The Economics of the Google Gigabit

In the excitement around Google's unveiling of the $70 gigabit broadband connection in Kansas City, some may be wondering how it is that Google can offer a gigabit for moderately more than what most of us pay for far slower cable broadband connections. On one side of the equation is the fact that big cable companies (Time Warner Cable, Comcast, etc.) have long been ripping off consumers by pricing their services far above cost -- something they can easily do because they face so little competition. But the more interesting side of the equation is how Google can make its gigabit price so low. Recall that Chattanooga made major waves with its gigabit service, priced then at the rock-bottom rate of $350/month. A gigabit is not available in many communities and where it is available, the price is often over $10,000 per month. We published an in-depth case study of their approach a few months ago. But, as Milo Medin -- the head of the Google Fiber project -- is fond of saying, "No one moves bits cheaper than Google." Google has built an incredible worldwide fiber optic network. Let's call this lessons 1 and 2.

Lesson 1: Google built its own network. It isn't leasing connections or services from big telecommunications companies. Building your own network gives you more control -- both of technology and pricing. Lesson 2: Google uses fiber-optics. These connections are reliable and have the highest capacity of any communications medium. The homes in Kansas City are connected via fiber whereas Time Warner Cable, CenturyLink, and others continue to rely on last-generation technologies because they are delaying investment in modern technology to boost their profits.

Others have already followed these lessons but are not able to offer their gig for such a low prices. To understand why, let's start with some basics. I'm hypothetically starting Anytown Fiber Net in my neighborhood and I want to offer a gig. Whenever any of my Anytown subscribers want to transfer files amongst themselves, the operating cost to me approaches zero because (aside from the capital costs of building the network), the cost of transfering those files is basically the electricity it takes to pulse lasers over the fiber and keep the fans on the routers humming.

Lesson 3: Local traffic on the network is essentially free. A local gigabit is no big deal on a fiber network. (Hat tip to Lafayette Utilities System for being the first to offer local 100Mbps traffic for free.)

We start talking about real operating costs when Anytown users want to connect to networks that are not on the Anytown network. When a user wants to watch a video on YouTube or download a patch from Microsoft, I need to interconnect with other networks that can get me there. For a small player like me, that means paying for transit. I pay Level 3 or some other major national network operator so my user can send a request to YouTube over the Level 3 network. The arrangement between me and Level 3 is interesting. I don't pay per bit that my customers use. Instead, I "commit" to a specific capacity. The higher the capacity, the lower my per bit charge. So if I commit to 500Mbps, I may be paying $7 for each Mbps but if I commit to 2000Mbps, I may pay $5 for each Mbps (these numbers are totally invented, not unlike how actual contracts seem to be made). But the interesting part is how it is measured and its implications. My committed rate determines my cost but not necessarily what I have access to. Let's say I commit to a 500Mbps connection to Level 3 for $7/Mbps. I have to pay for the amount of Mbps that corresponds to 95% of my peak demand. The cost comes down to how high the peaks are, not how many bits are transferred over the course of the month. So if my peak was 550 Mbps, then I have to pay for (550 * .95) * $7, or $3,657.50 On the other hand, if I allowed the combined usage of my users to hit a far higher peak, say 1,000 Mbps, my cost would be $6,650 and I would be kicking myself for not upping my committed rate. Fortunately, I can control the peak with my routers, allowing me some control (at the cost of alienating my users who will see worse performance individually). Another thing I can do is "peer" with others. For instance, if Anytown Net can get Google to connect directly to us, traffic to Google sites (ahem, YouTube) becomes free (as Google likely wouldn't charge because it wants to encourage everyone to use its services). This is why the Open Connect Netflix announcement is so important. Allowing users to hit popular sites without increasing the peak bandwidth saves real money. In fact, a sizable community fiber network reported to me that peering with a major source of video traffic dropped their monthly costs by tens of thousands of dollars. This brings us to Lesson 4.

Lesson 4: Scale matters. Big time. Everyone wants to interconnect with large networks and large sources of content. The larger Anytown Net is, the more others will be interested in connecting with me.

Google probably has the most favorable peering agreements with others because they all want cheap access to YouTube and the various other Google services. And Google can peer with others anywhere - they probably have a presence at every major interconnection location (to learn more about those fascinating places, read Tubes by Andrew Blum -- buy it through your local bookstore, not Amazon). What all of this means is that Google doesn't really have to worry about the cost of its peak because it already has advantageous relationships with the networks hosting the traffic that Google doesn't already have local to it. Let's go back to Anytown Net. If I offered a gigabit to my users, I would be exposing myself to a major peak in the evenings as most used that connection concurrently. As we now know, the cost to Anytown Net has much less to do with how much is transferred than to a few times when a lot of people happen to all be using a lot of their capacity at the same time. At present, there are a few other entities that have the kind of scale and relationships that could also do what Google is doing. They have names like Comcast, Time Warner Cable, Verizon, and AT&T. But they have little reason to invest because most of us are locked in to them. My neighborhood has one high speed Internet option - Comcast. We have a cheaper, slower DSL alternative from CenturyLink. This is why communities are increasingly building their own networks and policymakers need to pay attention the Looming Monopoly. These are the economics of Google's gig and an explanation of why it will be so hard to duplicate -- with one proviso. Dane Jasper of the incredible Sonic.net actually beat Google to the $70/gig. And I have no idea how he does it. If U.S. telecommunications policy were not so tilted in favor of the biggest corporations, perhaps we would see more successful local ISPs that subscribers actually liked. And let's not forget, though Google is offering a clearly superior product, Time Warner Cable is cutting its prices and locking customers into long-term contracts. It remains to be seen if this project will even be profitable for Google, though it is clearly already creating benefits for Kansas City as a whole. Though Google has a stated intention of demonstrating how this can be done so others can do it also, the lesson may be there very few can duplicate the full gig availability. But they can do what Chattanooga, Lafayette, and many others have done - build some of the best networks in the nation that are still accountable to the community. To learn more about the economics of networks, I strongly recommend this roundtable discussion: Photo of fiber deployment courtesy of Chattanooga EPB