Evaluating studies of the cost to serve all Americans with broadband

A few weeks ago, I updated my analysis of how far the BEAD funding will go, saying under ideal circumstances $42 billion could bring fiber-to-the-home broadband to 86% of the un- and underserved. More generally, the point was given RDOF funding, private capital matches, and assuming grant processes can engender some competition between ISPs that keep costs in check, the BEAD funding can go a long way to closing the digital divide. Coincidentally, at the same time, a study by VPS was put in a FCC docket on the same subject. Their estimate is it will take between $397 billion and $478 billion to reach all the underserved. 😮

It’s worth remembering there are only two numbers at play: the number of locations that don’t have access to 100/20 broadband service, and the average cost to bring fiber-to-the-home service to those locations. Let’s start with the number of underserved locations. I estimated 23.1 million un- and underserved locations. The VPS study estimated 45.5 million, or 32% of all United States housing units are underserved.

According to the FCC’s visualization of Form 477 data, only 8.2% of Americans lack access to either cable or fiber service at 100/10. Pretending DSL and fixed wireless don’t even exist, that’s approximately 11.6 million housing units that don’t have access to cable or fiber. Of course, it’s Form 477 data and it underreports.

Looking at it another way, how much of America is underserved if 45.5 million housing units are truly underserved? Here’s a map of the least dense census blocks in the country that represent 45.5 million housing units. If the VPS number is right, all the areas in green wouldn’t have access to 100/20 broadband. It’s a lot.

Zoomed in on Kansas City:

St. Louis:

and Philadelphia:

Here’s another data point: as both VPS and I referenced, Jim Stegeman, the CEO of CostQuest, said in an interview he thinks 23-25 million housing units are underserved. Given CostQuest’s contract with the FCC for new broadband maps, they have access to the best available data. I wouldn’t bet against the house.

To come up with their 45.5 million underserved number, VPS first multiplied 5.7 million unserved housing units (as of June 2020 Form 477 data) by 3.2, which is the factor by which unserved locations increased in Georgia’s broadband maps. They took the resulting 18 million housing units and multiplied it again by 2.5, which is the ratio between unserved and underserved housing units, resulting in an estimate of 45.5 million.

The latest June 2021 Form 477 data (VPS acknowledges there has been additional deployment, but doesn’t alter their numbers to reflect it) have 3.6 million unserved housing units. An updated estimate using their same methodology would have 28.8 million underserved housing units.

I believe a stronger methodology would have been to estimate the number of un- and underserved housing units and then apportion them between the two. The FCC’s new maps will light up partially served Census blocks we currently think of as served, but each location with light up as either unservered or underserved, but not both. Large amounts of urban America would need to be underserved for the underserved number to come anywhere close to 45.5 million — there just aren’t enough rural housing units.

Cost to serve

The second part of the equation is the cost to serve the average unserved or underserved location. The existing studies VPS references are below, and range from $1,818 per location to $5,714 per location. My model used an average cost of about $6,500 per location.

VPS discards these numbers in favor of their own experience in building networks. For unserved they use $15,000 per location, which includes electronics and drops to the house for those who subscribe. For underserved, they use $6,000 to $9,000 as the range. The total is between $397 and $478 billion.

The total capital required in my model was about $151 billion. The reason I believe $42.5 billion in BEAD funding can make a significant dent when $151 billion might be required is RDOF has already committed funds for more than 30% of the locations. These locations cannot be funded by BEAD projects. Also, in the RDOF auction ISPs provided more than 90% of the capital in 20% of cases. If ISPs bring 50% of the capital to BEAD projects, that’s 50% less that needs to come from the BEAD program.

Another point that’s important to remember is the FCC’s new maps will show census blocks that are partially served — there is an ISP providing service on the block, we just don’t know what percentage of the block they cover. The locations are almost certain to be less expensive to reach than the blocks we currently know to be unserved.

I believe the VPS estimate of the number of underserved is off by a factor of 2. I also believe the cost to serve each location is high, though given supply chains, inflation, labor costs, and other externalities there’s significant uncertainty at the cost to serve a location with broadband. For this reason, it remains critical that NTIA equip states (and the public) with a cost model, and encourage state rules that bring competition to the grant process.

Once RDOF and private capital matches are considered, it’s likely there’s enough money in the BEAD program to reach all the unserved and make a significant dent in the underserved, particularly in states with lower cost of deployment.