BroadbandClusters.org  ·  Research

Broadband adoption & digital equity research

Original analysis using FCC Broadband Data + Census ACS 2020–2024 across 500+ metros and all 50 states. ZIP-level data on internet adoption, device access, and the digital divide — freely available at broadbandclusters.org.

Track 1 — National snapshots
Track 2 — State deep dives

Interactive tool

State broadband explorer — 14 states

ZIP-level clusters, demographic filters, and access vs availability gaps across 14 states: Alaska, Arizona, Montana, North Dakota, New Mexico, Oklahoma, South Dakota, Wyoming, Minnesota, Alabama, Arkansas, California, Colorado, and New York. Filter by Indigenous, Hispanic, senior, veteran, disability, or any demographic group.

14 statesZIP-levelDemographic filtersAccess vs availability

Hari Narayanan  ·  ACS 2020–2024 + FCC Broadband Data

Data story series — Indigenous communities & broadband access

Arizona ★ Flagship Left behind: broadband access across Arizona's tribal nations 51% vs 76% statewide — 25-point gap across Navajo Nation, Tohono O'odham, and more Read story → New Mexico Tribal nations & broadband access in New Mexico Navajo Nation overlap with AZ — ZIP-level gap analysis across all counties Read story → New York Haudenosaunee & urban Native communities in New York 1,822 ZIPs analyzed — Haudenosaunee counties upstate + Buffalo urban Native communities Read story →

More state stories

South Dakota
Reservation communities & broadband
Indigenous · Pine Ridge
Montana
Tribal communities & broadband
Indigenous · Crow, Blackfeet
Alaska
Remote & Native communities
Indigenous · remote access

More state stories in progress — North Dakota, Oklahoma, and Minnesota coming soon.

Track 3 — Methodology & models

Working paper  ·  PDF

Decoding the digital divide: using autoencoders to uncover latent factors driving broadband adoption

A hybrid autoencoder trained on 15 demographic features identifies two latent dimensions explaining 53% of variance in broadband adoption rates across 22,676 U.S. ZIP codes — without using infrastructure data. Two pathways: rural composite disadvantage and direct device poverty.

Machine learningLatent factorsPolicy22,676 ZIP codes

Hari Narayanan  ·  December 2025  ·  ACS 2019–2023