Summary
- Hurricane Electric's Global IPv6 Deployment Progress Report, read on 19 November 2026, counts top-level domains, nameservers in the root zone and registered domains carrying AAAA records. In the text I received it carries no per-country adoption percentage, and its 2011 predecessor printed a different quantity entirely: the share of ASes in the routing table running IPv6.
- APNIC Labs builds a country metric from 25 to 30 million advertising impressions a day, reweighted to internet-user counts and averaged over thirty days. Google prints the share of its own users who reach Google over IPv6. LACNIC prints a country ordering with no measurement date and no window in the text I retrieved.
- Four publishers, four counted objects, one shared phrase. The published numbers are not the problem. The missing denominator, window and vantage are.
What the operator actually counts
The report is explicit about its purpose in its own first line: “This page was developed in order to provide metrics for global IPv6 deployment. These statistics are updated daily.” (Global IPv6 Deployment Progress Report) That is a statement about intent and about cadence. It is not a statement about who is counted.
What follows is a census of names. The retrieved text carries a root zone downloaded on Wednesday 18 November 2026 at 00:30:03 and processed the same day at 11:47:05, then 1,436 top-level domains, 1,420 of them with IPv6 nameservers at a printed 98.9 per cent, and 1,418 with IPv6 glue in the root zone at 98.7 per cent. Divide 1,420 by 1,436 and you get 0.98886, which reproduces the printed percentage — my division, not theirs.
The section after the domain list is headed “Registered domains with AAAA records” and describes itself in one sentence: “Generated by querying for A (IPv4) and AAAA (IPv6) records for all domains in the listed TLDs (top level domains).” That is a precise description of an input. In the .com row the adapter returned to me 169,267,016 domains, 149,232,466 A records and 31,854,060 AAAA records.
Two ratios come out of those three numbers, and both are mine. 31,854,060 divided by 169,267,016 gives 18.8 per cent of that row's domains carrying an AAAA record; multiply back and 169,267,016 × 0.18819 reproduces the stated figure to within 1,716 records at the precision printed. The second is the one worth keeping: 149,232,466 A records plus 31,854,060 AAAA records is 181,086,526, which exceeds 169,267,016 domains by 11,819,510. The two record columns are not a partition of the domain count.
The section's own sentence does not say whether a domain can contribute more than one record, so that excess is a fact about the columns and not a count of dual-stack domains. Anyone reading it as “11.8 million dual-stack domains” would be reading something the page never claimed.
The older version measured a different object, and said which one
The same report in its 7 October 2011 archive carried a section the current text did not show me: “Networks Running IPv6”, with a method stated in the same breath as the result. “We can measure the percentage of networks running IPv6 by comparing the set of ASes in the IPv6 routing table to those in the combined set of IPv4 and IPv6.” Then: “Percentage of ASes (IPv4 or IPv6) running IPv6: 12.0%.” (2011 archive)
Read the two versions side by side and the shape of the change is plain. The 2011 instrument named a denominator — autonomous systems visible in a routing table — and reported a share of it. The 2026 instrument names its inputs, a root zone and a DNS query, and reports counts. Neither is a national user sample. A 2011 figure of 12.0 per cent and a 2026 figure of 98.9 per cent of TLDs are not a growth curve for the same quantity; they are two different instruments that happen to share a URL family.
The country table is a routing footprint
The country-level table this operator does publish is a separate page, and its columns are Country, Prefixes, ASNs and Prefixes per ASN (BGP Prefix Report). That is a description of announced address space and the networks originating it. It is not a share of users, domains or traffic, and nothing here reads it as adoption.
It also does not hold still. Two retrievals of that page, hours apart, returned a United States row of 345,463 prefixes across 23,686 ASNs in one and a differently labelled table with 70,911 prefixes across 6,184 in the other. That is a factor of about 4.9 on the same country, with no printed report date in either text. I am not going to print a country number whose anchor I cannot state, so neither set of figures appears anywhere in this article. The disagreement is the finding: a page without a date cannot tell you what changed and what merely moved.
What APNIC counts, and the two windows it printed
APNIC Labs is where a per-country “IPv6 capable” percentage actually comes from. Its page labels the metric “IPv6 Capable Rate by country (%)”, offers plot types for IPv6 Capable, IPv6 Preferred and Measurement Sample Size, and prints a thirty-day average (IPv6 Measurement Maps).
Two retrievals of that same URL, minutes apart, printed two different windows: “30 day average (19/10/2026 - 17/11/2026)” and “30 day average (20/10/2026 - 18/11/2026)”. Neither retrieval carried printed per-country values for France, India, Germany, the United States or Japan — the numbers are rendered in a form the text I received did not include. So this article quotes the metric label, the plot types and the window, and no national rate at all.
The method is documented, and it is unusually specific about its vantage. The November 2025 post: “The measurement system is configured to present between 25M to 30M ad impressions per day. The ad impression pattern across the Internet is not uniform, so we use additional data from the UN Statistics Division and the ITU-T to relate the number of ad impressions per economy per day to the Internet user count per economy per day.” (How we measure: IPv6) The April 2025 post describes the sample differently: “matched DNS name sets with IPv4-only, IPv6-only, or dual-stack addresses”, each sample “tagged by its economy and origin-AS, then weighted using World Bank Internet user data”, results “averaged over the past 30 days” (IPv6 capability reaches 50%).
Seven months apart, two of APNIC's own posts name its weighting input two different ways — World Bank internet-user data in April, the UN Statistics Division and the ITU-T in November. Both are printed here as printed. Which is in force today is not established by anything I read, and nothing in this article is computed from either.
The two labels are also not interchangeable, and APNIC says so itself. “If APNIC Labs can fetch a URL using IPv6, then the host system is labelled ‘IPv6 capable’. However, if it uses IPv6 to fetch the dual-stack URL, it's labelled ‘IPv6 preferred’.” A capable rate and a preferred rate are different quantities measured on the same sample, and the page's own plot-type control lets a reader switch between them. A citation that says “IPv6 adoption” without saying which of the two it means has already lost one bit of meaning.
What Google counts
Google states its population in the first sentence: “We are continuously measuring the availability of IPv6 connectivity among Google users. The graph shows the percentage of users that access Google over IPv6.” (IPv6 Statistics) That is a clean, honest sentence, and it is a sentence about Google users reaching Google. It describes one service's reachable population. It is not a country's internet, and it is not a sample of the world.
What LACNIC prints without a date
LACNIC's own ranking page prints a country ordering whose top five read France 86.5, Russia 77.13, Germany 75.84, Belgium 74.84 and Saudi Arabia 69.2 (LACNIC Stats - IPv6 ranking). The retrieved text carries no measurement date, no averaging window and no definition of the quantity in the column. That is enough to display an ordinal list. It is not enough to rank that list against APNIC's thirty-day figures, and I do not do it.
Three clocks on one publisher's surface
The operator's own toolkit home prints “Updated 18 Nov 2026 15:22 PST” and links a quick item labelled IPv6 Progress (Hurricane Electric BGP Toolkit). The progress report prints its own 00:30:03 download and 11:47:05 processing stamps. The report also says its statistics are updated daily. Three clocks, one publisher, and the retrieved text does not state which time zone the report's stamps use. All three are reported. None is adopted, and the one that would matter most to a reader trying to reproduce a number — the window — is the one that is not there.
So what does a citation have to carry
A published measurement is auditable when it binds the result to its scope: the counted unit, the population, the exclusions, the observation window, the vantage that could see the sample, and the weighting that turned a sample into a population figure. Give a reader those six and two numbers that disagree become informative — they measure different things, or the same thing over different windows. Withhold them and the number can only be taken on the publisher's authority.
That is not an accusation, and this article makes none. Every one of the four pages is doing something defensible: an operator counting names in the namespace it can see, a research lab reweighting ad impressions, a service counting its own users, a registry printing a ranking it has not dated. The failure mode is entirely in the citation layer — in the habit of lifting a number out of one instrument and dropping it into a sentence written for another.
A regulator writing an IPv6 target, an operator sizing a transition programme, an investor benchmarking a network: each needs the population before the percentage, or the percentage is decoration.
The reserved premise for this piece said the operator publishes the widely quoted national ranking and that I would read it off the statistics page. The page does not carry one. Its 2011 predecessor carried a global share of ASes, and its current successor carries a census of names. That correction is the article: the name travels with the ranking, the ranking does not travel with the name.
Boundaries
- Every page here was read as text returned by a retrieval adapter. No page image was inspected.
- The operator's progress report came back truncated inside the long TLD list on both retrievals, so a section beyond that point cannot be ruled out. The absence of a per-country adoption table is an observation about the text I received on 19 November 2026, not a proof about the page.
- No printed per-country value from APNIC Labs was present in either retrieval of its map. The metric label, the plot types, the sample-size control and the thirty-day window are what the page text carried, and they are what this article quotes.
- The same APNIC URL printed two different thirty-day windows minutes apart, 19 October to 17 November 2026 and 20 October to 18 November 2026. Even the window label is a reading taken at a moment.
- Two retrievals of the operator's BGP Prefix Report returned country rows differing by roughly a factor of five for the United States, and neither printed a report date. Neither set of figures is used anywhere here.
- The BGP Prefix Report counts announced prefixes and originating ASNs by country. It is not a share of users, domains or traffic, and nothing here reads it as adoption.
- The 2011 archived report is a historical snapshot of an earlier format. It does not establish what the current page prints, and its 12.0 per cent is a 2011 routing-table reading, global rather than per-country.
- APNIC names its weighting input two different ways in two of its own posts seven months apart — World Bank internet-user data in April 2025, the UN Statistics Division and the ITU-T in November 2025. Both are quoted as printed; which is in force today is not established here, and nothing in this article is computed from either.
- LACNIC's ranking page states no measurement date and no averaging window in the text I retrieved, so its ordering cannot be ranked against APNIC's thirty-day figures and no such comparison is made.
- IPv6 capable and IPv6 preferred are two different labels in APNIC's own definitions: whether a tested client could retrieve an IPv6-only object, and whether it used IPv6 for a dual-stack object. Where both exist they are not the same quantity.
- Google's figure is a statement about Google users reaching Google over IPv6 — one service's user population, not a country's internet.
- The report's sentence for the AAAA section says it queries A and AAAA records for all domains in the listed TLDs. It does not state whether a domain can contribute more than one record, so the 11,819,510-record excess I computed is a fact about those columns and not a count of dual-stack domains.
- The toolkit stamp of 18 November 2026 15:22 PST, the report's 00:30:03 download and 11:47:05 processing stamps, and the report's sentence that its statistics are updated daily are three different clocks on one publisher's surface, and the retrieved text does not state which time zone the report's stamps use. All three are reported; none is adopted.
- No claim here is that any publisher's figure is wrong, that any country's readiness is overstated or understated, that any operator manipulated a statistic, or that any of these four instruments is better than another. The difference between them is a difference in what each one counts.
- Prior coverage named rather than repeated: BTW's 10 July 2026 company research on this operator read its transit scale, shared hosting and long-standing IPv6 reputation; six upstream-relationship profiles name it as a counterparty; a 20 August 2025 event item recorded a Toronto point-of-presence launch. None of them read a publisher's measurement definition, which is the delta this briefing claims.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






