adopt, they seem as if they were all of one mind.

        Likewise, the model assumes that “the internet” is *a* thing, while handleym notes how its usage is still changing; late adopters will confront a rather different set of choices than earlier ones did.

      • Ian Ollmann

        Really what we are looking for is a feedback mechanism whereby users beget more users. This is what gives you early exponential growth. I related an anecdote about users seeing other users using the product, but clearly that is not the only way this can happen. (Advertising agencies will no doubt tell you this is critical, but they can do it better!) There are also well documented network effects of having more people on a platform, which increases the value of that platform. I am certain you can come up with more.

        The other key feature is that towards the end of the growth curve the rarity of non-users becomes dominant. This gives you the 1-f(x) portion of the logistic curve. It should be fairly self-evident.

        I think that the ubiquity of these two conditions, self-reinforcing growth and zero sum populations is why the logistic curve is so common. Why should exponential growth be ubiquitous? It probably isn’t. However, if you don’t have that, the growth likely won’t be noticed because it is too slow.

      • Horace Dediu

        The history of logistic functions is interesting.çois_Verhulst

      • Walt French

        {Disqus mangles your link. Until Disqus gets smarter, you can replace c-cedilla with a simple “c”.}

      • Daniel Tello

        Bacteria grows faster (in a limited environment) when there’s more food and less competition for it. Consider the proportion p of bacterial mass to total mass (bacteria and food) in the petri dish, then the relative growth of p at any given moment, p’/p, is proportional to the remaining ratio of food mass 1-p.

        A contagious disease within a limited population will propagate faster as there are relatively more healthy individuals to infect. If p is the proportion of sick individuals, the relative propagation rate p’/p is proportional to the remaining ratio of healthy individuals 1-p.

        Consider the ratio p of cancer cells to all cells within some constrained tissue or organ or individual. The relative rate of growth of the cancer p’/p is proportional to the remaining ratio 1-p of healthy cells.

        In all cases, the equation is logistic: p’/p=r(1-p) where r is constant.

        The adoption of some technology (or any idea) is just like bacteria, or a virus, or a cancer, “eating up” or “infecting” a limited population of users.

      • Ian Ollmann

        > they seem as if they were all of one mind.

        I’m sure there is a basic tenet in both basic economics and statistics that says that is in fact exactly what happens. Nobel prizes are awarded for people who show it isn’t true.

  • Jack

    Penetration would peak, but total data consumption is nowhere near its peak. So then why would investors flee tech/internet when the goldmine is really the consumption? I have statistics, but just a hunch.

  • Rogers

    Been thinking about your penultimate paragraph since this article’s publication ….

    Your article is about population using the internet, but titled “… Internet and everything.” You show that the internet user population growth will slow, but like others here, it appears to me that devices/uses of the internet (“the Internet and everything”) are growing ever faster.

    Maybe this foretells simply a shift in investing activities to where the growth is?

  • Arley Andres Cespedes Herrera

    which is the mathematical model used for forecasting?