IPv6 Adoption in the Top 100,000 Web Hosts

A measurement study of IPv6 adoption among the 100,000 most-linked hosts on the web, using the Common Crawl Web Graph.

Live report: commoncrawl.github.io/ipv6-analysis

Key findings

• 36.9% of the top 100,000 hosts are fully reachable over IPv6 (AAAA record + successful HTTPS probe).
• 1.2% publish AAAA records but fail reachability checks — these hosts actively hurt dual-stack users via Happy Eyeballs fallback delays.
• 61.9% have no AAAA record at all.
• Adoption correlates with rank: 71% of the top 100 hosts support IPv6, falling to 32% at ranks 50,001–100,000.

Methodology

Hosts are ranked by harmonic centrality from the Common Crawl Web Graph release cc-main-2025-26-dec-jan-feb, built from three crawls spanning December 2025 through February 2026. Each host was assessed in two stages:

1. DNS AAAA lookup — dig +short AAAA to check for published IPv6 records.
2. IPv6 reachability probe — curl -6 --head over HTTPS with a 10-second timeout, following up to three redirects. Any HTTP response (including 4xx/5xx) counts as reachable.

Probes were issued from a host with native IPv6 connectivity in northwestern Santa Clara County, California. DNS lookups ran at up to 100 concurrent tasks; curl probes at up to 50.

Files

• docs/index.html — The full interactive report with charts, tables, and methodology details. Served by GitHub Pages.
• docs/ipv6-results-100k.json — Raw per-host results (rank, hostname, AAAA records, reachability status, HTTP response code).
• v6_probe.py — The probe script used to collect the data.

Data format

Each entry in docs/ipv6-results-100k.json looks like:

{
  "rank": 1,
  "host": "www.facebook.com",
  "has_aaaa": true,
  "aaaa_records": ["2a03:2880:f36c:1:face:b00c:0:25de"],
  "ipv6_reachable": true,
  "http_code": 200
}

Author

Common Crawl Foundation

Measured 16 March 2026.

License

This project uses a dual-license model:

• Data (ipv6-results-100k.json) is released under CC0 1.0 Universal (Public Domain Dedication).
• Code (index.html and everything else) is released under the Apache License 2.0.