dns.pizza research
The State of IPv6 2026: 31.4% of the top 100,000 domains answer AAAA at their apex
IPv6 was specified in 1998 and the free IPv4 address space ran out in 2011. We asked 100,941 domains of the Tranco top 100,000 for an AAAA record on their own name. 31,692 answered. Which ones did barely moves with how popular a site is; it varies sharply across the companies running their DNS.
Published November 15, 2026 · data collected June 2026–November 2026 · download the data
31.4%
answer AAAA at the apex
36.4%
of the 86,952 domains whose apex answers any address at all
9×
more AAAA on Cloudflare than on Amazon Route 53, the two largest DNS operators here
35
domains answer AAAA and no A record
Key findings
- Roughly two in three of the web's most-visited names still have no IPv6 address of their own. 55,260 of 100,941 domains (54.7%) answer an A record and no AAAA. Another 13,989 (13.9%) answer neither at the apex. Only 31,692 (31.4%) publish an IPv6 address — 36.4% if you count only the 86,952 domains whose apex is addressable at all.
- Almost nobody publishes IPv6 on its own. 31.4% of the corpus publishes both families; 35 domains do publish AAAA and no A, about one in 2,884. In this snapshot IPv6 is something a domain has in addition to IPv4, essentially never instead of it. That makes this a report about who has bothered rather than about a transition, though a single snapshot cannot show movement in either direction.
- There is no rank gradient — and the one the raw numbers seem to show is an artifact. Measured against every domain, adoption looks like it rises as sites get less popular: 27.8% in the top 1,000, 30.5% from 1,001–10,000, 31.5% in the long tail. That is the addressless apexes talking. 22.7% of top-1,000 domains publish no address on the bare name at all, against 13.5% in the long tail. Why the biggest names skew that way is not something this data answers — serving from
wwwand being an API or CDN endpoint rather than a website are both plausible, and we measured neither. Count only addressable apexes and the three tiers land within half a point of each other: 36.0%, 36.4% and 36.5%. Within this corpus, rank tier is not where the variation lives. - IPv6 tracks the DNS operator far more sharply than it tracks anything else we measured. The two largest DNS operators here are 9× apart: 72.3% of the 32,294 domains whose nameservers are all Cloudflare answer AAAA, against 8.0% of the 18,257 on Amazon Route 53. On the same denominator these two figures use, the whole rank-tier range spans 3.7 points (27.8% to 31.5%, and under half a point once you restrict to addressable apexes); the operator moves it by a factor of 9. What we are looking at is an association, not a demonstrated cause. Possible explanations include platform defaults and customer self-selection — the kind of customer who picks a CDN-shaped DNS provider may simply be the kind who wants IPv6 — and nothing in this data distinguishes them. We do not measure what any provider ships by default. At the other end, 2 qualifying operators have not one AAAA-answering domain between them, which is why this report quotes no best-versus-worst ratio: the floor is a real zero.
- 77.8% of the IPv6-answering domains in the single-operator cut sit behind one company. Of the 29,991 single-operator domains that answer AAAA, 23,344 are on Cloudflare. That share is of the single-operator population only — not of all 31,692 AAAA-answering domains in the corpus, because the domains excluded from that cut have no published operator here. Even so, IPv6 in this corpus is not spread evenly across the industry: it piles up behind a few large platforms.
- Where IPv6 exists, it is usually more than one address. 85.7% of AAAA-bearing domains publish two or more distinct IPv6 addresses (27,149 of 31,692), for 73,520 published AAAA records across the corpus (a count of records, not of distinct addresses — one CDN address shared by a thousand domains is a thousand records) — an average of 2.3 per domain. The two families are usually published at the same width: of the 31,657 dual-stack domains, 29,121 publish exactly as many AAAA records as A records, and only 571 (1.8%) publish fewer. That is a statement about records only — we resolve names and never connect, so this says nothing about capacity, redundancy or whether anything answers on those addresses.
- Adoption is a registry story too. Among the 20 largest TLD samples we publish, .nl leads at 49.2% against 7.3% in .ru — a 6.8× spread across namespaces running the same protocol.
- 13,989 apexes answer no address at all, and that is two different things. 1,199 of them came back NXDOMAIN on every query: the name itself is gone. The other 12,790 did not, and simply published no address on the bare name. What those names are used for instead — a
wwwhostname, mail only, something else — is not measured here; all we know is that the apex carries no address. Those are the domains this report cannot see the IPv6 of, which is why every share is also published against the addressable denominator.
Address families by popularity tier
Share of domains in each Tranco rank tier by what their apex publishes. The three segments partition each bar exactly; the AAAA segment folds the dual-stack and IPv6-only families, which the table below separates.
Source: dns.pizza DNS record checks · snapshot November 15, 2026
IPv6 adoption by DNS operator
The 10 largest operators by corpus size, ranked by adoption. Each bar is a share of that operator's own single-operator domains, so the bars are independent populations and not slices of one total.
Source: dns.pizza DNS record checks · snapshot November 15, 2026
Adoption by tier and TLD
| Segment | Domains | Has AAAA | Of addressable | Dual-stack | IPv4 only | No address |
|---|---|---|---|---|---|---|
| All domains | 100,941 | 31.4% | 36.4% | 31.4% | 54.7% | 13.9% |
| Top 1,000 | 1,002 | 27.8% | 36.0% | 27.8% | 49.5% | 22.7% |
| 1,001 – 10,000 | 9,332 | 30.5% | 36.4% | 30.5% | 53.4% | 16.0% |
| 10,001 – 100,000 | 90,607 | 31.5% | 36.5% | 31.5% | 54.9% | 13.5% |
| .com | 46,884 | 33.2% | 38.2% | 33.2% | 53.8% | 13.0% |
| .net | 6,527 | 26.5% | 37.3% | 26.4% | 44.5% | 29.0% |
| .org | 4,181 | 45.4% | 48.3% | 45.4% | 48.7% | 5.8% |
| .ru | 3,858 | 7.3% | 8.0% | 7.3% | 84.3% | 8.4% |
| .io | 1,929 | 34.0% | 44.5% | 34.0% | 42.4% | 23.6% |
| .jp | 1,811 | 7.9% | 11.1% | 7.9% | 63.0% | 29.1% |
| .de | 1,793 | 30.5% | 32.1% | 30.5% | 64.4% | 5.2% |
| .br | 1,409 | 33.9% | 36.6% | 33.8% | 58.6% | 7.5% |
| .uk | 1,348 | 27.7% | 31.3% | 27.7% | 60.8% | 11.4% |
| .in | 1,264 | 27.3% | 29.9% | 27.3% | 63.9% | 8.8% |
| .fr | 992 | 23.9% | 25.3% | 23.9% | 70.4% | 5.7% |
| .cn | 972 | 13.4% | 22.0% | 13.2% | 47.4% | 39.2% |
| .it | 865 | 21.4% | 22.8% | 21.4% | 72.3% | 6.4% |
| .co | 853 | 37.2% | 42.8% | 36.8% | 49.6% | 13.2% |
| .edu | 829 | 22.4% | 22.9% | 22.2% | 75.4% | 2.2% |
| .pl | 768 | 32.8% | 33.6% | 32.8% | 65.0% | 2.2% |
| .nl | 648 | 49.2% | 51.0% | 49.2% | 47.4% | 3.4% |
| .es | 554 | 24.2% | 26.2% | 24.2% | 68.1% | 7.8% |
| .au | 533 | 25.7% | 26.6% | 25.5% | 70.9% | 3.4% |
| .tv | 528 | 39.0% | 42.5% | 39.0% | 52.8% | 8.1% |
“Has AAAA”, “Dual-stack”, “IPv4 only” and “No address” are shares of the segment's domains; the last three plus the IPv6-only sliver partition it exactly. “Of addressable” is the AAAA share of the segment's domains that answer any address. TLDs need at least 200 measured domains to appear, and the 20 largest qualifying samples are shown; 618 smaller TLDs covering 10,299 domains are not broken out.
IPv6 by DNS operator
Each row counts the domains whose entire nameserver set belongs to that one operator — 89,733 of the 100,941 measured domains. Restricting the cut this way makes the rows a partition: every domain appears in exactly one, nothing overlaps, and the column really does add up. The 9,322 domains spread across two or more operators are left out of this table rather than counted in several rows, and 1,840 more had no parseable nameserver recorded.
| DNS operator | Domains | Has AAAA | Dual-stack | IPv4 only | 2+ IPv6 addresses |
|---|---|---|---|---|---|
| Cloudflare | 32,294 | 72.3% | 72.3% | 21.0% | 97.1% |
| Amazon Route 53 | 18,257 | 8.0% | 8.0% | 70.4% | 82.8% |
| Akamai | 2,457 | 28.5% | 28.5% | 56.1% | 93.7% |
| GoDaddy | 2,042 | 7.2% | 7.2% | 87.8% | 44.9% |
| 1,975 | 28.8% | 28.8% | 56.8% | 34.8% | |
| Microsoft Azure DNS | 1,457 | 5.5% | 5.4% | 76.3% | 47.5% |
| Alibaba Cloud DNS | 1,371 | 7.1% | 7.0% | 60.9% | 57.7% |
| Tencent DNSPod | 944 | 7.4% | 7.2% | 63.1% | 45.7% |
| Namecheap | 649 | 8.3% | 8.3% | 71.0% | 64.8% |
| CSC | 601 | 14.8% | 14.8% | 81.4% | 67.4% |
| UltraDNS (Vercara) | 579 | 22.1% | 21.9% | 68.0% | 75.8% |
| NS1 (IBM) | 463 | 22.0% | 22.0% | 61.6% | 84.3% |
| DNS Made Easy | 413 | 7.7% | 7.7% | 84.5% | 65.6% |
| RU-CENTER | 376 | 1.6% | 1.6% | 88.3% | 50.0% |
| Yandex | 367 | 13.6% | 13.6% | 77.9% | 0.0% |
| ClouDNS | 320 | 13.4% | 13.4% | 73.4% | 48.8% |
| Gandi | 318 | 13.2% | 13.2% | 83.3% | 45.2% |
| Selectel | 298 | 1.0% | 1.0% | 86.2% | 33.3% |
| REG.RU | 275 | 3.3% | 3.3% | 87.6% | 0.0% |
| OVHcloud | 252 | 15.1% | 15.1% | 84.1% | 13.2% |
| pbsg.com | 230 | 0.0% | 0.0% | 98.3% | 0.0% |
| DigitalOcean | 227 | 19.8% | 19.4% | 62.1% | 57.8% |
| Network Solutions | 225 | 8.9% | 8.9% | 89.8% | 80.0% |
| parklogic.com | 201 | 0.0% | 0.0% | 99.0% | 0.0% |
| All smaller operators | 23,142 | 12.3% | 12.3% | 74.4% | 36.9% |
Operators need at least 200 single-operator domains to get a row; “All smaller operators” folds 10,152 of them covering 23,142 domains. “2+ IPv6 addresses” is a share of that operator's AAAA-bearing domains. A row named like a bare domain rather than a company is an operator we do not map to a brand, which keeps its own nameserver zone as its identity rather than being folded in with others; the two such rows showing 0% are domain-parking services, and neither sample contains a single apex that answers AAAA. An operator's name here describes who answers DNS for the domain, which is not necessarily who hosts the website — though for the CDN-shaped operators at the top of this table it usually is both.
Download the full dataset (CSV)Methodology
dns.pizza continuously resolves DNS records — including A and AAAA — while generating domain intelligence reports across the Tranco top 100,000. This report takes the latest DNS report for each domain and then drops the domain if that report's A or AAAA lookup errored — note the order: a domain whose newest report errored is excluded rather than falling back to an older, cleaner one, so the corpus is always the most recent observation. That leaves 100,941 domains, measured between June 10, 2026 and November 15, 2026.
What is measured is the apex name, and only the apex name. We ask example.com for AAAA, not www.example.com and not any other hostname. A site served only at www can have a dual-stack www and a bare name with no address at all; this corpus holds no observation of those hostnames, so such a domain reads here as “no address”. Nothing in this report should be read as “this site is unreachable over IPv6” — the claim is narrower and literal: this name does or does not answer AAAA. Where a CNAME sits at the apex the resolver follows it, so a domain aliased to a dual-stack CDN target does count as having AAAA.
Two denominators, because one of them is honest about that limit. “Has AAAA” is of all 100,941 conclusive domains. “Of addressable” is of the 86,952 whose apex answers A or AAAA — a denominator that cannot be deflated by a www-only or dead apex, because every domain in it resolves to something. The two differ by 13.9 points of the corpus, and the second is the number to quote if you want the share among domains that are actually using their bare name.
One vote per domain. The corpus is every domain our rank table places at 100,000 or better. That is 101,638 distinct domains rather than exactly 100,000, because the ranking has ties — 96,825 distinct rank values are shared among them — which is also why the “Top 1,000” tier holds 1,002 domains. Domains themselves are unique, so no domain is counted twice.
An empty answer is an absence; only an error is unknown. Of 101,637 ranked domains with a DNS observation, 696 are excluded from every denominator because the A or the AAAA lookup itself errored — 688 on AAAA, 644 on A, mostly SERVFAIL. This report compares the two families within each domain, so either one being unknown makes the pair unusable. Everything else stays in: ENOTFOUND, ENODATA and NXDOMAIN are definitive absences rather than failures, and are recorded as a null record with no error attached. Treating those nulls as “unknown” would discard 68,621 genuine no-AAAA domains and report adoption close to 100%, which is the single easiest way to get this measurement spectacularly wrong. A further 628 domains returned an explicitly empty AAAA set and are likewise counted as having no AAAA.
Per-domain address counts are counts of distinct addresses: a resolver that returned the same address twice does not read here as a redundant pair.
The operator table is a partition by construction. A domain can use two DNS operators — 9,322 in this corpus do — and a table built from “domains that use this operator at all” would count such a domain in two rows, so its column could not be added up. (Our nameserver concentration report published exactly that mistake and had to correct it.) Rather than manage the overlap, this report avoids it: the operator cut takes only the 89,733 domains whose entire nameserver set maps to one operator. “Entire” is literal — every nameserver host we hold for the domain has to classify, not merely one of them, so the 46 domains with a half-readable set are excluded alongside the 9,322 multi-operator ones rather than quietly admitted on the strength of their readable half. Both the pipeline and the test suite assert that the rows sum to exactly that population and that the four coverage states partition the corpus.
Nameserver hosts are grouped to a provider zone (last two labels, three under a multi-part public suffix) and then mapped to an operator, because one operator owns many zones: Amazon Route 53 alone is spread over hundreds of awsdns-NN.* zones. The mapping is imported from the same module the nameserver concentration report and the nameserver-stats job use, so it cannot drift between reports. Zones the mapping does not recognize keep the zone itself as their operator identity and are never folded into a shared bucket — folding them would make a domain on ns1.foo.com plus ns2.bar.net look single-operator and wrongly admit it to a cut that is supposed to contain only single-operator domains. Because two zones owned by the same unrecognized company count as two operators, the single-operator population is a floor, and the excluded multi-operator count is a ceiling.
The nameserver set comes from the denormalized record on our rank table, refreshed by the prewarm and nameserver-stats jobs, while the address records come from the domain's latest report. The two are selected independently, so an operator attribution and an address observation need not come from the same moment. Every count in this report comes out of a single SQL statement, so no two figures in it are read from different database snapshots.
Limitations: the corpus is the top-ranked slice of the web (Tranco aggregates several popularity signals; it is not a literal visit count). Every figure here describes popular domains and nothing else — we measured no comparison population, so none of it should be restated as a number for the web at large. An AAAA record says an address is published, not that it is reachable, not that the server behind it answers, and not that the path to it works end to end — we resolve records, we do not connect to them. The reverse direction is not measured at all: whether a domain's own nameservers are reachable over IPv6 is a different question from whether its website is, and this corpus does not persist those lookups. Mail is also out of scope, since we hold no AAAA observation for MX hosts. And an address family is not a security posture: a dual-stack parked domain is not better run than an IPv4-only one that is.
The data is free to cite and reuse (CC BY 4.0, credit "dns.pizza"). For questions or the raw per-domain data, get in touch.
Does your domain answer on IPv6?
A health check reports your apex AAAA record and whether your nameservers have IPv6 addresses of their own — or confirms you're one of the 54.7% running IPv4 only.