Google’s IPv6 Milestone Exposes a Split Internet

Google’s data hit a threshold many had waited years to see. On March 28, 2026, native IPv6 accounted for 50.10% of users reaching its services. The figure, pulled from the company’s long-running measurement graph, marked the first time the protocol crossed the halfway mark.

Why the Numbers Don’t Line Up

But look closer and the picture fragments. APNIC Labs pegged global IPv6 capability at 42% around the same period. Cloudflare reported 40.1% of HTTP requests over IPv6. The Register noted the gap in an April 17, 2026 article. (The Register)

These differences stem from methodology. Google tracks actual connections to its own vast properties. APNIC weights samples by estimated internet population per economy using advertising-driven tests. The two data sets bracket reality. One shows momentum. The other tempers enthusiasm. And as of mid-June 2026, Google’s own page listed 48.68%. (Google IPv6 Statistics)

The Internet Society called the March crossing “a significant milestone in the very long road to more widespread IPv6 deployment.” Mat Ford wrote the analysis for the group on April 21. He pointed to 18 years of Google data. Native IPv6 only pulled ahead of transition technologies after 2010. (Internet Society Pulse)

Adoption never spread evenly. France sits near 73%. India reaches 72%. Saudi Arabia hits 65%. Yet Italy lingers at 17%, Spain at 10%, and Egypt at 4%. Most of Africa and Central Asia trail further behind. The global average from multiple sources still hovers around 43% according to the Internet Society.

George Michaelson captured the nuance in the APNIC Blog post that crystallized the moment. “Reaching the 50% mark is a significant milestone, demonstrating that IPv6 is a mature, fully capable protocol that is being deployed at a global scale and used effectively in real-world networks.” He authored the April 28 piece. (APNIC Blog)

Michaelson stressed measurement mechanics. APNIC’s approach applies statistical weighting drawn from World Bank population estimates. Raw ad impressions vary daily by region. Large populations in India, China, and Indonesia therefore carry more influence in the final number. This produces a lower global figure than Google’s unweighted view of its own traffic. The two measurements, taken together, offer bounds on the true state.

China adds another layer. Google statistics largely miss it. Commenters on the APNIC post and Hacker News threads estimate that including China’s rapid progress could push effective global usage to 65% or higher. One analysis suggested the real user count already exceeds three billion on IPv6.

Mobile operators drove much of the early surge. Reliance Jio in India built its network IPv6-first. Major U.S. carriers followed with heavy IPv6 ratios. T-Mobile shows 88% in some reports. AT&T and Verizon sit in the mid-70s. These networks rely on 464XLAT to give customers IPv4 where needed while routing internally over IPv6. The economics proved decisive. IPv4 addresses now carry explicit costs at cloud providers. AWS charges roughly $43 per public IPv4 address annually. That price signal accelerated decisions that protocol purity arguments never could.

Content providers and CDNs amplified the effect. Google, Meta, Akamai, and Cloudflare deliver enormous traffic volumes. Their caches sit close to users. A single cache hit serves thousands of downstream requests. This means measured IPv6 at internet exchanges understates actual edge delivery. As more ISPs enable IPv6, the multiplier grows.

Yet enterprise networks lag. Many VPC configurations still default to IPv4-only. Older Helm charts and Terraform modules ignore dual-stack options. Corporate IT often treats IPv6 as optional. The consumer side crossed 50%. The business side has not. That split defines the current internet. One half runs efficiently on abundant addresses. The other clings to NAT layers, address sharing, and workarounds.

The protocol itself no longer needs justification. IPv6 supplies 340 undecillion addresses. IPv4 offered 4.3 billion. The math was clear decades ago. What changed is operational maturity. TCP, UDP, and QUIC function across both versions. Higher-layer applications rarely notice the difference. Dual-stack services handle the rest.

Some services still omit IPv6. Certain code repositories and national broadcasters cite operational overhead or regulatory demands around geolocation and content filtering. These gaps frustrate users on IPv6-only connections. But they reflect pragmatic limits more than outright rejection.

Critics once called the slow pace failure. Michaelson rejects that view outright. Deployment demanded real capital and engineering effort. Incumbent providers maximized returns on existing IPv4 assets. Newer entrants chose IPv6 to cut long-term costs. The internet evolved through market choices, not central decree. Progress therefore varies by region, sector, and business model.

Today the network runs on two protocols. NAT complexity on the IPv4 side rivals translation or encapsulation on the IPv6 side. Neither offers obvious simplicity. Interoperability happens at the transport layer. Intermediaries fill the gaps. The arrangement works. It just isn’t elegant.

Recent discussions on X echo the milestone. Engineers note that consumer migration looks complete while enterprise defaults remain stuck in 2019. Others highlight weekend spikes. Google’s graph often climbs on Saturdays when mobile and residential traffic dominate. The June 14 reading of 48.68% reflects a weekday view.

Linear growth models from 2018 onward suggest continued gains. Both Google and APNIC trends fit roughly 5% per year in recent periods. If the slope holds, majority status across all major measurement platforms could arrive before 2030. But entrenched systems in some economies may flatten the curve.

The milestone matters. Half the users visible to one of the internet’s largest destinations now prefer IPv6. The protocol runs on fixed lines and mobiles, in developed markets and emerging ones, for apps and massive data-center flows. It has moved from experiment to daily infrastructure.

Recognition of that fact does not imply completion. Plenty of work remains on enterprise adoption, service coverage, and measurement consistency. Still, the data shows a clear direction. IPv6 no longer waits in the wings. It shares the stage. And on some days, it commands more than half the audience.