Unfortunately, there is little reliable information on actual costs with ISPs typically claiming that such information is a trade secret that should be kept confidential. My report noted that there are two elements to the actual costs. One is the Internet-facing data costs, which arise once a user’s traffic travels onto the public Internet. This cost is very low, estimated in the report at about one cent per GB and falling. This is consistent with public transit arrangement pricing and is likely even cheaper for large ISPs that use peering arrangements to cover off most actual costs.
The more difficult calculation involves the internal ISP network leading to the public Internet. As CRTC Commissioner Candice Molnar noted during the usage based billing hearing, “we all, I think, can hopefully agree that there is no marginal cost to using the network when you are not causing augmentation.” While there are no marginal costs, there is a capital cost of building the network and ongoing maintenance and augmentation costs when congestion arises due to traffic growth. My report used Bell’s data in the deferral account case (one of the only ones to put information on the public record) to estimate that seven cents per gigabyte (for a total of eight cents) was a best guess among a range of possibilities.
Yet Bell’s recent comments about actual costs suggests that it relies on similar ranges to provide a best guess at cost per gigabyte. Over the span of one week, it effectively provided three different numbers. On July 13th, Bibic told the Wire Report that 19.5 cents per gigabyte, the original number in its AVP proposal, was close to its actual costs. Bibic claimed “they bear a fairly close relationship to the nineteen-and-a-half cents. The nineteen-and-a-half cents [per gigabyte] is justifiable based on the actual costs of a gigabyte on the network.” A week later, Bibic told the CRTC that actual wholesale costs were $39.15 per Mbps, which works out to 12.4 cents per gigabyte at full capacity. A day or two later, it submitted that 17.5 cents per gigabyte reflected actual costs plus markup. The 17.5 cent number appears to be based on fairly low wholesale usage, however (roughly 63% usage). Increasing estimated usage would result in a significantly lower estimate on the cost per gigabyte, likely closer to 12.4 cent estimate.
Bell’s inconsistency aside, the company submitted a response in April on my report after Telus conveniently asked about it during the interrogatory phase. It claims that my estimate is based on assuming maximum usage every month (900 GB), which obviously results in a lower cost per gigabyte. It notes that average users do not come close to using that maximum capacity, particularly since it is a shared network. My report included a pricing scenario based on theoretical maximum output (F) and it sets the lowest possible price per GB for each equipment configuration [$0.30 for a pure OC3 configuration; $0.08 for a pure OC12 configuration, and $0.02 for a pure OC48 configuration). Bell seems to assume that this is the OC12 F scenario that is the source of the eight cents per gigabyte figure.
Yet that is not how the report arrived at the figure. It identifies a range of pricing options based on different equipment configurations and usage patterns which range from as high as from $0.34 to as low as 0.019/GB (see Table 5 at page 34). The report notes how the pricing data relied on for the study was created for rural areas and for a complete buildout of the network (as opposed to simple network maintenance through provisioning for increased flows on an existing network). This suggests that actual prices will be at the lower end of the spectrum presented in table 5 (the costs for an urban network are likely far lower than a network that Bell itself says would not be built based solely on market considerations).
The report never suggested that the cost should be based on the assumption of maximum usage. Instead, it concluded:
So, in the long run, the low end of the range would again seem to be the better estimate as demand for peak bandwidth increases. Consequently, while there is a range of plausible values, a comparison of Tables 5 and 7 suggests a figure of $0.07/GB may be the best single estimate for the costs of the internal network.
In other words, there are a number of possibilities, but the seven cents/gigabyte figure appears to be the best guess based on Bell’s own numbers. Bell’s claims now cover a range as well, reflecting the reality that per gigabyte pricing is subject to many variables that result in a spectrum of possibilities (the most important variable is capacity/usage since the higher the assumed capacity and usage, the lower the cost per gigabyte – fixed costs to build divided by larger or smaller amounts of data depending on use). Arriving at a common metric for identifying per gigabyte pricing will be a crucial aspect to resolving the wholesale usage based billing issue since the issue is subject to considerable variation. It is hoped Bell will step up with public data for all to scrutinize, or at the least that the Commission treat Bell’s proposed figures on cost with the skepticism it deserves.