On March 28, 2026, Google’s IPv6 adoption tracker ticked past 50.10%. For the first time since the search giant began measuring in 2008, more than half of its global users reached its services over Internet Protocol version 6. The Internet Society called it a milestone 18 years in the making. APNIC ran a blog post headlined, simply, “Google hits 50% IPv6.” The Hacker News thread filled with the usual mix of congratulations and war stories.
Here is what nobody said: 50% is not the finish line. It is the point at which we should admit the race will never end.
IPv6 was designed to succeed IPv4 — to replace it, not to coexist with it indefinitely. The original vision, stretching back to the 1990s, was a clean migration. Every device gets a globally routable address. NAT boxes go the way of token ring. The internet becomes end-to-end again. That vision has now been formally, statistically abandoned. Half the world’s Google users are on IPv6. The other half are not. And if 18 years of deployment effort could not push the number past 50%, there is no reason to believe the next 18 will finish the job.
The aggregate number hides the real story, which is a bifurcation so sharp it amounts to two internets.
The Geography of the Missing Half
Google does not publish per-region breakdowns on its public dashboard, but APNIC’s own measurements tell the story. India runs above 70% IPv6 adoption. The United States hovers around 55%. Germany, France, and Japan all clear 60%. Then there is the rest of the world: large stretches of Africa, the Middle East, and Southeast Asia where IPv6 adoption sits below 5%, sometimes below 1%. These are not temporary lags. They are structural. The networks in those regions are not waiting for a software update; they are operating on business models and infrastructure cycles that make IPv6 deployment a cost with no revenue upside.
A mobile carrier in a price-sensitive market does not deploy IPv6 because it is elegant. It deploys IPv6 because it has exhausted its IPv4 allocations and has no choice. If it can still buy IPv4 addresses on the secondary market — where a /24 block trades for roughly $50 per address, according to recent brokerage data — or if it can continue stretching its existing pool with carrier-grade NAT, it will. The economic case for IPv6 in the developing world is, and has always been, that IPv4 scarcity will eventually force the issue. But scarcity has not bitten hard enough, and CGNAT has turned out to be a surprisingly tolerable workaround for the applications that dominate these markets: messaging, social media, mobile payments.
“We did the math on a network upgrade in one West African market,” one telecom engineer told me in a hotel bar at a regional connectivity conference. “The IPv6 migration would have cost us about $4 million in core and RAN reconfiguration. Or we could buy another /22 of v4 space for about $120,000 and kick the can down the road another three years. The CFO didn’t even let us finish the slide deck.”
The result is not a temporary gap. It is a permanent division. The internet’s wealthy half runs dual-stack, leaning increasingly toward IPv6-native. The internet’s poorer half runs IPv4-only, behind layers of NAT that make it functionally invisible to the rest of the network. And the gap is self-reinforcing: content providers optimize for where the traffic is, which means they optimize for the IPv6-rich markets. The IPv4-only markets become, over time, second-class citizens on a network that was supposed to be universal.
The Costs Nobody Is Accounting For
A permanently bifurcated protocol landscape imposes real costs, and they are not the ones the IETF working groups spend their time on. Every major content delivery network — Cloudflare, Akamai, Fastly — maintains dual-protocol stacks. Every cloud provider runs translation layers. Every enterprise that wants to reach the full global internet has to support both protocols in its edge infrastructure. This is not free. It is a tax on every new service, every new application, every new startup that cannot afford to target only the IPv6-majority half of the world.
The more insidious cost is in the asymmetry of reach. An IPv6-only user can reach IPv4 content through translation mechanisms like NAT64. An IPv4-only user cannot reach IPv6-only content at all — not without a tunnel, not without a proxy, not without someone on the other side deliberately degrading their service to accommodate the legacy protocol. As more services go IPv6-native — and the trend is accelerating among cloud-native startups that see no reason to provision IPv4 addresses they do not need — the IPv4-only half of the world becomes a walled garden in reverse. It is not that they are locked in. It is that they are locked out of the parts of the internet that have moved on.
This is not a technical problem with a technical solution. It is a policy problem that no policy body is treating as urgent. The ITU talks about digital inclusion in terms of fiber and spectrum. The IGF talks about it in terms of content moderation and platform governance. Nobody is holding hearings on what happens when a billion people are stuck on a protocol that the rest of the internet is quietly deprecating.
What a Real Plan Would Look Like
The standard response from the IPv6 advocacy community is that we just need more deployment — more incentives, more mandates, more government procurement requirements that specify IPv6 support. This is, by now, a form of magical thinking. The incentives have been in place since the US government’s IPv6 mandate in 2005. The IETF declared IPv4 a “legacy protocol” in 2016. The market has had all the information it needs for two decades, and it has delivered 50%.
A real plan would start from the premise that the split is permanent and design policy around that fact. It would mean requiring that every government digital service be reachable over both protocols, not just IPv6. It would mean funding translation infrastructure — NAT64 gateways, DNS64 resolvers — as public goods in the regions where private ISPs have no incentive to deploy them. It would mean treating IPv4-to-IPv6 reachability as an access problem, the way we treat last-mile connectivity, rather than assuming migration will eventually solve it.
None of this is glamorous. None of it fits on a conference slide with an upward-sloping green line. But the alternative is a two-tier internet where half the world’s users can reach the other half’s services only by grace of a protocol that was supposed to have been retired before most of them were born.
The 50% mark was not a celebration. It was the moment the two-stack internet became a permanent feature of the landscape. The question now is whether anyone in a position to do anything about it is willing to say so.