Which ISPs are still dragging their feet on IPv6?


Which ISPs are actually deploying IPv6 — and which aren’t?

Country-level IPv6 adoption figures obscure more than they reveal. Japan sits at ~58%, the US at ~59% — but those numbers aggregate across every network in each country. The decision to deploy IPv6 is made by ISP engineering teams, and a network-level view tells a different story.

We looked at the world’s largest ISPs by IPv4 address allocation and compared how many IPv4 versus IPv6 prefixes each announces in the global routing table. A note on what these numbers represent: these are BGP announcements — routes seen by BGP collectors and validated against RPKI and IRR records — not RIR allocations sitting on paper. A prefix that has been allocated but never announced is not carrying traffic, and doesn’t appear here.

This comparison is deliberately symmetric: both are routing table entries, both reflect operational decisions, and an ISP that has deployed IPv6 to its customer base at scale will have announced customer-facing IPv6 blocks at roughly the same order of magnitude as its IPv4 footprint. Comcast, with 1,132 IPv4 prefixes and 1,078 IPv6 prefixes, is close to that parity. Most large ISPs are not.

ASN ISP Country IPv4 prefixes IPv6 prefixes
AS7922 Comcast US 1,132 1,078
AS9808 China Mobile CN 15,569 9,148
AS4134 China Telecom CN 4,229 2,088
AS4837 China Unicom CN 1,446 548
AS3320 Deutsche Telekom DE 885 151
AS7018 AT&T US 1,691 39
AS4766 Korea Telecom KR 3,111 36
AS701 Verizon Business US 1,626 42
AS3269 Telecom Italia IT 362 1
AS4713 NTT Docomo JP 183 6
AS9318 SK Broadband KR 1,324 4
AS2856 British Telecom GB 502 7
AS17676 SoftBank JP 480 3
AS1221 Telstra AU 280 2

Routing table presence is a prerequisite for customer-facing IPv6 deployment, not a sufficient condition — but its absence is conclusive. SoftBank has 480 IPv4 prefixes and 3 IPv6 prefixes — Japan’s second-largest mobile carrier is essentially absent from IPv6 routing, which explains why Japan’s 58% national figure is carried heavily by other networks. Telecom Italia has 362 IPv4 prefixes and one IPv6 prefix announced.

Accounting for historical fragmentation

Raw prefix counts have a known limitation: an ISP that grew through decades of acquisitions inherits many small, non-contiguous IPv4 blocks. IPv6 allocations are planned from scratch and tend to be far more aggregated. Comparing raw counts is therefore not always apples-to-apples.

To address this, we re-ran the analysis against the unfiltered BGP table — routes regardless of RPKI or IRR status — and computed contiguous address blocks per ASN, merging adjacent prefixes into single blocks. Two things came out of it.

First, RPKI and IRR filtering had removed zero IPv6 prefixes for any of the ISPs in this table. The prefixes that exist in the unfiltered view but are absent from the filtered one turn out to be routes seen by a single BGP peer, or announcements with IRR origin mismatches — neither of which represents real operational deployment. The filtered view is actually the more honest measure.

Second, the aggregation check does change the picture for some ISPs. The fairest comparison is below:

ASN ISP IPv4 contiguous blocks IPv6 /32 blocks
AS7922 Comcast 187 18
AS7018 AT&T 369 16
AS3320 Deutsche Telekom 695 21
AS4766 Korea Telecom 515 21
AS3269 Telecom Italia 94 3
AS4713 NTT Docomo 118 4
AS9318 SK Broadband 259 4
AS2856 British Telecom 344 5
AS17676 SoftBank 86 3
AS1221 Telstra 197 2

AT&T’s 681 raw IPv4 prefixes collapse to 369 genuinely separate address blocks — the fragmentation argument holds there. And routing table presence alone doesn’t tell the whole story: APNIC’s client-side measurement puts AT&T at 84% IPv6-capable users, confirming that 16 IPv6 blocks is enough to serve their customer base. AT&T belongs closer to Comcast in this picture.

For SoftBank, Telecom Italia, British Telecom, and Telstra, the aggregation argument doesn’t change the conclusion. Their IPv4 side collapses too, but IPv6 stays near zero regardless.

A note on APNIC’s measurement coverage

APNIC’s per-ASN client-side data is a strong cross-check where available, but it comes with a geographic caveat worth understanding. Their method works by embedding a small piece of JavaScript in Google ads — when a user loads a page showing a Google ad, the script silently attempts to fetch a resource over both IPv4 and IPv6 and reports back which succeeded. This gives APNIC broad, passive coverage without any ISP cooperation.

The catch is that coverage is only as strong as Google’s ad presence. In the US, Europe, India, and Japan, Google ads are everywhere and the samples are large and reliable — AT&T’s 84% figure is almost certainly solid. In China, where Google is largely blocked by the Great Firewall, APNIC’s sample is tiny and almost certainly unrepresentative. China’s 44% APNIC figure should be treated with caution. Russia, Iran, and parts of Africa and Southeast Asia fall somewhere in between.

The practical implication: the ISPs where APNIC data is most trustworthy happen to be in markets where our routing table analysis is also strongest. For the laggards in markets with thinner Google coverage, routing table presence remains the more reliable signal — and for those networks, the numbers above still stand.

Global IPv6 adoption will not reach the levels needed to reduce IPv4 dependence through organic growth alone. It requires specific large networks to move. Routing table data shows which ones haven’t started.