IPv6 has been around for decades, yet adoption on the public web remains surprisingly patchy. To get a concrete snapshot, we tested the top 100,000 hosts from the Common Crawl Web Graph for IPv6 support using DNS lookups and live HTTPS probes. The results paint a clear picture of where the transition stands in early 2026.
What we measured
We took the 100,000 most-linked hosts from the latest Common Crawl Web Graph release (cc-main-2025-26-dec-jan-feb), ranked by harmonic centrality (a measure of how central a host is in the web's link structure). For each one we asked two questions: does it publish a AAAA DNS record (the IPv6 equivalent of an A record), and can we actually reach it over HTTPS on an IPv6 connection?
Probes ran from a host with native IPv6 in Santa Clara County, California. Any HTTP response, even a 404 or 500, counted as reachable. We weren't testing whether the site works well, just whether IPv6 packets can get there and back.
The headline numbers
36.9% of the top 100,000 hosts are fully reachable over IPv6. A further 1.2% publish AAAA records but failed our reachability check, likely due to misconfiguration or firewall rules. The remaining 61.9% have no AAAA record at all and are IPv4-only.
One encouraging finding: when operators do publish AAAA records, the deployment is almost always complete. 96.8% of hosts with a AAAA record responded successfully to our IPv6 probe. Broken or half-deployed IPv6 is the exception, not the rule.
Bigger sites do better, but not all of them
IPv6 adoption correlates strongly with rank. Among the top 100 hosts, 71% support IPv6. By the time you reach ranks 50,001–100,000, that falls to 32%. The pattern makes sense: larger operators tend to have more mature infrastructure, and major CDN providers like Cloudflare and Fastly have offered IPv6 by default for years.
But there are some notable holdouts even at the very top of the list. Twitter/X (ranks 7, 14, and 33) has no IPv6 at all. Neither does GitHub (rank 19), Reddit (rank 58), TikTok (rank 31), or OpenAI (rank 82). Even the Internet Archive's web.archive.org (rank 81) lacks IPv6 support. Meanwhile, Facebook, Google, YouTube, Instagram, LinkedIn, and Wikipedia all respond happily over IPv6.
The Cloudflare plateau
One interesting pattern shows up in the mid-ranks. Between roughly ranks 2,500 and 10,000, IPv6 adoption holds steady at 46–50% instead of declining smoothly. The likely explanation is Cloudflare, which enables IPv6 by default for all hosted domains. A concentration of Cloudflare-proxied sites in this band pushes the measured rate up regardless of whether the site operators themselves have done anything deliberate.
This is worth keeping in mind when interpreting the numbers. Our measurements reflect end-to-end reachability from the public internet, including IPv6 termination at a CDN edge. A site proxied through Cloudflare may appear IPv6-capable even if the origin server behind it has no IPv6 connectivity. From a user's perspective that site is reachable over IPv6 (which is what matters for real-world performance), but it does blur the line between native deployment and CDN-mediated IPv6.
Why this matters for performance
This isn't just about protocol completeness. Modern browsers implement the Happy Eyeballs algorithm (RFC 8305), which races IPv6 and IPv4 connection attempts simultaneously, with a slight head start for IPv6. When a site is fully reachable over IPv6, dual-stack users get the fastest available path automatically.
But when a site publishes a AAAA record and then fails to respond over IPv6, it's actively worse than having no AAAA record at all. The browser tries IPv6 first, waits roughly 250 milliseconds for it to fail, and then falls back to IPv4. That penalty hits on every new connection. The 1.2% of hosts in our survey with broken IPv6 are imposing this cost on dual-stack users with every visit.
Why aren't more sites enabling IPv6?
Common barriers include the cost of upgrading legacy network hardware and software, the complexity of running a dual-stack environment, and a lack of perceived urgency while IPv4 still works (often extended by carrier-grade NAT). Many operations teams simply don't have deep IPv6 expertise, and when everything works fine over IPv4, the migration stays at the bottom of the priority list.
Explore the data
The full interactive report, with charts, a searchable table of the top 1,000 hosts, and detailed methodology, is available at commoncrawl.github.io/ipv6-analysis. The raw dataset (all 100,000 hosts with their DNS records, reachability status, and HTTP response codes) is released under CC0 and available in the GitHub repository.
The web graph data used for host ranking comes from Common Crawl's open dataset. If you're interested in doing your own analysis, whether on IPv6 or anything else about the structure of the web, the web graph data is freely available at commoncrawl.org/web-graphs.
Erratum:
Content is truncated
Some archived content is truncated due to fetch size limits imposed during crawling. This is necessary to handle infinite or exceptionally large data streams (e.g., radio streams). Prior to March 2025 (CC-MAIN-2025-13), the truncation threshold was 1 MiB. From the March 2025 crawl onwards, this limit has been increased to 5 MiB.
For more details, see our truncation analysis notebook.
.svg)
