Home » Internet

The problem with cost per Mbps comparisons

By 13 October 2009 5 Comments

internet-mapWhenever people quote broadband prices from abroad, they often cite some ridiculously low figure that are either flat out wrong or misleading.  One frequent assertion is that a person could get 100 Megabit Per Second (Mbps) broadband service in Japan for $14.  But that only covers the “provider” charge but doesn’t cover the actual facilities charge of $50 that has to be paid to Nippon Telegraph & Telephone (NTT).  Other examples of misleading data is the the OECD broadband rankings which among other issues don’t factor in crucial details like usage caps when comparing cost per Mbps.  Since US broadband services typically have 10 times more usage allowance than nearly all other OEDC nations, the OECD cost per Mbps ranking is extremely misleading.

Other assertions we hear about is that broadband costs only $0.27 per Mbps in Japan but costs $3.33 per Mbps in the United States or that it’s $0.13 per Mbps in Japan and $2.83 per Mbps in the United States.  The problem is that these are comparing multi-dwelling advertised (theoretical) Mbps in Japan compared to single family home cost per delivered Mbps in the United States.  So while it’s possible that someone might be able to get 1000 Mbps access in an apartment complex somewhere in Japan for $130 a month which works out to $0.13 per advertised Mbps, real world usable speeds don’t come anywhere close to that performance.  We know that anything above 100 Mbps is suspect since major Japanese sites offer less than 4 Mbps content because server bandwidth is much more expensive in Japan than the United States.  Moreover, many computers only come with 100 Mbps network interfaces or far less with 802.11 Wi-Fi so there’s no way they could use 1000 Mbps even if the service could deliver those speeds.  These comparisons also ignore the fact that it is also possible to get very cheap 100 Mbps broadband access in the United States if a person doesn’t mind paying sky high rent for a tiny apartment in an urban area and these numbers weren’t factored in, and high rent and small living quarters aren’t exclusive to American cities.

Table 1 below shows what people will typically get in the U.S. and Japan and it shows that the differences are not that high when comparing apples to apples.  Furthermore, the cost per delivered Mbps is going to be even closer.

Table 1: Advertised speeds versus monthly cost

Internet Service Provider Down Mbps Up Mbps Price $/Mbps *
AT&T DSL 6 0.768 $35 5.83
Yahoo! BB DSL 8 0.9 $36.60 4.58
YouTV standard cable 10 1 $38.65 3.87
Comcast standard cable 15 3 $42.95 2.86
YouTV premium cable 160 10 $65.58 0.41
Comcast premium cable 50 10 $99.95 2.00
BB.excite (NTT Single Family FTTH) 100 100** $66.72 0.67
@nifty (NTT Single Family FTTH) 100 100** $74.95 0.75
Verizon FiOS (Single Family FTTH) 25 15 $64.99 2.60
Verizon FiOS (Single Family FTTH) 50 20 $139.95 2.80
@nifty (NTT Multi-unit FTTH) 100 100** $45.68 0.46
Multi-unit access in San Francisco 100 100 $30 0.30

* $/Mbps is computed using the advertised download speed, but the numbers are not directly comparable between different providers or countries because it does not factor in the differences in oversubscription and actual delivered speeds.  The number also doesn’t factor in the differences in upload speeds.

** NTT caps upstream at 2.78% duty cycle which means upstream can only average 2.78 Mbps.  Non-compliance could mean account termination.  This is very relevant to peer-to-peer (P2P) performance because upstream performance determines the downstream performance of the P2P network.

Suspect data from New America Foundation Report

The New America Foundation report on broadband prices between U.S. and Japan claims that @nifty is $44/month for 100 Mbps access, but they do not cite or link to any source for their data.  I’ve been monitoring Japanese prices since the beginning of this year and I’m fairly certain the prices have not changed that much since the beginning of the year.  My guess is that the New America Report may have failed to include the “provider” charge and only quoted the NTT facilities charge.  Table 1 of this article shows @nifty is actually $74.95 and I provide the link to the site that lists the prices.  This differential is not explained by the ~7% difference in exchange rates from June 2009 and October 13, 2009 because even a 10% differential in exchange rates would only explain a $4.4 differential.

Furthermore, I can’t find any reference to a $39 1000 Mbps plan and if such a service is offered, it is likely heavily oversubscribed where you have a single gigabit connection feeding hundreds of units each of whom have a gigabit Ethernet feed.  The New America report fails to mention that you can get very cheap 100 Mbps access for $30 in multi-dwelling units in American cities like San Francisco.  Objectively speaking, none of these multi-dwelling prices are comparable to single family broadband prices because a significant portion of the “last mile” infrastructure is borne by the high rent paid by tenants.  If we put this in the context of paying $2300/month for a small 900 square feet apartment, $30 for 100 Mbps Ethernet broadband doesn’t seem like such a great deal any more.

US broadband providers have the most honest advertised speeds

Based on an Ofcom study (see Figure 1), Japanese broadband services have much lower real-world average speeds than their advertised burst speeds compared to American broadband services.  So despite the fact that they have a theoretically lower cost more per advertised Mbps on the higher end service plans, their actual cost per delivered Mbps is much closer to American broadband plans.

We can look at economist Scott Wallsten’s 2008 paper: Understanding International Broadband Comparisons to see how misleading advertised speeds can be in Japan.  One of the charts Dr. Wallsten shows in his paper is a comparison of advertised performance versus actual performance and the effect this has on customer satisfaction.  The data is derived from Ofcom (2007) survey data.

Figure 1: Advertised Mbps versus deliver Mbps versus customer satisfaction

Wallsten 2008 – Understanding International Broadband Comparisons

Based on this actual data, we can see that US broadband providers deliver the closest actual throughput to what is being advertised, and it is well above 90% of the advertised rate.  When we consider the fact that advertised performance is often quoting raw signaling speeds (sync rate) which include a typical 10% protocol overhead and the measured speeds indicate payload throughput, US broadband providers are delivering almost exactly what they are advertising.  By comparison, consumers in Japan are the least satisfied with their broadband service despite the fact that they have the fastest broadband service.  This dissatisfaction is due to the fact that Japanese broadband customers get far less than what was advertised to them.  The Ofcom data is further backed up by Akamai data which provides the largest sample of real-world performance data in the world.  When real-world speeds are compared, the difference is nowhere near the difference in advertised speeds.

Cable broadband companies in Japan don’t really have more capacity than their counter parts in the United States yet they offer “160 Mbps” service.  While that sounds even more impressive than the Fiber to the Home (FTTH) offerings in Japan or the U.S., it’s important to understand that the total capacity of the network may only be 160 or 320 Mbps shared between an entire neighborhood with 150 subscribers.  Even if only 30 of those customers in a neighborhood takes up the 160 Mbps service, just 2 of those customers can hog up all the capacity on the network.

While Verizon may charge more per Mbps for FTTH service than their Japanese counter parts, they aren’t nearly as oversubscribed and they usually deliver advertised speeds or even slightly higher than advertised speeds.  More to the point, Verizon’s 50 Mbps service is a lot closer to the 100 Mbps services they offer in Japan than the advertised speeds suggest.  Oversubscription is a very important factor and it is the reason businesses will pay 20 times more per Mbps for dedicated DS3 circuits with no oversubscription on the access link.  If Verizon were willing to oversell and exaggerate advertised bandwidth as much as FTTH operators in Japan, they could easily provide “100” Mbps or even “1000” mbps FiOS service over their existing GPON infrastructure.  The risk in doing this is that Verizon might lose some customer satisfaction and trust when the actual performance (while higher than current levels) don’t live up to the advertised speeds.

No justification for overreaching conclusions

There are those who want to draw far reaching conclusions on the negative economic impact of broadband prices based on these misleading comparisons, but the reality is that broadband prices are only a small factor in Internet growth. The computer’s ability to use these high speeds especially on the upstream e.g., limited real-time HD video encoding performance or the ability of content companies to pay for hosting higher bandwidth content are much more significant factors.

It’s also important to note that costs generally stay the same as bandwidth grows exponentially since prices are set to what consumers are generally willing to pay.  If any thing, prices decline over time as uptake increases because of the increased economy of scale.  Cost per Mbps will always decline along an exponential curve just like cost per MB storage declines as storage devices grow in capacity.

As for super low prices like the ones found in multi dwelling units, factors like rent prices and quality of life exceeds any savings one may get.  And when it comes to business innovation, broadband costs aren’t even a factor because businesses generally don’t use broadband to host websites and applications.  They colocate in datacenters and server colocation costs are up to 10 times lower in the United States than Japan.  But even that didn’t play too much of a factor in American success because the major obstacles startups face are getting the right idea in the first place, paying the developers, and getting venture capital funding.  The $50 to $300 per month server colocation and bandwidth costs pale in comparison to the development and legal costs of a new startup.


  • Nick Brown said:

    Nice post George. It may also be noted that average wholesale backhaul prices for connection to an Internet backbone can run below $10/megabit in urban areas while rural area providers may be paying up to $300/megabit. It is possible that since the US has less densely packed urban environments in comparison to Japan, the average cost per megabit is raised in some organizations research tables. In future studies from organizations like the ones mentioned in your post, it may be more balanced to separate urban and rural rate structures.

  • George Ou (author) said:

    Rural areas are more like $100 per Mbps for backhaul but this is largely a function of economy of scale and the cost of the long-haul connection to somewhere remote. If you tried to buy dedicated backhaul bandwidth in urban centers at quantities of 1.554 Mbps, you’re still going to pay $200 per Mbps.

    If you’re in a major metropolitan area where the fiber has already been laid by someone like Verizon Business Network, I’ve just gotten a quotation for dedicated bandwidth at $28 per Mbps for a fully saturated 155 Mbps OC-3 circuit. But again, you’re not paying for a last mile(s) circuit because the fiber is already there. If you were 1 mile off the fiber “grid”, you’d have to pay another company for a lease line to connect to close the gap to get to the backhaul provider and that would cost you more money. If you’re 100 miles away, you can expect that lease line to cost a LOT more money because someone has to trench 100 miles of fiber for you.

    So the two factors that determine very large differences in backhaul cost is the distance of the lease line and the quantity that you purchase. If you’re a small rural operator and you only need 4 Mbps of backhaul and you’re very far away, it’s normal to be paying high rates. This is no different in any other country including Japan.

    The main point I tried to raise in this article is that you cannot compare multi-dwelling unit costs to single family home costs. The MDUs already laid the access portion of the network using Category 5 or 6 Ethernet cabling (which only works when your distances are less than 100 meters), but this cost is borne by the rent that tenants pay and it’s hidden from the cost of the Internet connection.

  • Bret Swanson - Maximum Entropy » Blog Archive » Preparing to Pounce: D.C. angles for another industry said:

    […] George Ou explains, the international data on broadband “speed” and prices are highly suspect. Benkler […]

  • Preparing to Pounce: D.C. angles for another industry — Technology Liberation Front said:

    […] George Ou explains, the international data on broadband “speed” and prices are highly suspect. Benkler […]