Issue 084 - Cellular infrastructure - Coverage geometry
How many cell sites does it take to cover the United States?
SpaceX selected Qualcomm equipment for a terrestrial Starlink Mobile 5G network, with deployment planned to begin in 2027.
Energy and infrastructureAbout 1 minute
Sources checked November 5, 2026. Figures and circumstances may have changed.
The problem
Starting with land area and an estimated radio coverage radius, about how many macro cell sites would provide broad geographic coverage across the contiguous United States?
Allow for overlap rather than dividing the land into perfectly nonoverlapping circles. The guess unit is sites for one network, not antennas, sectors, satellites or all operators combined. Use the lower 48 states as a clearly defined first approximation; Alaska and Hawaii are an extension.
After estimating, compare the result with the announced build-out. Optionally increase site density for urban capacity, or choose a power draw per site and estimate the network's electricity demand.
Fermi estimates target scale rather than precision. The Fermi reference collects useful facts and equations.
Grounding facts
Census area tables put contiguous-U.S. land near 7.7 million km2; all 50 states have roughly 9.1 million km2 of land. Matt's 15-million-km2 rectangle therefore overstates the lower-48 land area by nearly a factor of two.
A GSMA spectrum report gives an illustrative rural-network radius range of about 3.5 to 10 km. Ericsson has demonstrated rural 5G coverage beyond 10 km under specific test conditions. Neither establishes a universal site radius: frequency, antenna height, handset uplink, buildings, terrain and the desired data rate all matter.
Qualcomm's announcement specifies tens of thousands of sites, deployment from 2027 in multiple waves, and 64-transmit/64-receive Massive MIMO equipment. Reuters separately relayed Bloomberg reporting that SpaceX was in talks to secure access to at least 80,000 sites. Access discussions are not completed installations or a promise of gap-free land coverage.
WIA reports about 255,000 macrocell sites and 158,500 purpose-built towers across the U.S. industry at the end of 2025, plus small cells. Those are different categories across multiple networks; an operator can colocate equipment on an existing structure.
After checking sources
Checked answer and calculation
contiguous-U.S. land area
~= 8 x 10^6 km2
assumed useful rural-style radius
~= 8 km
circle area per site
= pi r^2
~= 3.14 x 8^2
~= 200 km2
effective coverage allowance
~= 80% of circle area
~= 160 km2/site
geographic site count
~= 8 million km2 / 160 km2/site
~= 50,000 sitesThe checked target is an illustrative geometric estimate, not an optimized engineering design. For gap-free ideal flat-land coverage, circles centered on a triangular grid can serve hexagonal cells with area about 2.6r^2, roughly 83% of pi r^2. The 80% allowance is a rounded version of that geometry; actual landscapes need additional margins and careful placement.
The radius is the dominant uncertainty because area scales as r^2. With the same 80% allowance, 5 km gives about 130,000 sites, 8 km about 50,000, and 13 km about 19,000. Covering all 50 states at the central assumptions would raise the geometric count to around 57,000, with Alaska's terrain making uniform-radius coverage especially unrealistic.
area per site at 80,000 sites
~= 8 million km2 / 80,000
~= 100 km2/site
implied ideal effective radius
~= sqrt(100 / (0.8 x pi))
~= 6 kmSo tens of thousands of macro sites make geometric sense for broad coverage. That does not prove continuous indoor service, adequate urban throughput, or purely terrestrial coverage everywhere. Satellite service and terrestrial service are complementary parts of the wider Starlink Mobile plan.
Using Matt's 13 km radius but correcting only the land area gives about 14,500 sites before overlap, or about 18,000 with the allowance. This explains the smaller figure in his reflection. Reducing the radius at the same time pushes the count upward again; 12,000 to 16,000 should not be treated as a more accurate nationwide requirement.
A broadcast-radio analogy is a weak range peg for cellular service: receiving a broadcast is different from sustaining a two-way handset link at a useful data rate.
capacity sensitivity example, not measured land-use shares
dense areas: 1% of 8 million km2 / 2 km2 per site
~= 40,000 sites
suburban areas: 4% / 20 km2 per site
~= 16,000 sites
rural areas: 95% / 160 km2 per site
~= 48,000 sites
illustrative total ~= 100,000 sitesThe invented density categories show how a small share of land can add many sites. They are not a forecast of SpaceX's design. Real capacity planning needs traffic, spectrum, simultaneous users and performance targets, not just area.
optional electricity scenario: assume 5 kW per site
50,000 sites x 5 kW
= 250 MW continuously
~= 2.2 TWh/year
at an assumed 1.2 kW average per household
~= 210,000 household-equivalentsThe 5 kW assumption represents an illustrative whole-site electrical load, not radio-frequency transmit power or a measured Qualcomm specification. Backhaul, cooling, equipment configuration and utilization change it.
Before checking sources
Matt's original estimate
This is Matt's original, unverified estimate, preserved before the source check.
I assumed that the land area of the continental US is about 1.5 * 10^13 m2 (basically an approximation based on an average east/west coastal distance of about 5,000 km and an average north-south coastal distance of about 3,000 km).
I also assumed the average cell tower might have a reliable range of about 8 mi or about 13 km (based on how well my radio functions based on known distances from the station).
using 13 km as a radius, a single tower's range of coverage would be pi * r^2 or about 3 * (13,000 m)^2 = about 5 * 10^8 m2.
dividing that coverage into the total land area of the US approximated above, that gives about 30,000 cell towers needed if they're all perfectly arranged for maximal coverage and minimal overlap, and if both my land area and per-tower coverage areas are correct.
Reasoning score
Matt's reasoning score: 55 / 100
Useful facts: 10/30. The land rectangle was nearly twice the actual contiguous land area. A 13 km reach is possible under some conditions but optimistic as a universal useful cellular radius; broadcast reception does not verify it.
Reasoning approach: 15/30. Dividing land by pi r^2 is a useful first estimate, but covering gaps requires overlap and a real network needs capacity and terrain allowances.
Math: 10/10. The circle area, square-meter units and roughly 30,000-site division followed correctly from the assumptions.
Final estimate: 20/30. About 30,000 is within a factor of two of the 50,000-site geometric scenario and within an order of magnitude of the reported 80,000-site access discussions.
Post-check reflection
Matt's reflection
I was in the right ball-park even though I overestimated land area in the US and overestimated the likely coverage area per tower.
If my assumptions were more accurate I might have finished with a calculation of about 12,000 - 16,000 towers, but the actual reported number of tower access requests by SpaceX is more like 80,000.
Recommended memory peg
Contiguous-U.S. land ~= 8 million km2. For ideal broad cellular coverage, use effective area ~= 2.5r^2 with radius in kilometers. Halving radius roughly quadruples site count, before adding urban capacity or terrain constraints.