|
VeriSign makes a great deal of money from the .COM and .NET registries. Can we tell how much they make, and how much that might change if the CFIT lawsuit succeeds? It’s not hard to make some estimates from public information.
The largest gTLD registry that VeriSign doesn’t run is .ORG, which was transferred a few years ago to the Public Internet Registry (PIR) which pays Afilias to run the registry, and uses whatever is left over to support the Internet Society (ISOC).
PIR is a 501(c)(3) charity, so their annual financial filings are public information. The most recent one I can find is from 2007, which says that PIR’s total revenue was $33M, they spent about $23M on running the registry and other expenses, and gave $9.7M to ISOC, leaving a reasonable end of year surplus of $500K. A quick look through the IRS Form 990 didn’t show any notably large or odd expenses, so I it’s close enough to say that what ISOC gets is what would be PIR’s profit.
ISOC is budgeting $15M from PIR this year. The prices for the domains are pretty close: .COM is $6.86, .NET is $4.98, and .ORG is $6.75. VeriSign has about 12.15 times as many domains in .COM and .NET as PIR does in .ORG, so do the arithmetic and the estimated profit for VeriSign comes out to $180M, not a small number.
This assumes that the cost per domain of running .COM is about the same as running .ORG, which seems plausible. On the other hand, the cost of running a database increases with size faster than linearly, and .COM is has some unique costs due to it being the main target for drop catching, tasting, and the like. I would trust the order of magnitude of this estimate, but not much more than that.
There’s a more direct way to estimate VeriSign’s profit. VRSN is a public company, so their finances are also public. For 1q09, their revenue was about $255M and their pretax profit comparable to the tax-exempt PIR was about $70M, a robust 27% profit margin. Annualize that and you get $1000M revenue and $280M profit, in the same ballpark as our other estimate. (VRSN has other businesses but they’re all trifles compared to the registry business.)
To estimate what they would make if the registry contract had been priced competitively, let’s wave our hands, and pretend that their revenue was the $5 they get for .NET, rather than $6.86, with the same costs.
This drops the revenue to $740M, leaving about $150M profit, a still respectable 17%. So we can estimate that VRSN’s profit is about double what it would be if there’d been a competitive bid to run the registry like there (sort of) was for .NET.
Thanks to Gordon Cook of the Cook Report for suggesting the comparison to .ORG’s surplus.
Sponsored byVerisign
Sponsored byDNIB.com
Sponsored byRadix
Sponsored byIPv4.Global
Sponsored byWhoisXML API
Sponsored byCSC
Sponsored byVerisign
One point to tease out is the assumption that running a database with a linear increase in size is results in an exponential cost increase. Why do you think that is true? Running a global DNS network is going to look more like a step function with capacity increasing by magnitude and scales pretty well. When the re-bid process occurs, I think that this point is left out and skews the competitive evaluation and is at least a reason for the differences in profit margin.
There are costs that expand exponentially such as network effect (more popular the TLD, the more popular an individual name is) and complexity with data management (harder to estimate, looks like a step function or evaluated as a fixed cost). Then you have linear effects like TLD size and hardware costs (routing/switching and server hardware). Then you have logarithmic costs like bandwidth and site size. Many of these costs are fixed though like software and physical sites (although they might look more like a step function depending on how you evaluate it). In the end, less than linear rates outweigh the exponential costs.
Another place to look at is some type of capacity analysis to derive a more accurate cost/name. I would assume (or at least I would highly consider it if I were in the position), that Verisign is more over provisioned than other networks which is helping to make the scaling effects look more palatable for comparisons like this.
Perhaps pricing should be included in any TLD rebid process. Otherwise, I doubt that you will see any gTLD rebid process go to someone other than an incumbent.
Keeping in mind that we’re all just waving our hands here, I agree that the read-only part of running a TLD, serving the DNS zone, probably scales no worse than linearly. On the other hand, the read-write part, the provisioning system where registrars, add, delete, and change, has to scale worse than linearly because of the locking required to keep a single master image of the database consistent. (This is why airline reservation systems use farms of cheap PCs to search for available flights, but tightly clustered mainframes to maintain the database.) VRSN has a particular challenge due to all of the drop catchers hammering on them 24/7 trying to be the first to grab a domain in the millisecond after it’s freed.
I do agree that VRSN is, if not overprovisioned, very heavily provisioned. During the arguments about tasting, the .ORG registry said they had to add extra capacity to handle tasting and wanted ICANN to adjust the rules to make tasting go away. We didn’t hear a peep from VRSN, presumably because they were already provisioned heavily enough to handle it. This also gives us a hint about the actual cost of running the registry: we can conclude from VRSN’s lack of complaints that they made enough from the tiny fraction of tasted domains that weren’t cancelled to cover the incremental costs of the ones that were.