IPv6 in Russia in 2026: why four measurements give four different answers

"IPv6 adoption" names three different quantities, and confusing them produces most of the contradictory figures:

• User side — what share of people can reach IPv6. Measured by Google and APNIC.
• Traffic side — what share of requests actually arrived over IPv6. Measured by Cloudflare.
• Site side — how many domains carry an AAAA record. Measured by W3Techs and vyncke.org.

For a single country these differ by tens of percentage points. That is not measurement error but different denominators.

APNIC’s sample on the measurement day was 32,996,250 tests. Google publishes no sample size, but a strong weekly seasonality is visible: the figure runs higher at weekends.

This is where it gets interesting. Four independent measurements give incompatible answers:

Google and APNIC differ twentyfold, and both figures rest on weak ground.

Google’s Russian figure is extremely unstable. From archived copies of the same data file: August 2025 — 19.3%, December 2025 — 8.6%, March 2026 — 27.0%, July 2026 — 34.4%, 25 August 2026 — 59.94%. A collapse followed by a climb inside nine months marks this as a single-day reading, not a settled level. Publishing "Russia is at 60% IPv6" as fact would reproduce exactly the error we are correcting.

APNIC’s Russian sample has collapsed. In 2020 it ran 203,607 tests a day; in 2026, 2,546. Tests are delivered through Google’s advertising network, so wherever those ads are not shown the sample degrades. On 2,500 tests a day for a country with a hundred million users, swings of 2.6% → 9.7% → 3.1% are noise, not trend.

One measurement is free of survey-sampling problems: how many IPv6 prefixes allocated to Russia are announced in routing at all. Per vyncke.org on 27 August 2026, of 10,183 prefixes:

• announced individually — 3,318 (32.6%);
• announced within an aggregate — 1,947 (19.1%);
• not announced at all — 4,918 (48.3%);
• real traffic observed — 900 (8.8%).

Almost half the address space allocated to the country lies unclaimed, and traffic is seen from under nine percent of it. This is the least ambiguous figure in the whole set.

APNIC’s per-operator breakdown explains the low overall level: Rostelecom (AS12389) sits at 9.52%, while the autonomous systems with the most users sit at 0.04–0.05%. The high shares belong to hosting and transit: Timeweb 59%, FIORD 81%. Russia has IPv6 in its infrastructure and almost none at the consumer edge.

The honest conclusion for Russia is not a number but a disagreement: three of four independent metrics indicate low penetration (3–13%), with Google the sole outlier. Meanwhile nearly half the prefixes allocated to the country go unannounced — so the constraint is not a shortage of addresses.

In practice that means IPv4 cannot be switched off, while adding IPv6 alongside it is both possible and useful.

1. Check whether your domain has an AAAA record in the DNS lookup.
2. If it does, confirm the site actually answers on that address. An AAAA pointing nowhere is worse than none: some browsers try IPv6 first and only then fall back.
3. Check mail separately: among Russian top-100 domains, IPv6 is markedly more common in mail (29%) and DNS (39%) than on the website itself (13%).

Caveat on that last point: vyncke.org’s site list is built on the Alexa ranking and has been frozen since November 2016 — only the scan results are current, not the composition of the hundred.