Technical background
IPv6 adoption can be measured in at least three ways, and they rarely agree. User-side measurements count the fraction of real clients that reach a test service over IPv6; Google's public statistic, for instance, hovered in the mid-40s percent in 2024 and 2025, with strong weekend peaks because home and mobile networks are more IPv6-enabled than offices. Network-side measurements count ASNs and prefixes in the BGP table. Content-side measurements count how many popular domains publish AAAA records. This tool works at the network layer: it reads the prefixes an AS announces and reports how many are IPv6.
Prefix counts must be read with care, because the two families are allocated very differently. A mid-size ISP may hold dozens of IPv4 blocks acquired over twenty years and announce each one, or deaggregate them into /24s for traffic engineering, while a single IPv6 /32 covers its entire customer base with room to spare. RIR policies give LIRs a /32 or larger by default, and BCP 157 (RFC 6177) recommends giving end sites enough for many subnets, commonly a /48 for businesses and a /56 for homes. In the global table, /48 is the longest IPv6 prefix that is widely accepted, just as /24 is for IPv4. So an AS with 120 IPv4 prefixes and 2 IPv6 prefixes shows under 2 percent here and may still deliver IPv6 to every customer.
Operators reach IPv6 through several transition models. Dual-stack runs both protocols natively and is the common choice for fixed networks. Mobile carriers often run IPv6-only access and handle IPv4-only applications with 464XLAT (RFC 6877), which combines a client-side translator with NAT64 (RFC 6146) in the network; DNS64 (RFC 6147) synthesizes AAAA answers for IPv4-only names. DS-Lite (RFC 6333) tunnels IPv4 over an IPv6 access network to a provider-side NAT. On the client, Happy Eyeballs version 2 (RFC 8305) races IPv6 and IPv4 connections so a broken IPv6 path does not hurt the user experience.
Regional differences are large. LACNIC finished distributing its last IPv4 in August 2020, which pushed many Latin American ISPs to CGNAT and, increasingly, to IPv6. Brazil and Mexico have reported user-side IPv6 rates among the highest in the region, while several smaller markets remain in single digits. The regional cards on this page provide a fixed reference for that spread; for a real-time view, run the ASN check on the specific networks you care about.
Common errors and how to fix them
- Problem Concluding that a network has poor IPv6 adoption because its IPv6 percentage of prefixes is low.
- Fix Compare address coverage instead of route counts: one IPv6 /32 can serve more customers than all of an ISP's IPv4 blocks. Confirm with a user-side test or by checking whether customers actually receive IPv6 addresses.
- Problem Announcing IPv6 longer than /48, such as /56 or /64, expecting it to be reachable globally.
- Fix Most networks drop IPv6 routes longer than /48. Announce at least a /48 per site that needs independent routing, and aggregate customer /56s inside the provider's /32.
- Problem Publishing AAAA records for a service whose IPv6 path is broken, so some users see long connection delays.
- Fix Test the service over IPv6 from outside your network, check firewall rules and path MTU (ICMPv6 Packet Too Big must not be blocked), and only then publish AAAA.
- Problem Blocking all ICMPv6 at the firewall the same way ICMP was blocked for IPv4.
- Fix Allow the ICMPv6 types IPv6 needs to work, especially Packet Too Big and Neighbor Discovery, following RFC 4890 recommendations; blocking them breaks PMTUD and connectivity.
- Problem Numbering customers with a single /64 or a changing prefix on every reconnect.
- Fix Delegate a stable /56 or /48 per customer via DHCPv6-PD so they can build subnets and keep firewall rules valid, as BCP 157 recommends.
More questions about IPv6 Adoption Tracker
What percentage of the Internet uses IPv6?
It depends on what you measure. Measured by users, large public statistics put IPv6 reachability in the mid-40s percent globally in 2024 and 2025, with some countries well above half. Measured by networks, a growing majority of ASNs that announce routes include at least one IPv6 prefix, but IPv6 prefixes are still far fewer than IPv4 ones because a single IPv6 block covers much more.
Why do IPv6 prefix counts look so low compared with IPv4?
Because IPv6 blocks are huge and rarely need splitting. An ISP that announces dozens of IPv4 prefixes, accumulated over years and deaggregated into /24s, can cover its whole network with one IPv6 /32. Prefix ratios therefore understate adoption. A better question is whether the network announces any IPv6 at all and whether its customers receive IPv6 addresses.
Is IPv6 adoption in Latin America growing?
Yes. After LACNIC exhausted its IPv4 pool in 2020, many operators had to choose between buying addresses, deploying CGNAT or moving users to IPv6, and many did all three. Brazil has long been one of the region's leaders thanks to coordinated work by NIC.br and large ISPs, and Mexico, Uruguay and others show strong figures, while some smaller markets are still early in deployment.
Do I need IPv6 on my website if users can still reach it over IPv4?
It is strongly advisable. Many mobile users are on IPv6-only networks with NAT64, so an IPv4-only site reaches them through a translator that adds latency and a shared exit address. Publishing AAAA records on a properly working IPv6 path removes that hop, reduces dependence on CGNAT reputation and future-proofs the service at little cost.
What IPv6 prefix size should my company announce?
An organization with its own provider-independent space usually receives a /48 from its RIR, and /48 is also the longest prefix most networks accept, so announce it whole. If you have several sites that need independent routing, each needs its own /48. Providers receive a /32 or larger and should announce that aggregate, adding more-specifics only when necessary.
