IPv6 in 2026: closer than you think, still not there
Forty-five percent of Google traffic is now over IPv6 — up from 36% three years ago. The protocol that's been "the future" since 1998 is finally, slowly, becoming the present.
If you opened a networking textbook in 2010, it told you IPv4 would be exhausted "within a year or two" and that IPv6 was about to take over. That prediction was half right. IPv4 addresses did run out — IANA assigned its last block in 2011 — but IPv6 didn't take over. It just sort of... showed up at the party, found a corner, and stayed there for a decade.
Then, somewhere between 2022 and 2026, things started moving. Google's IPv6 adoption metric — which measures the fraction of users reaching Google services over IPv6 — has been the gold-standard public indicator for adoption since 2008. In June 2023 it sat at about 36%. In May 2026 it's 45%. That nine-point gain in three years is the largest sustained climb the metric has seen since 2016.
So what changed? And how much further is there to go?
Who's actually using IPv6
The headline 45% figure hides enormous variation by country. The leaderboard is dominated by three groups: countries with strong national broadband policy (France, Germany, Belgium, Greece), countries with massive mobile networks that bypassed IPv4 entirely (India, Vietnam, Saudi Arabia), and countries with progressive telcos (the United States, partly).
| Country | IPv6 adoption | Driver |
|---|---|---|
| France | ~78% | Free / Orange aggressive defaults |
| India | ~75% | Reliance Jio mobile-only |
| Germany | ~72% | Telekom dual-stack on every consumer line |
| Vietnam | ~68% | National IPv6 policy 2018-2025 |
| United States | ~55% | Comcast, Verizon, T-Mobile |
| United Kingdom | ~43% | Sky, BT |
| China | ~38% | Government mandates |
| Brazil | ~30% | Mobile operators |
| Russia | ~14% | Slow telco rollout |
| Italy | ~12% | Almost zero ISP movement |
The gap between France at 78% and Italy at 12% is bigger than the gap between France and the IPv4 internet itself. There's no technical reason for the divergence — both countries run on the same hardware vendors, the same operating systems. It's almost entirely a question of whether their dominant residential ISP made dual-stack the default ten years ago or never bothered.
Why the second curve is mobile
The biggest single contributor to global IPv6 growth this decade hasn't been desktops, fiber, or enterprise — it's been mobile operators. The reason is simple and unromantic: they ran out of IPv4 addresses for their customers.
A US mobile carrier has tens of millions of active devices. If each device needs its own IPv4 address, that's tens of millions of IPv4 addresses — and the carrier can't buy them anymore at any sane price. The market price for a single IPv4 address has hovered around $50-60 since 2023. A million addresses costs you $50M. For a carrier with 100 million devices, that's $5 billion in addresses they don't actually own.
The math forced a choice: deploy CGNAT (carrier-grade NAT, which puts a million customers behind one shared public IPv4 and breaks half the internet for them), or deploy IPv6 (which gives every device its own address and doesn't break anything). Most large mobile networks chose dual-stack with IPv6 preferred, falling back to CGNAT only for the IPv4-only services. Today, when you check whether you have IPv6 on an LTE or 5G connection in the US, the answer is "yes" about 80% of the time. On wired home broadband, it's still closer to 55%.
Why residential broadband is slower
Home internet ISPs don't have the same forcing function. They can hold onto their IPv4 blocks for decades. A consumer customer with a /29 of IPv4 isn't generating new addressing pressure. So adoption depends on whether the ISP cares to enable IPv6 — and many don't, because the customer doesn't notice.
Comcast in the US is the standout exception. They started dual-stacking their residential network in 2011 and were essentially done by 2017. The user doesn't see anything different; their router gets both an IPv4 and an IPv6 prefix, and Comcast's edge routes the bits accordingly. The IPv6 is just... there. Today about 85% of Comcast residential connections have working IPv6.
By contrast, Time Warner Cable / Spectrum was famously slow. Charter (which acquired both) is now better, but the legacy footprint is still mixed. Cox is also mixed. Smaller regional ISPs are mostly IPv4-only because there's no business case for them to invest in the dual-stack upgrade.
The CGNAT problem you don't see
Carrier-grade NAT is the alternative to IPv6 that no one talks about, partly because it works fine for most browsing. You can load any website, watch video, do video calls. What it breaks: anything where someone else needs to find you. Hosting a game server. Running a self-hosted nextcloud at home. Inbound SSH. WebRTC voice calls between two CGNAT-ed parties. Port-forwarding for a security camera. Bittorrent seeding.
If you're behind CGNAT and don't know it, the symptoms feel random — some apps just don't work right, some games show high latency, some services say you "appear to be on a VPN." You can check if you're on CGNAT by looking at your public IP and comparing to what your router shows. If they differ and your "public" IP is in the 100.64.0.0/10 range, that's CGNAT's reserved space; you're behind one. Our homepage shows your public IPv4 and your router will show the local one — compare them.
IPv6 sidesteps this entirely. Every device gets its own globally routable address. The internet's original peer-to-peer architecture, the one we lost to NAT in the 1990s, comes back.
What's still standing in the way
Three things, in roughly this order:
1. The cost of changing what's working. An ISP with five million customers on stable IPv4 has no immediate revenue case for spending engineering quarters on IPv6 rollout. The customer wouldn't pay extra. The CFO calls it a maintenance cost.
2. Enterprise IT. Corporate networks, especially in regulated industries, are full of equipment that "supports IPv6" but was never actually tested in production. Banks, hospitals, government agencies tend to stay on IPv4 for the same reason they stay on COBOL: the working system is more valuable than the modern one. A surprising amount of enterprise software still has hardcoded IPv4 assumptions (logging, monitoring, ACLs) that surface only when you flip the switch.
3. The training and tooling gap. Most ops engineers under 35 have never debugged an IPv6 routing problem. The error modes are different, the addresses are unmemorable, traceroute6 outputs look weird. Most monitoring dashboards still default to IPv4 columns. The skill gap is closing — slowly — but it's real.
What this means for you
If you're a normal user, IPv6 is increasingly working without you noticing. Test if you have it: check on our IPv6 page. If you have it, your computer has just quietly been using it for many of the websites you load every day. If you don't, that's almost certainly because your home ISP doesn't offer it (your computer, router, and most of the internet all support it).
If you're a developer, the practical implication is: stop assuming IPv4. Every new service you build should accept connections on both stacks. Log the source IP including IPv6. Make sure your firewall rules cover both. If you have a "block all from China" rule, double it for IPv6 ranges (it's a different RIR allocation). If you have monitoring that breaks on long IPv6 addresses, fix it now — the addresses aren't going to get shorter.
If you're an ISP, you already know whether you're behind. The forcing function will probably arrive in the form of regulator pressure (many countries are now writing IPv6 minimums into broadband subsidies) and customer churn to ISPs that did the work. The runway between now and "your competitor markets IPv6 as a feature" is shorter than it used to be.
The slow protocol that won
IPv6 hasn't replaced IPv4. It probably won't for another decade. What it's done is grown into a parallel internet that now carries roughly half of all traffic. The decade of pundits saying "IPv6 is dead, no one's adopting it" has been quietly wrong all along — it's been growing at 5-10% adoption per year, every year, for fifteen years. That doesn't make a headline. It just makes a fact.
The 100% IPv6 internet is still a long way off. The 50% IPv6 internet will probably be here in 2027. The half of users without IPv6 today are the half whose ISP didn't get around to flipping the switch — and the pressure on that ISP to do so grows every year.