In reviewing coverage of the proposed Comcast-TWC merger last week, I came across a blog post in The New Yorker, “We Need Real Competition, Not a Cable-Internet Monopoly” which made the following assertion:
Take the “triple-play” packages—cable, phone, and high-speed Internet access—that tens of millions of Americans buy from companies like Comcast and Time Warner Cable. In France, a country often portrayed as an economic and technological laggard, the monthly cost of these packages is roughly forty dollars a month—about a quarter of what we Americans pay. And, unlike in the United States, France’s triple-play packages include free telephone calls to anywhere in the world. Moreover, the French get faster Internet service: ten times faster for downloading information, and twenty times faster for uploading it.
These figures are taken from an informative 2012 book, “The Fine Print: How Big Companies Use ‘Plain English’ to Rob You Blind,” by David Cay Johnston, a Pulitzer Prize-winning financial reporter. In response to Johnston and other critics, the cable and telecommunications industry commissioned its own research, which, predictably enough, made the U.S. performance look a bit better. But more recent independent reports, from the Organisation for Economic Co-operation and Development and the New America Foundation, have confirmed what anybody who has spent some time abroad already knows. “Americans in major cities such as New York, Los Angeles, and Washington, DC are paying higher prices for slower Internet service.”
This struck me as an extraordinary claim. Could it really be correct?
Certainly, the assertion of such dramatic US inferiority in broadband didn’t square with my own knowledge of the industry, after working in it for many years. And some anecdotal experiences of my own led me to wonder whether something had been lost in translation.
For example, at a meeting with an Asian cable operator many years ago, we compared our broadband service offerings. At the time, my employer was offering service with download speeds of 10 Mbps or 30 Mbps. We asked the Asian operator what speeds they offered. After what seemed like some puzzlement over understanding the question, presumably due to the language barrier, they told us 100 Mbps.
That sparked more questions about how they were delivering such a high-speed, since at the time, standard cable technology (DOCSIS) only supported speeds of about 38 Mbps.
It turned out that (perhaps actually as a result of significant competition), download speeds weren’t a selling point in the market. Customers lived in apartment buildings that were wired for ethernet. The Asian cable operator served those customers by placing a cable modem in the basement, feeding an ethernet switch that connected each apartment. The ethernet switch used the “Fast Ethernet” standard, which has a nominal speed of 100 Mbps, known as the PHY rate. Now, the PHY rate is significantly higher than the actual capacity of the ethernet link, but regardless, the cable modem was capable of just the normal ~38 Mbps, as expected.
So in reality, the service they were offering maxed out at 38 Mbps service—not much different from our own 30 Mbps service. Because we had asked a question that wasn’t terribly relevant to their business, we got an answer that didn’t make sense—that they used 100 Mbps ethernet. Something had gotten lost in translation—not just language translation, but market translation.
Could similar misunderstandings play a role in the extraordinary claims made above? They certainly could be an element of it. Regardless, the New America Foundation report referenced above does indeed make extraordinary claims along the lines described in The New Yorker blog post. Their methodology for arriving at those claims is not fully clear to me, but it appears to focus on the “best” service available anywhere within a given city, based solely on marketing claims.
I adhere to the philosophy that extraordinary claims require extraordinary proof. The New America Foundation report certainly seems quite far from extraordinary proof.
So I decided to see what publicly available data I could find to examine the issue, myself.
Global Broadband Performance Data
It turns out there is a very robust source of information available from Ookla—the independent company that operates Speedtest.net, supplied the FCC’s consumer broadband test, and which is cited by the FCC as a leading source of speed data. Ookla makes public data collected using what they summarize as 5 million tests daily, across 350 million users in the past year, using 2600 testing servers deployed worldwide. Ookla collects both real-world speed performance data, and through post-test surveys, data about the cost and marketed speeds of the broadband packages consumers use for the tests.
We should expect that the performance data collected by this methodology is a very reliable indicator of real-world performance. It’s harder to be certain about the data collected via self-reported surveys, but Ookla has been remarkably transparent with their data generally, making the full (anonymized) data set available. And as we’ll see below, Ookla has also been transparent in identifying cases where they lack sufficient survey data to report cost information. Taken together, I believe we can be confident that the data from Ookla is reliable (and certainly far more reliable than modest surveys of market offers).
So let’s take a look at what the Ookla data actually shows us about how the US compares to the rest of the world, by looking at how it stands in relation to other OECD countries. (I have adopted Peter Nowak’s effective presentation format, although I must note that in his analysis of Canada’s relative broadband performance, he reaches conclusions that simply aren’t supported by the data he presents.)
Looking at the actual download and upload speeds supplied by country, there can be no question that the US does not lead. It is very much in the middle of the pack.
On the other hand, US speeds (and costs, as we’ll see below) are well within the mainstream, comparable to the averages for other similar countries, and better than the overall OECD averages:
Furthermore, let’s note that the Ookla data directly contradicts the assertion that French broadband speeds are 10-20 times faster than US speeds. Rather, the Ookla data shows actual broadband performance in France to average just 27% faster for downloads, and 32% faster for upload—a far cry from the 10x and 20x advantages, claimed.
Regardless, let’s be very clear: the speeds being provided by US providers are well above the requirements for applications deployed today. The speeds reported above reflect combined performance across all speed tiers. Even so, they indicate speeds easily capable of carrying multiple high-definition video feeds, simultaneously. And even higher speeds—typically 50 Mbps or even 75 Mbps download—are widely available at consumer price points.
Suggesting that US speeds are inadequate is like suggesting that highways built for 150 mph are inadequate, and should instead be built for 500 mph, when cars normally drive at 65 mph, and hardly any are capable of exceeding 100 mph.
But of course, speed is only one half of the criticism. The other half of the criticism is that broadband service is dramatically more expensive in the US than in other countries. So let’s move on to look at what the Ookla data says about cost.
Global Broadband Cost Data
We saw above that the US appears to compare reasonably to averages for other similar countries in terms of cost, but let’s take a look at cost on a country-by-country basis.
Based on Ookla data, broadband costs in the US are at the high end of the range. (Note, Ookla does not report cost data for South Korea or Japan, indicating that they lack sufficient data to do so. So the remaining charts exclude those two countries.)
However, when these costs are normalized to each country’s per capita GDP—effectively normalizing the costs of broadband to user’s incomes—US broadband costs look very mainstream. Notably, the normalized US cost is almost identical to that for France!
Editorial note, 2/18: Cost data has been presented here both in US dollars and normalized to per capita GDP because Ookla supplies the data in those forms, and per capita GDP normalization seemed a reasonable way to account for both the cost of broadband as a percentage of consumer income, as well as higher costs for ISPs to serve on account of higher prices and labor costs. One reader subsequently complained that this normalization is inappropriate. We are noting this, but leaving the article intact, with both sets of data. We would also point out that none of the analysis or conclusions below fundamentally depend on GDP-adjusted figures, and the same basic conclusions apply, even without GDP-adjusted figures.
Ookla also provides data on costs per Mbps of download performance. However, it’s important to note that, whereas the package cost data is based on averages, the per Mbps cost data provided is for the median—or the point at which half of the users pay more, and half pay less.
When we look at absolute cost per Mbps, the US again falls in the middle of the pack.
When we look at this data normalized to per capita GDP, on the other hand, the US looks comparable to all but a few of the lowest cost countries.
An interesting note here is that the per Mbps cost median for France is nearly double that of the US! This divergence between the mean of French broadband package prices and the median of French per Mbps prices suggests that France does indeed have some users paying very low rates for fast broadband, but that the majority of French users are actually paying much more for broadband than we pay in the US, and at much slower speeds.
On balance, claims that the US market is far slower and more expensive than other comparable countries appear to be simply wrong. Accordingly, it’s hard to give credence to the associated claims that different regulatory policies adopted by other countries have proven more successful.
And you might think that the story ends there, but in fact, there’s more to consider. To this point, we really haven’t had information to compare how much it actually costs providers to serve their customers. After all, if there are significant cost differentials, we should expect that to have some influence on prices.
ISP Cost to Serve
Now in looking at consumer costs normalized to per capita GDP, the data we’ve seen so far has already been adjusted, in some form, for general cost of living differences that presumably impact provider costs. We also need to look, however, at whether there are differences in usage patterns across the different countries, because the cost to serve customers is driven less by the headline speeds delivered than they are by the customers’ actual usage—and in particular, usage during so-called peak, when overall usage is highest.
To understand why, consider a broadband system with a capacity of 100 Mbps, shared by many customers. If during peak, the average customer uses 1 Mbps, then the system can support 100 customers. If however, the average customer uses 2 Mbps during peak, then the system can support just 50 customers. Broadband is a capital-intensive business, where the cost to build broadband capacity is the dominant component of ISP costs. So the costs to serve each customer in these two different scenarios are very different. Where customers use much more bandwidth, the ISP cost to serve is much higher.
It turns out that there is useful, publicly available data to evaluate the question of how the cost to serve in the US compares to elsewhere. For example, Sandvine, a company that supplies widely-used network management systems to many worldwide ISPs, publishes a twice-yearly Global Internet Phenomena Report, summarizing data trends actually observed in their ISP customers’ networks.
According to Sandvine’s most recent report (2H13), North American broadband users average 37.9 GB of data downloaded, and 6.6 GB uploaded, each month. The corresponding average usage for European broadband users are 14.0 GB downloaded and 3.4 GB uploaded—much less than half that of US users. Download utilization (the primary cost driver in the US) in Europe is just 37% of that in the US. Or conversely, US utilization is nearly three times European utilization!
The basic drivers for this differential are pretty clear: internet video is utilized far more in the US than in Europe. According to Sandvine’s data, in the US, Netflix accounts for more than 31% of peak utilization, and Netflix and Youtube combined account for just over 50% of peak utilization. “Real-time entertainment” altogether (all sources of streaming audio and video) accounts for 67% of peak utilization.
In Europe, on the other hand, over-the-top video services are far more nascent, with Netflix accounting for a mere 3.5% of peak utilization, and “Real-time entertainment” (primarily Youtube and BitTorrent) accounting for less than 50% of peak utilization.
Asia-Pacific users skew more toward peer-to-peer file-sharing/peer-casting services, leading to a higher proportion of upload usage compared to US users. Asia-Pacific users still use less data overall, and more than 35% less in downstream utilization, with average utilizations of 24.0 GB downloaded and 11.4 GB uploaded. (For the remainder of this discussion, we’ll focus on Europe, since Asia-Pacific is an aggregation of a couple of very mature broadband markets, in Japan and South Korea, together with many very nascent broadband markets. The aggregation of markets that behave very differently make it difficult to draw useful conclusions.)
Another useful source of data is Cisco’s Visual Networking Index forecast. Cisco is, of course, the leading worldwide networking vendor, with great visibility into network utilization trends, worldwide. Their most recent report and forecast for 2012-2017 (released May 2013) including data for fixed networks reported that peak period internet traffic increased 41% in 2012. (That’s a worldwide average figure. Growth in the US has been even higher.) It projects a 5-year annualized growth rate of 26% for the North American consumer internet traffic vs. 17% for Western Europe and 22% for Eastern Europe. Note that these figures include all traffic—peak and non-peak, fixed and mobile. (Frankly, if US growth in peak demand does come in at 26%, that would be a much, much lower growth rate than US ISPs have seen over the past 5 years.)
The bottom line is that the cost to serve broadband customers in the US has been, and presently is, much higher than the cost to serve customers in the rest of the world, and that is especially true when compared to European countries. Furthermore, we can expect that US customers will continue to be more expensive to serve—requiring more ongoing capital investment—particularly compared to European countries often cited by critics for comparison.
In light of that, the cost for broadband in the US looks very competitive, indeed. If, for example, we estimated that the nearly 3x utilization differential between the US and Europe results in a 2x differential in cost to serve, then among OECD countries, only Luxembourg would beat the US in normalized cost for broadband (and then just barely). And just Luxembourg and Denmark would beat the US in normalized cost per Mbps.
This is certainly not the dramatically inferior cost performance that critics of US broadband have claimed.
That’s not to say that the US broadband market is optimal. It surely isn’t. Even as a long-time cable guy, I wish that I had more truly competitive alternatives for broadband where I live. Competition has a remarkable way of driving even good performance and costs to be better.
On the other hand, it surely is to say that claims that the relative lack of broadband competition in the US have led to remarkably inferior performance and cost, compared to many other countries, are simply not supported by the facts. And it is noteworthy that US broadband performance clearly is not inhibiting the growth of broadband applications, given that the US leads in utilization across the board, as well as in utilization of the current killer broadband application: video.
Profitability
Some critics will surely be quick to dismiss the real-world data reviewed above, and instead point to high estimated gross margins as evidence of consumer gouging, as numerous critics have done in the past.
Of course, these claims don’t hold up any better to objective scrutiny. As we explained above, broadband is a capital-intensive business, where the ongoing capital required to build more and more capacity dominates the cost structure. But those costs are capital costs, not operating costs, and along with the costs to service the large amounts of debt ISPs take on to fund those capital investments, are completely excluded from the gross operating margin estimates that have been used to claim ISPs are overcharging exorbitantly.
If critics were correct that these estimated gross margins prove broadband costs are unreasonable, then we ought to see that ISPs are earning excess profits compared to other industries.
But do they?
Well, 2007 data compiled by CNNMoney (old, but the only usable industry ranking data I could find easily) ranked Telecommunications 43rd out of 52 industries, based on return on assets. Return on assets measures the amount of profit generated for each dollar of assets the business uses to generate those profits.
Since this is old data, and Telecommunication is a broad category, I also looked to individual company data for the 2013 Fortune 500. Here’s what I found:
Now, none of these businesses are pureplay ISPs, for sure. And we know that at least some of their other lines of businesses (the cable television operator business, in particular) are seriously challenged. But by any measure, these are not excess profits. In fact, they represent a much smaller return than you or I expect when we invest our own money.
Competing telco ISPs, don’t seem to be doing any better:
The telcos aren’t pureplay ISPs either, of course. But interestingly, their other, and dominating, line of business is mobile, which is quite robust. Yet they still show low returns on assets, in comparison to other businesses. I would expect this is a reflection of how capital-intensive and debt-intensive the mobile business is, with both network enhancements and device subsidies commanding massive ongoing capital investments.
The story looks only slightly better if we look at return on revenues (profit as a percentage of revenues), where the 2007 CNNMoney data ranks Telecommunications 27th out of 52 industries, and the individual company data for 2013 is:
With returns on revenue for Fortune 500 companies ranging from a high of 79.1% to a low of -59.6%, and with a median of 5.5%, it’s quite clear that US ISPs aren’t earning excess profits compared to other industries.
Accordingly, the bottom line is that, as frustrated as any or all of us may be that we lack more options for broadband service, claims of gouging by US broadband ISPs simply aren’t supported by the facts.