February 19th, 2026
IPv6 was defined in RFC1883 in December of 1995. That's right, IPv6 is now 30 frickin' years old, and we're still nowhere near universal adoption. What's worse, I still have to argue with people about why they should treat IPv6 as a first-class requirement, not a "nice to have" optional feature. Akamai, Cloudflare, and Google all show similar adoption rates of around 40% - 45% of HTTP traffic to the sites they each serve:
These companies provide the server-side view, with the ability to assess which networks or countries support IPv6, and one interesting observation there is that it looks like IPv6 traffic rises and falls following workday pattern, suggesting possibly higher availability of IPv6 on mobile or residential networks compared to business or perhaps off-hour usage patterns that favor services with higher IPv6 availability (e.g., streaming) compared to "business traffic":
But that peculiarity aside, this traffic is all HTTP as observed by the big service providers. What kind of IPv6 traffic adoption did I see on my puny little domain? Well, the answer was disappointing: only about 20% of HTTP traffic, 3.43% of SMTP traffic, and 2.34% of DNS traffic to this server is IPv6. Abysmal.
But this got me to wondering wondering just about
service availability in general. Rather than
focusing on clients, I was looking to answer the
question: what percentage of servers offers
IPv6? For that, I once again scanned the Tranco Top 1M
domains and checked each for the availability of
A and AAAA records for their www service names, their MX records, as well as their NS glue records.
NS records / DNS
Let's start at the DNS. The use of NS records and the concentration
of DNS providers is is a topic I've explored previously, but not
from an IPv6 angle. The root
zone is, of course, fully dual stack, but what
about the top-level domains (TLDs)? As of February 2026, there
are 1,436 TLDs in the root zone. Looking only at the
NS glue records in the root
zone:
| Total | % of TLDs | |
|---|---|---|
| Only IPv4-only glue records | 18 | 1.25 |
| Only IPv6-only glue records | 0 | 0 |
| At least one IPv4-only glue record | 240 | 16.71 |
| At least one IPv6-only glue record | 22 | 1.53 |
Ok, so that's fairly good: 98.75% of all TLDs use IPv6 for their name servers.
Now for the Top 1M Domains, I looked up the NS records and used their
additional data if provided by the name
server and otherwise manually looked up the NS's A
or AAAA
records.1
| Total | % of TLDs | |
|---|---|---|
| All NS IPv4 only | 260,739 | 27.99 |
| All NS IPv6 only | 10 | 0.001 |
| All NS dual-stack | 645,560 | 69.30 |
| At least one dual-stack NS | 670,054 | 71.93 |
| At least one NS IPv4-only | 281,475 | 30.22 |
| At least one NS IPv6-only | 834 | 0.09 |
| Dual-stack + at least one IPv4-only NS | 24,170 | 2.59 |
| Dual-stack + at least one IPv6-only NS | 104 | 0.01 |
| Dual-stack + IPv6-only + IPv4-only | 36 | 0.003 |
| Dual-stack + only IPv4-only | 24,134 | 2.59 |
| Dual-stack + only IPv6-only | 68 | 0.007 |
| At least one NS IPv6-enabled | 670,784 | 72.01 |
So in summary, for the Top 1M Domains, slightly over 70% of name servers have IPv6 addresses. That's a far cry from near universal availability as for the TLDs, but still not terrible.
WWW records / HTTP
Now on to HTTP traffic. As noted above, large content delivery networks obviously support IPv6, and as I noted previously, the majority of the Top 1M Domains are served by just a few large service providers, so you would expect pretty much all of them to likewise speak IPv6. But it turns out that even on those CDNs that offer IPv6, customers do indeed choose to actively disable it.
| Total | % of Top 1M2 | |
|---|---|---|
| IPv4 only | 557,996 | 55.80 |
| Dual-Stack | 353,223 | 35.32 |
| IPv6 only | 157 | 0.2 |
That is: only slightly more than a third of the Top 1M Domains are IPv6 enabled!
Now "Happy Eyeballs" has been around since "World IPv6 Day" 15 years ago (!), and RFC8305 dates back to 2017, but this doesn't assuage many conservative and change-averse industry verticals who, perhaps years ago, observed a minor performance penalty during early IPv6 testing and since then have left their "Disable IPv6" code, configs, and convictions in place unchanged.
MX records / SMTP
But of course the internet is more than just the
web, and HTTP is not the only protocol we care about.
I've covered email before (e.g., with respect to PQC support or the
distribution of email
service providers), so it made sense to take a
look at IPv6 support across the Top 1M Domains' MX records as well. For each
domain's MX records, I
checked their A and AAAA records to see how many mail
servers are, in theory at least3, reachable via
IPv6 and IPv4 on all, some, or none of their named
mail exchanges.
Looking at the Top 1M Domains, 643,241
domains have explicit MX
records4; for those, I've
found:
| Total | % of MX records | |
|---|---|---|
| All MX IPv4 only | 334,696 | 52.03 |
| All MX IPv6 only | 74 | 0.01 |
| All MX dual-stack | 288,849 | 44.91 |
| At least one MX dual-stack | 301,273 | 46.84 |
| At least one MX IPv4-only | 347,119 | 53.96 |
| At least one MX IPv6-only | 271 | 0.04 |
| Only IPv4-only and IPv6-only MXs | 89 | 0.01 |
| Dual-stack with at least one IPv4-only MX | 12,334 | 1.92 |
| Dual-stack with at least one IPv6-only MX | 108 | 0.01 |
| Dual-stack + IPv4-only + IPv6-only MX | 18 | 0.002 |
| Dual-stack + only IPv4-only MX | 12,316 | 1.91 |
| Dual-stack + only IPv6-only MX | 90 | 0.01 |
| At least one MX IPv6 enabled (dual-stack or IPv6-only) | 301,436 | 46.86 |
This comparison does not take into account the mail exchange's priority, but in summary: fewer than half of all of the Top 1M Domains' mail servers support IPv6!
There are many reasons why mail servers might conservatively favor IPv4-only configurations. "Happy Eyeballs" has always seemed to me primarily an HTTP focused approach, even though the RFC is protocol agnostic. I just haven't seen many SMTP servers implement it, which, in turn, means that any IPv6 misconfiguration leads to performance impact or even lost mail.
Additionally, a large number of SMTP abuse prevention mechanisms -- requiring matching reverse DNS records, IP block reputation scoring, SPF/DKIM/DMARC configuration -- see challenges with IPv6 adoption. But while I'm sympathetic to such concerns, they can be overcome: Google and Microsoft offer IPv6 for Gmail and Outlook respectively, so I'm going to have to give Yahoo, GoDaddy, and Namecheap (to name just a few of the big providers) some serious side-eye for remaining IPv4 only here.
Conclusion
The lack of IPv6 adoption is frustrating. For the two most widely used protocols, SMTP and HTTP, organizations choose to consciously disable IPv6, in part, I'm certain, based on decisions made ten, fifteen, or more years ago.
None of the problems service providers might encounter are impossible to overcome, but all of them suffer from the lack of critical mass adoption problem. As long as IPv6 is not seen as a fundamental requirement to do business, people will continue to disable it; as long as large businesses disable IPv6, it will not be seen as a fundamental requirement.
This is exacerbated by our educators failing our next generation: even today, in 2026, there are many engineering schools and universities that do not have IPv6 on campus! Computer Science and Engineering students are not exposed in their homework, projects, labs, and even routine use to IPv6, so when they enter the industry, they will continue to favor what's been "working" for them so far: NAT and RFC1918 overuse. Why bother with something they've never engaged with?
<Insert inspirational "let's roll up our sleeves and enable IPv6 everywhere" message here>, but I suspect we'll still not be there in another ten years, either. Universal IPv6 seems as elusive as Linux on the Desktop -- only that is actually an entirely viable alternative at this point. And people are clearly being overly hopeful about IPv6 adoption when I asked on Mastodon, at least in comparison, I suppose:
Wish I was so optimistic...
February 19th, 2026
Footnotes:
[1] Some DNS lookups simply failed due to a server side failure or yielded no results, so I ended up with a total of 931,515 domains instead of an even one million.↩
[2] Around 8.86% of top 1M domains
do not have any A / AAAA records for their
second-level or their www.<domain> name. This
includes domains such as, e.g., akamai.net or akadns.net, which are used by
Akamai for customer subdomains but don't themselves
serve as HTTP endpoints.↩
[3] I say "in theory", because I have observed some
widely used mail services to have AAAA records that are not
actually reachable. This is, of course,
worse than not having any AAAA records at all, because now
IPv6 enabled mail servers will attempt and then time
out on trying these addresses before hopefully failing
over to IPv4.↩
[4] For this exercise, I'm ignoring
NULL
MX records as well as no MX records. That is, I'm
not looking at what the raw domain resolves
to.↩
Links:
- Discussion on HackerNews
- Akamai IPv6 Statistics
- APNIC labs IPv6 stats
- Cloudflare IPv6 Statistics
- Google IPv6 Statistics
Related: