Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods

Household-weighted shares from the American Community Survey, 2020 to 2024, across 82,000 census tracts. Left: neighborhoods in ten equal groups by median income. Right: six density bands.

Cite these charts: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

Before any modeling, the raw count says most of what follows. In the poorest tenth of neighborhoods, 59 percent of households have a wireline subscription; in the richest tenth, 89 percent. Another 11.6 percent of households nationally rely on a cellular plan alone, and 6.6 percent have no internet access at home. The density picture is flatter than most operators assume: the suburbs lead, the countryside trails by a wide margin, and the densest cities sit a few points below the suburbs, not above them.

The spread between neighborhoods is about thirty points. Everything that follows is an attempt to say how much of that spread is the households, how much is the network, and how much is the brand.

Cite this chart: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

The model behind the rest of this article takes 21 household characteristics from the Census, the neighborhood’s density, and how much of it each network passes, and predicts the subscription share. To test it honestly, it is fitted five times, each time with a fifth of the states held out entirely, and judged only on the states it did not train on. Each dot above is a neighborhood predicted that way.

On states excluded from training, the model explains 0.68 of the variation in neighborhood subscription rates. Survey sampling error limits precision. Our estimated noise ceiling of 0.66 is approximate; the model’s slightly higher score calls for checking that estimate, rather than concluding that all remaining variation is survey noise.

Household characteristics and network availability provide a useful benchmark for neighborhood subscription rates. The next sections examine which factors contribute to those predictions; they do not establish causal effects.

Cite this chart: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

A neighborhood with a median income near $34,942 subscribes 7.4 points below what the rest of its profile predicts. One near $166,920 subscribes 4.2 points above. Education runs a parallel course of similar size. Between them, income and education do more work than every network variable combined.

Demographics and density alone explain 0.63 of the neighborhood-to-neighborhood variation in subscription on neighborhoods the model did not train on, compared with an approximate noise-ceiling estimate of 0.66 derived from Census sampling errors. The number of competitors at the address adds nothing once the technologies present are known, the same pattern that holds for availability in the metro competition ranking.

Set penetration targets against what each neighborhood's households predict, not against a national average. The gap between the two is where acquisition budget is misallocated, and it is measurable before the first truck rolls.

Cite this chart: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

Cable is the network that decides whether a neighborhood subscribes at all. Where it passes almost nobody, subscription runs nearly nine points below what the households would otherwise predict; once it passes about nine in ten addresses, the effect stops growing. Fiber adds to that, but gently, and most of its contribution arrives only where it passes nearly every address. Fixed wireless, measured before the carriers’ 2023 to 2025 scale-up, does not raise wireline subscription anywhere and edges below zero where it is universal; where fixed wireless went, and whose footprints it presses, is its own study.

Adding the four networks to the model lifts its out-of-sample fit from 0.63 to 0.68. That is the tenth the network adds. The count of competitors at the address adds nothing further.

For a cable operator, coverage of the last tenth of a neighborhood is worth more than any other network investment in this data. For a fiber builder, presence lifts total subscription only at saturation; below that, fiber’s customers come from the incumbent, not from new households.

Cite these figures: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

Cite this chart: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

The gain was a catch-up. The lowest-income fifth of neighborhoods added 10.9 points; the highest-income fifth added 3.0. Neighborhoods that were already above 85 percent did not move. And the catch-up is explained by where a neighborhood started and who lives there, not by what was built: sorting neighborhoods by how much fiber or fixed wireless reaches them today changes the excess gain by less than a point in either direction.

One caution belongs here. The Affordable Connectivity Program subsidized broadband for low-income households from 2021 until it ran out of funds in mid-2024, inside the later window. The Census data cannot yet say how much of the low-end gain survives its end; the next five-year release, due December 2026, will.

In affluent metropolitan footprints the catch-up is over and growth is share, one customer taken for one customer lost; plan retention and acquisition spend on that basis. Whatever organic headroom remains sits in lower-income and rural neighborhoods, and the end of the Affordable Connectivity Program puts it at risk.

Cite this chart: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

Of the 32 million households Spectrum passes, 78.9 percent subscribe to wireline broadband, 1.0 points above what those households predict. Of the 42 million Xfinity passes, 79.5 percent subscribe, exactly at expectation. Mediacom sits 3.4 points below and Sparklight 2.6 below, after the lower incomes of their markets are already accounted for. The brands above two points are regional cable and telephone companies that upgraded their own plant to fiber.

The test of whether this is noise, or a leftover of something about the markets rather than the brands, is a completely different measurement: how much each footprint gained over five years relative to what its composition and starting point predict. The two readings share no data except the footprint itself, and they rank the brands the same way, with a rank correlation of 0.88. The five-year figure for every brand is in the table below.

One thing that no longer separates these brands is the speed tier on the rate card. Every major wireline brand offers gigabit on 93 percent or more of the addresses it passes. What differs is who buys speed: in a county model of take-up at 100 Mbps or faster, cable and fiber reach outrank income, and the median US fixed connection now runs 500 Mbps down and 35 up. A high-income footprint that under-buys speed has a packaging problem, not an availability problem.

Two readings this chart does not support. It is not brand equity: price, bundling, product and competitors’ response are all inside a brand’s number. And it is not market share: the Census counts a household subscribing to wire, not the company it chose, so where a second wireline network overlaps a footprint, a brand that wins share without growing the base registers no lift. That is also why fiber entrants that built most of their footprint after 2022 sit at or below zero here; the households were counted before the network reached them.

Cite this table: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

These gaps identify footprints for further investigation. They may reflect pricing, execution, competitors, deployment timing or local conditions the model does not capture. They do not measure an operator’s causal contribution or its own subscriber penetration. For scale, one percentage point of household subscription across Spectrum’s footprint represents about 322,000 households, regardless of which provider they choose.

Cite this table: G2M Insights, "Who Buys Broadband: Household Take-Up Across 82,000 US Neighborhoods," G2M Communications Research, edition 2025-12-31-v7, g2m.ai/research/communications/who-buys-broadband-household-take-up-brand-footprint.

Ziply passes the richest footprint in the table and Vyve and Sparklight the poorest, and the expected column has already adjusted for that: it is what those exact households would be predicted to do with the same networks present. Even so, the brands that beat their expectation tend to serve higher-income markets, and the correlation between footprint income and lift is 0.28 across all 65 footprints. The model is not leaking income; at the neighborhood level its errors are unrelated to income. The pattern is between brands, and the likeliest reading is selection: the operators that chose to upgrade affluent markets to fiber are also the operators that execute well. Public data can describe that. It cannot separate it.

If your footprint sits below expectation, the measured demographics and networks do not fully explain the gap. That is a reason to investigate local conditions and commercial performance. The networks present in any US market can be checked in the broadband market explorer.

• The subscriber ceiling is set by who lives there. Set penetration targets against what the households predict for each neighborhood, not against a national average.
• Cable reach decides whether a neighborhood buys at all; fiber reach adds only near saturation. For a builder, that means customers come from the incumbent, not from new households.
• The catch-up is over. In affluent metropolitan footprints growth is share, one for one, and the retention-versus-acquisition math changes with it.
• Every brand offers gigabit almost everywhere. If a high-income footprint under-buys speed, that is a packaging problem, not an availability problem.
• A footprint below expectation is not explained by its households or its networks. Look at local execution, price and competitors.

Every effect here is a conditional association on cross-sectional data, not a cause, and brand lift is not brand equity or market share.