Waiting for IPv6

Prepared by Claude Code using Fable 5.1
Published: 2026 October 12
Revised: 2026 October 12
Cite as: ftorrent. Waiting for IPv6. October 2026. https://docs.ftorrent.com/waiting-for-ipv6.html

You were at a conference in the middle 2000s. One session was about IPv6: the internet was running out of addresses, the new protocol had enough for every grain of sand, the changeover was coming, and the government had set a date. At drinks afterward someone said a single American university held more addresses than all of China. Then you never heard about it again. The question this article answers is the one that memory leaves behind: the rise of IPv6, what happened with that? Why isn't it at ninety-nine percent right now?

Two things are true. Google's IPv6 chart, the one everyone cites, has climbed from a fraction of a percent in 2008 to about half of Google's users in 2026, and that is a real change in how the internet is built. And the transition, as it was announced, did not happen: the old protocol never went away, no deadline was met in the sense its authors meant, and the engineers who predicted when the addresses would run out, correctly, to the month, were wrong about what running out would force. What follows is the record, with the sources beside it, told in order: the addresses that were handed out, the forecast, the protocol, the promise, the deadlines, the switch-on days, the reasons, where it stands, and what we expect the next ten years to look like.

1. The university and the country

Start with the anecdote, because it is mostly true, and the way it is true explains the rest. An IPv4 address is 32 bits, which gives about 4.3 billion of them. Before 1993 they were handed out in three sizes only: a Class A block of 16.8 million addresses, a Class B of 65,536, or a Class C of 256. The 1990 edition of the internet's address registry, RFC 1166, lists who had a Class A block: IBM, Xerox, Hewlett-Packard, Digital, Ford, Apple, AT&T, the US military under a dozen names, and the universities that were building the network, among them MIT as MIT-TEMP and Stanford as SU-NET-TEMP. Even the registry marked those two as provisional.

So in the 1990s the claim was true. China's holdings passed one Class A block, by APNIC's own account, "sometime in 2000 or 2001," and the registry's allocation file, summed by country, agrees: about 14 million addresses at the end of 1999, 20 million a year later. Until then, MIT, with one block of 16.8 million, had more than China. Stanford did too, until it gave its block back. The university's networking group announced in 1999 that "because of IP address scarcity, Stanford is giving back the 36 net in the year 2000," and renumbered every building into four Class B blocks on a published schedule. The 36 block is registered today to China Telecom's Fujian branch.

The anecdote outlived its facts. A CNET story in March 2004, reporting from a United Nations forum, said China "has been allocated only 9 million global Internet addresses, less than Stanford University's total of 17 million." By then China held about 49 million and Stanford about 330,000; the figures were wrong on both ends. Microsoft's IPv6 team wrote a correction in 2007 that called it "a factoid that became untrue with age." The India version held longer. India reached its first Class A equivalent only in 2007 or 2008, and at the end of 2025 holds about 41.7 million IPv4 addresses, two and a half blocks: fewer than Apple or Ford, each of which still holds a whole block, and a fifth of what the US Department of Defense holds in thirteen. MIT kept its block until 2017, when it sold eight million of its fourteen million unused addresses to Amazon.

None of this was a scandal. The blocks went to the institutions that were building the network, in the only sizes the scheme allowed, when four billion looked like more than anyone would use. What the anecdote records is that the address space was spent before most of the world arrived, and that the arriving world would have to share, buy, or build something new. India is the clearest case. In September 2016 a new carrier, Reliance Jio, launched a mobile network and signed up 200 million subscribers in about nine months, five times the country's entire IPv4 holding, and ran them on IPv6. The anecdote and its resolution are the same arithmetic.

2. The forecast

The people who built the protocol saw the end of the address space coming early and said so in writing. In August 1990, at an IETF meeting in Vancouver, three engineers projected that the Class B blocks would be exhausted by March 1994. A workshop of the internet's architecture board the following January produced RFC 1287, which expected the internet to "run out of the 32-bit IP address space altogether" and argued against small fixes: "We should embark now on a plan that will take us past the exhaustion of the address space," one that "should take us all the way to replacing the current 32-bit global address space."

Two stopgaps came first, and both described themselves as stopgaps. Classless routing, RFC 1519 in September 1993, let blocks be any size and so stretched what was left. Its authors wrote:

This plan should be viable for at least three (3) years, after which time, deployment of a suitable long term solution is expected to occur.

RFC 1519, Classless Inter-Domain Routing, September 1993

Address translation, RFC 1631 in May 1994, let many machines share one address. It called itself "another short-term solution" offering "temporarily relief while other, more complex and far-reaching solutions are worked out," and ended with a warning:

NAT has several negative characteristics that make it inappropriate as a long term solution, and may make it inappropriate even as a short term solution.

RFC 1631, The IP Network Address Translator, May 1994

When that document was rewritten in 2001 as RFC 3022, the words "short-term" and "temporary" were gone. NAT had become an ordinary mechanism.

The forecast of the end itself came from an IETF working group on address lifetimes, which met through 1993 and 1994 and projected that the internet would exhaust IPv4 "between 2005 and 2011." The recommendation that chose IPv6 recorded the projection and the reception: "It has been criticized as both too optimistic and as too pessimistic." Geoff Huston, who would spend the next thirty years measuring this, restated it that December in RFC 1744 as six to twelve years, and added a prediction of his own: "a free trading environment in address blocks is a probable outcome, where address blocks would be bought and sold between trading entities." The free pool ran out on 3 February 2011, inside the window. The market he predicted opened a month later. Hold on to that: the forecast was right.

3. The protocol

Vint Cerf has told the origin of the 32-bit address many times, and the version with a transcript is from a US government workshop in September 2010:

So somewhere around 1977, I said, okay, enough, I'm trying to get this thing underway. We haven't -- you know, we have to decide. I said how about 32 bits? It's 4.3 billion addresses. It's enough to do an experiment. We didn't know whether this was going to work. This was an experiment, folks. The problem is that the experiment never ended and so here we are. It's 2010 and we're now into the production version of the Internet.

Vint Cerf, NTIA IPv6 workshop, Washington, 28 September 2010

He also recalled that a faction in 1977 had wanted 128 bits, and that "most of the other engineers said you're crazy, why do we need 340 trillion, trillion, trillion addresses? It doesn't pass the red-face test." That is the size IPv6 chose.

The choosing took three years. By 1992 the IETF had four proposals for a next-generation IP; by the end of that year, seven; by 1994 they had merged and fallen away to three, and the IPng area directors, Scott Bradner and Allison Mankin, ran an open contest with written criteria. The criteria document, RFC 1726, is worth reading for what it already knew. Criterion 5.5: "The protocol must have a straightforward transition plan from the current IPv4." And in the discussion: "If we can't transition to the new protocol, then no matter how wonderful it is, we'll never get to it." It also saw the incentive problem: "new features provide a motivation to deploy IPng more quickly."

The recommendation, RFC 1752 in January 1995, picked the 128-bit variant of a proposal called SIPP, described it as "an evolutionary, rather than revolutionary, step from IPv4," and set four goals for the transition: any host or router could be upgraded at any time without depending on any other; new IPv6 systems could be installed with no prerequisites; upgraded systems could keep their existing addresses; start-up costs would be low. It also recorded one thing left undone. "The biggest problem the reviewers had with SIPP was with IPAE, SIPP's transition plan. The overwhelming feeling was that IPAE is fatally flawed." The chosen protocol shipped without its transition plan: "transition and address autoconfiguration do remain to be finalized." A companion paper by Brian Carpenter, RFC 1671, gave the only timescale anyone in the IETF put in writing: "the next decade (or more)," and "during the many years of transition the Internet will contain black holes."

The specification itself, RFC 1883, is dated December 1995. The IETF revised it in 1998 and, in July 2017, twenty-two years after the first edition, declared it an Internet Standard as RFC 8200. No IETF document ever gave a date for the transition to complete. The dates came from elsewhere.

4. The promise

For the next thirteen years the new protocol was sold, by people who believed in it, as the next thing. The quotations below are exact, each with the page it can be read on, and each sounded reasonable when it was said. Google's measurement, which begins in September 2008, puts a number under the last of them.

In April 1997, Steve Deering, IPv6's co-author, gave an invited talk at a USENIX conference; the published summary records his expectation of "wide availability by 1998-1999." In August 1999 Esther Dyson, then interim chair of ICANN, took a reader's question in an online chat about the address shortage:

There's going to be a whole new addressing system called IPv6 (vs. the current IPv4) that will solve that problem in spades.

Esther Dyson, ABC News chat from the ICANN meeting in Santiago, 26 August 1999

That same month Microsoft told a trade paper that "due to the experimental nature of IPv6, Microsoft will not support it in Windows 2000," as The Register reported when the company reversed itself with a preview release the following spring. Japan's cabinet put IPv6 into its national technology strategy in January 2001, and the implementing program that March promised that "an IPv6-based Internet environment will be realized by CY2005." In July 2003 Christian Huitema, by then Microsoft's networking architect and the author of the first book on the protocol, told a conference that "the Dream Network of the future requires IPv6," because "NAT is evil."

The selling had an organization. In the spring of 1999 Latif Ladid, then at a Danish router company, began recruiting founding members for an IPv6 Forum, registered it in Luxembourg that July, held the first Global IPv6 Summit in Paris that October, and gave it a mission: "to promote IPv6 by dramatically improving the market and user awareness of IPv6, creating a quality and secure Next Generation Internet and allowing world-wide equitable access to knowledge and technology, embracing a moral responsibility to the world." Its home page in March 2000 carried a press link titled "Get Rid of Pesky NATs." By 2004 it listed thirty-one national task forces and ran an "IPv6 Ready" logo program for equipment. The vendors wrote the copy. Nokia's IPv6 pages in January 2001 are the earliest use this research found of the comparison everyone would repeat: "enough capacity for an estimated 340 undecillion IP addresses, one for every grain of sand on the planet." Sun's page in 2000 was headed "Renewing the Net" and promised that "there's no limit to the number of people you can reach -- and who can reach you -- through the Net." Microsoft's, in 2003: "It is the future of networking." Toshiba showed a refrigerator with its own IPv6 address in June 2001, and Japan's government estimated in 2004 that businesses built on the protocol would be worth 84 trillion yen by 2010.

The press supplied the deadline. In May 2001 a Reuters story carried by USA Today reported that "the Web will run out of IP addresses by the year 2005 if a new standard is not adopted soon." In October 2003 the BBC ran a feature on "what is going to happen when the number of net addresses runs out sometime in 2005," in which a magazine lab director offered, "We could assign IP addresses to every grain of sand in the Sahara and still have extra." The people who ran the registries objected. Andrew McLaughlin wrote on CircleID that the 2005 claim was "patently absurd," and APNIC published a brochure on the myth of address shortage that named the BBC story as its typical example. Both sides were partly right. The addresses did not run out in 2005. They ran out in 2011, and by then the public had heard the warning at so many wrong dates that the true one landed as one more.

The governments followed. In June 2003 the Pentagon's chief information officer, John Stenbit, issued a memorandum: "The DoD goal is to complete the transition to IPv6 for all inter and intra networking across the DoD by FY 2008." The date was chosen, he told reporters, because experts expected commercial adoption between 2005 and 2007, so the department would "pull the switch in 2008." In August 2005 the White House budget office extended the order to every agency in memorandum M-05-22: "we have set June 2008 as the date by which all agencies' infrastructure (network backbones) must be using IPv6."

In September 2005 Cisco's Tony Hain published a forecast that the remaining pool "will be exhausted in about 5 years," and advised that "perhaps in the 2008–2010 timeframe, we should plan to turn on IPv6 networking capabilities throughout our networks." Huston answered in the same issue with 2014. Three years later, in September 2008, he revised it: the pool would "reach the point of exhaustion in February 2011." It did, on 3 February.

The European Commission's version is the fullest statement of the promise, and the one with the grains of sand. On 27 May 2008 Viviane Reding, the commissioner for the information society, announced a target of 25% of European users on IPv6 by 2010:

This is very much a case of a stitch in time saves nine. In the short term, businesses and public authorities might be tempted to try to squeeze their needs into the strait jacket of the old system, but this would mean Europe is badly placed to take advantage of the latest internet technology, and could face a crisis when the old system runs out of addresses. IPv6 provides more addresses in cyberspace than there are grains of sand on the world's beaches. If Europeans are to use the latest internet devices such as smart tags in shops, factories and airports, intelligent heating and lighting systems that save energy, and in-car networks and navigation systems, then we already face a thousand-fold increase in demand for IP addresses. I call on Member States to make sure that public authorities and industry have IPv6 widely sown up by 2010.

Viviane Reding, European Commission press release, 27 May 2008

Two months earlier the comparison that outdid the sand had appeared on an electronics-industry blog, where Steve Leibson passed along a note from a fellow museum docent, Dick Guertin: "we could assign an IPV6 address to EVERY ATOM ON THE SURFACE OF THE EARTH, and still have enough addresses left to do another 100+ earths." The note's arithmetic assumed a reserved first byte that IPv6 does not have and so computed two to the 120th power rather than the 128th; the claim is true anyway, by a wider margin. Both comparisons count addresses rather than networks, and since IPv6 hands every network a block of two to the 64th, the practical space is smaller than either suggests, which network operators pointed out at the time.

Then the measuring began. Google's chart starts on 4 September 2008 at 0.14% of its users. In October 2008 Lorenzo Colitti of Google told a RIPE meeting why the company would not yet publish an IPv6 address for its main site: "0.2 percent of users have IPv6 and will use IPv6 if you publish a AAAA," and "0.09 percent of users, that's one in 10,000 have broken IPv6 connectivity. We don't want to break that user in 10,000." Five months after Europe set its target at a quarter of users, one user in five hundred could reach a site over IPv6, and one in ten thousand would be cut off by the attempt.

5. The deadlines

Here is what the dates became. The right-hand column is what a measurement found when the deadline arrived, or the nearest one available.

Table 1. IPv6 deadlines set by governments, and what was measured when they arrived. Sources are linked in the text and listed in the references.

Who	Set	Required	By	What was measured
Japan, e-Japan program	March 2001	"an IPv6-based Internet environment"	2005	No measurement exists for 2005. Google measured Japan at 1.6% of users in June 2012.
US Department of Defense	June 2003	"complete the transition to IPv6 for all inter and intra networking"	FY 2008	A demonstration in 2008, after which, the inspector general found, the department turned IPv6 off; a 2020 audit counts three separate attempts, two abandoned.
US Office of Management and Budget	August 2005	agency backbones "must be using IPv6"	30 June 2008	Every agency reported a "successful demonstration" of passing IPv6 traffic. Google's world figure two months later: 0.14%.
Malaysia	November 2006	"an IPv6-enabled nation"	2010	Google measured Malaysia at 0.05% in June 2012.
European Commission	May 2008	25% of users	2010	In May 2010 the Commission's own official told a RIPE meeting, "I don't think we will reach this target." Google's world figure at the end of 2010: 0.23%. Germany 0.2% and the UK 0.1% in June 2012.
India, Department of Telecommunications	July 2010	all major providers to offer IPv6 service	December 2011	The department's own status table in February 2012 reads "final report still awaited" for most providers. Google measured India at 0.04% in June 2012.
US Office of Management and Budget	September 2010	public services on native IPv6; internal networks by FY 2014	30 September 2012	16% of federal domains enabled at the deadline; 55% of public domains by 2015; no figure for internal networks.
South Korea	March 2014	commercial service in 2014; half of traffic by 2018	2018	Google measured South Korea at 3.6% in December 2018, and 18.9% in October 2026.
China, State Council	November 2017	20% of users by 2018, 50% by 2020, the transition complete by 2025	2025	Independent measurement put China at 3% to 7% at the end of 2018 and 21% at the end of 2020. For 2025 three figures exist: China's own count of 865 million users, 77%; APNIC's sample, about 50%; Google's, 7%, from the few users who reach Google from inside the country.
US Office of Management and Budget	November 2020	80% of federal networked assets running IPv6 only	30 September 2025	No government-wide result has been published.

The pattern repeats across four continents and twenty years: a date set when adoption was under one percent, a deadline that passed with adoption still under one percent, and a new date. The United States alone did it four times. The 2003 order said 2008; the 2005 order said June 2008; the 2010 order said 2012 and 2014; the 2020 order said 2023, 2024, and 2025. Each was met in a narrow sense and missed in the plain one. The 2008 deadline was satisfied by agencies demonstrating that an IPv6 packet could cross their backbone, and the Pentagon's own auditors later recorded that the department demonstrated it in 2008 "but then turned it off."

The dates also had something in common besides passing. None of the documents in Table 1 appropriated money for the work. They are memoranda, action plans, and roadmaps: the American ones direct agencies and set milestones; the European one "calls upon" content providers and "encourages" internet providers; the Indian one sets targets for private carriers; the Malaysian minister "directed." An audit of the American agencies in 2006 found their own cost estimates ranging from under a million dollars to over twenty million each, with no line in any budget to pay it. A deadline without a budget is a hope with a date on it, and in every case the people it addressed treated it as one.

The documents also show the authors learning. The 2010 memorandum carried a footnote: "it is expected that agencies will also continue running IPv4 into the foreseeable future." The 2020 one, M-21-07, reversed it: "OMB previously issued policy discussing the expectation for agencies to run dual stack (IPv4 and IPv6) into the foreseeable future; however, in recent years it has become clear that this approach is overly complex to maintain and unnecessary." Its strategic intent is to run "using only IPv6." Five years on, no agency-wide figure says how far that got.

6. The day the pool ran out, and the two switch-on days

On 3 February 2011, at a ceremony in Miami, the central registry handed its last five blocks to the five regional registries. The press release from the registries' coordinating body said:

Deploying IPv6 is now a requirement, not an option.

Raúl Echeberría, chairman of the Number Resource Organization, press release, 3 February 2011

The Asia-Pacific registry reached its last block in April 2011, Europe's in September 2012, Latin America's in June 2014, and North America's ran dry on 24 September 2015. Each announcement used the imperative: "mandatory," "imperative," "as soon as possible." Each was addressed to an internet where, by Google's count, fewer than one user in a hundred could use the new protocol. The address market Huston had predicted in 1994 opened the next month, when Microsoft paid 7.5 million dollars for 666,624 addresses from the bankrupt Nortel, 11.25 dollars each. That January, at a conference in Brisbane, Huston had put his own position in one sentence: "Vint Cerf is a professional optimist. I am not."

The content side had a reason for waiting, and it was the one Colitti had named in 2008: a small fraction of users whose IPv6 was misconfigured would be cut off from any site that published an IPv6 address. The Internet Society's answer was a rehearsal, which it called a test flight. On World IPv6 Day, 8 June 2011, from midnight to midnight UTC, about four hundred organizations, Google, Facebook, Yahoo, and Akamai among them, published IPv6 addresses for their main sites alongside the IPv4 ones and measured what broke. The organizers had estimated that 0.05% of users, one in two thousand, would find the participating sites slow or unreachable. Google's chart for the day reads 0.34%. Afterward Google reported that it "carried about 65% more IPv6 traffic than usual, saw no significant issues and did not have to disable IPv6 access for any networks or services." Facebook saw "over 1 million users reach us over IPv6" and no rise in requests for help. Yahoo's breakage rate stayed at 0.022% of users, and the IETF's chair, Russ Housley, concluded "we've pretty much fixed the IPv6 brokenness problem." A company that monitored six carriers' traffic put the other side: "IPv6 traffic roughly doubled during the v6-day period. However, doubling a fraction of a percent, is still a fraction of a percent," and at midnight "the web traffic falls off the cliff." An IETF measurement published later as RFC 6948 counted the top ten thousand websites with an IPv6 address: 245 before the day, 491 at its peak, about 310 a month after. It added, "The event was largely unnoticed by the general public." The headlines agreed. The Register: "World IPv6 Day fails to kill the internet. Publicity stunt over, now the work begins." eWeek: "World IPv6 Day Ends, Everyone Goes Back to IPv4." Wired had listed the date in print as the "Nerdiest Holiday Ever." The American registry's John Curran called it "a non-event, which is what we were hoping for," and Comcast's John Brzozowski gave the operator's version at a conference that month:

World IPv6 day was pretty much a non-event, from a brokenness perspective. It was a wasted 37 hours of not sleeping, for me.

John Brzozowski, Comcast, Multichannel News, June 2011

A year later came the launch, and the organizers were careful about the word. World IPv6 Launch, 6 June 2012, was announced in January under the heading "This Time It is For Real," and its commitments were three: websites would leave IPv6 on permanently; home-router makers would ship it enabled by default; and internet providers would enable it for enough customers that at least one percent of their residential subscribers reached the participating sites over IPv6. The organizers' own FAQ anticipated the obvious question: "Why only 1%? In many cases, users may need to upgrade or replace hardware and software, such as operating systems or home routers, to use IPv6. Over time, as users upgrade, IPv6 adoption will increase without any changes in the ISP's service or equipment." Four router makers made the default-on commitment. About three thousand websites took part, the four most visited among them. "IPv6 is no longer a lab experiment," said the Internet Society's Leslie Daigle, "it's here," and in May, "it's not just for a day — it's forever." Google's Erik Kline wrote the sentence that best catches the moment: "The original IPv6 specification was published more than 15 years ago, but for the entire career of most Internet engineers its deployment has always been in the future. Now it's finally here." Cerf wrote the launch post on Google's blog:

Today we launch the 21st century Internet: you ain't seen nothin' yet.

Vint Cerf, Google's World IPv6 Launch post, 5 June 2012

Google's chart for launch day reads 0.65%. At the Amsterdam exchange, where much of Europe's traffic changes hands, IPv6 rose that day to 0.4% of the total. Akamai, looking back ten years later, counted 3.9 billion IPv6 requests on launch day from about 0.6% of its clients, against more than four trillion a day in 2022. The chart crossed one percent that December, ten percent at the end of 2015, and twenty percent in 2017, and the Internet Society marked each anniversary with the figure: twelve percent in 2016, over a quarter in 2018. The Register's headline for the sixth anniversary was "Worst. Birthday. Ever.", and Cerf's video message that day said, "it's time to get with the program." The launch site was archived and stopped updating. Huston had set a bar in June 2012, in an interview: "If 20 percent of the users in America use IPv6, the future of IPv6 is assured." By his own test the future was assured within five years, and the transition still did not end.

7. Why

Four reasons, each with people who said it at the time, and a verdict at the end. They are not alternatives. Together they describe one mechanism.

7.1 The new protocol could not talk to the old one

An IPv6-only computer cannot send a packet to an IPv4 address, and an IPv4-only computer cannot send one to an IPv6 address. The designers chose this. The sharpest statement of what it cost is Dan Bernstein's, written around 2002:

The IPv6 designers made a fundamental conceptual mistake: they designed the IPv6 address space as an alternative to the IPv4 address space, rather than an extension to the IPv4 address space.

D. J. Bernstein, "The IPv6 mess," about 2002

The consequence, in his words: "Right now, public IPv6 addresses are practically useless," and making them useful meant convincing "millions of computer administrators to do something with no immediate benefit." Randy Bush, an operator who had chaired the IETF's transition working group, put the same point on a slide at an APNIC meeting in 2007: "IPv6 is on-the-wire incompatible with IPv4," which "could have been avoided, e.g. if IPv6 had variable length addressing, IPv4 could have become the 32 bit variant." His slide titled "Why Is This Happening?" listed: "No transition plan. Declared victory before the hard part started. No real long term plan. No realistic estimation of costs. No support for the folk on the front lines. Victory will be next month." In 2009 the Internet Society's Daigle conceded it on a panel: "The lack of real backwards compatibility for IPv4 was the single critical failure," adding, "There were reasons at the time for doing that." Bradner, who had co-run the selection, named his regret in 2012: "did not do variable length addresses in IPv6 & punted on dealing with routing during the IPng process." Radia Perlman went further: the format proposed by the ISO in 1992, with twenty-byte addresses, had been technically better, and "the only reason it was not adopted is because standardization agencies compete with one another."

The record is fairer to the designers than the critics allow. They knew. Their criteria demanded a transition plan; their recommendation recorded that the one on offer was judged unworkable; the first transition standard in 1996 opened with "the key to a successful IPv6 transition is compatibility with the large installed base." Their plan was dual stack: every upgraded machine would speak both, and the old protocol would be retired when nothing needed it. That plan honors every goal they set for the machine being upgraded, which keeps its address and depends on no one. It does nothing for the machine that has no IPv4 address at all, which is the only machine the transition was for. Bernstein's alternative, embedding the old addresses in the new space so that an old server needs no change, was rejected rather than disproved; no one tested it at scale. What is certain is the shape of the incentive the shipped design created. The first mover paid and gained nothing. The last mover paid nothing and lost nothing.

7.2 It was nobody's job

Huston wrote the economics down in 2009: the two protocols "are not perfect substitutes," the cost of running both "would need to be borne by the service provider," the benefit "lies in no individual consumer's sphere of immediate self interest," and so "we may be facing a situation of a market failure with this transition." Daigle's version the same year was shorter: "nobody wants to go to IPv6 unless they think they're friends are doing it, too." Lee Howard, then at a cable company, priced the choice in 2013: about 30 dollars per new subscriber per year to put them behind carrier-grade NAT, against about 7.50 to run dual stack. IPv6 was already the cheaper option for a provider, and providers still waited, because, as Huston said in 2012, "the customers are not going to fund this with incremental payments," and "the last-mile access providers want to wait until their competitors deploy IPv6." Howard's 2019 talk title gave the structure its name: "the prisoner's dilemma, where deploying IPv6 is only helpful if everyone else deploys IPv6." The IETF's own 2023 status report, RFC 9386, says of enterprises that they "are shielded from IPv4 address depletion issues" and "do not have the business requirement or technical justification to transition."

7.3 The stopgap worked

This is the reason underneath the others. NAT, the 1994 "short-term solution," turned out to carry the internet. Huston's running count: in 2017 the average address was shared by about ten devices; by January 2022, "in excess of 20 billion connected devices can be squeezed into some 2 billion active IPv4 addresses"; by January 2025, thirty billion into three billion. The forecast of 1994 was right about the date and wrong about the consequence, because the people making it treated address sharing as an architectural wrong and did not model it. Exhaustion arrived on schedule and nothing visibly broke for the ordinary user, who was behind a home router already. Cerf himself said in 2011, "now that we really are in the IPv4 end-game, there is not much choice but to deploy NATs to try to make dual-stack work as a transition plan."

NAT's success is lopsided, and the lopsidedness is this project's whole subject. Sharing an address solves the problem for a machine that only reaches out, and for an established server that keeps the address it has. It puts the cost on whatever needs to be reached: a new server, a game, a call, a peer. Huston, in 2012: "All of a sudden, content is no longer open and uniformly accessible. Content has to pay a tax to the carrier." A swarm is made of peers that accept connections. Address sharing removes that ability from the shared side, quietly, and no one who only browses ever notices.

7.4 The bootstraps broke, and the carriers did not

The mechanisms meant to let a single user start, by tunneling IPv6 through the IPv4 network on their own, failed in use. Huston measured in 2010 that about 13% of connections through the automatic tunnel called 6to4 failed, against 0.2% for plain IPv4, and later found that Teredo, Microsoft's tunnel, failed more than 40% of the time in 2011: "more of a liability than a positive contribution to IPv6." The IETF deprecated the 6to4 relay in 2015. What worked was the opposite shape: translation run by an operator that controlled both the handset and the core. T-Mobile US, which was running out of IPv4 addresses, put eight million phones on an IPv6-only network in 2014 with a translator in the phone and another in the core, a design called 464XLAT, and reached 93% of its subscribers on IPv6 by 2018. Jio did it at Indian scale from September 2016, and India's measured IPv6 share went from one percent to twenty in a year and to 77% by 2022. Neither waited for a killer application. Each had a killer constraint: tens or hundreds of millions of new devices to connect, each needing its own address, in a region where the pool was already closed, and the power to configure every device itself.

7.5 Who did it, and why

The networks that moved did so when moving was cheaper than not moving, and the record has them saying so. Free, a French provider, put more than 1.5 million customers on IPv6 in five weeks at the end of 2007, using a tunnel of its own design, because it built the box in every customer's home and the change was a firmware update. Comcast's Jason Livingood told the 2010 government workshop why his company had begun: "I actually think in the long term it's less expensive to have native access because you have fewer middle boxes in the network." Google's engineers made the case to themselves on a 2010 slide headed "Why IPv6? Cost": "Buying addresses will be expensive, more so over time. Carrier-grade NAT may be expensive." T-Mobile US, in the Internet Society's account, "was running out of IPv4 addresses." Jio had two hundred million customers to connect in a country holding forty million addresses. Amazon charges for IPv4 because its own cost to buy addresses rose three hundred percent in five years. And the content sites turned IPv6 on for good in 2012 when the number of users it would break fell below the number it would serve.

None of these is a company doing the public a favor, and none is a villain. Each is a company doing arithmetic. That is the answer to who did what and why: the engineers of 1995 wrote a protocol with no business case, as Hinden says below, and the business cases arrived twenty years later one network at a time, each in the form of a bill. Where the bill never arrived, because sharing was cheap and the old addresses were already paid for, nothing happened, and that describes Telecom Italia, most offices, and two thirds of the web's servers today.

7.6 Botched, mismanaged, or foolish to expect?

The engineering was not botched. The protocol works, has worked since the 1990s, and carries nearly half of Google's users today. The forecast of exhaustion was not botched; a window set in 1994 held. What was wrong was one assumption on which every deadline rested: that running out of addresses would force the migration. It did not, because sharing made running out survivable, and because the design made the first mover pay for everyone else's benefit. Two men who were there named the error in their own words. Bob Hinden, the protocol's other author, put on a slide in 2018: "We did not anticipate how Internet would change. No longer 'build it and they will come.' Now there has to be a business case." And "A lot of the industry was in denial for a long time." Cerf, asked for a regret in 2021: "I wish that we had been able to get everybody to use IPv6 right away and realize how useful IPv6 was before the network grew, but this was during the boom of the Internet bubble, so people had other things on their minds."

So the honest verdict is not failure. It is that the transition was planned for a network of a few million machines run by people who would upgrade because the standard said so, and was carried out in a network of billions run by companies that upgrade when it pays. Under the first assumption the deadlines were reasonable. Under the second they were never going to hold, and the people setting them learned that one deadline at a time.

8. Where it stands, October 2026

The companion report, Open Address, Closed Device, measures the present in detail; here is the shape, with the sources.

Set against what was promised, the record splits cleanly. Better than planned: the protocol itself, which performs comparably to IPv4, as RIPE's measurements found on launch day in 2012 and a 2014 study confirmed; the brokenness problem, which fell from about one user in a thousand to one in five thousand in two years and then stopped mattering; and the mobile networks, where one carrier could move a hundred million people in a year. Worse: the tunnels, every one of them; the home router, which held residential adoption back for a decade after the carriers were ready; the enterprise network, which has not moved; and the long tail of websites, which has no reason to. Faster than expected: India, France, the American carriers. Slower: fixed broadband nearly everywhere else, offices everywhere, and the content side as a whole. The hype, in the strict sense of claims that never came true at all, was narrower than memory suggests: the refrigerator, the 2005 deadline, the "new Internet." Most of what was promised arrived. It arrived for half the users, on the devices least able to use it, and twenty years late.

Google's monthly average for September 2026 is 47.5% of its users. The chart shows single days above half, because weekends run about three points above weekdays, with people at home and on their phones rather than on office networks; the first day above 50% was 28 March 2026, a Saturday, and the highest so far, 51.7%, was Saturday 3 October 2026. The month is still under half. And the slope is flattening. Table 2 is computed from Google's daily data file, as the monthly mean for each December and the change from the December before, and the companion report's data folder reproduces it.

Table 2. Google's users on IPv6, mean of each December, and the gain over the year. Source: Google's adoption data file.

December	Share	Gain in the year
2012	1.0%	0.6
2013	2.5%	1.5
2014	4.9%	2.4
2015	8.8%	3.9
2016	14.0%	5.2
2017	19.8%	5.7
2018	24.1%	4.4
2019	27.5%	3.4
2020	31.5%	4.1
2021	35.1%	3.6
2022	39.7%	4.5
2023	42.3%	2.6
2024	44.6%	2.3
2025	46.6%	2.1

The gain peaked at 5.7 points in 2017, the year Jio's subscribers arrived, and has run between two and three points a year since 2023. From December 2025 to September 2026 it was 0.9.

The reason is visible in the per-country numbers, because adoption moves in steps when a carrier switches, not in a slope when users choose. Google's October 2026 figures put France at 87%, Germany 76%, Belgium 74%, Saudi Arabia 69%, India 68%, Malaysia 63%, and Japan, Brazil, the United States, and the United Kingdom in the mid fifties. APNIC's per-network measurements show why: T-Mobile US at 94%, Jio 96%, Free 93%, Orange 92%, BT 90%, Comcast 85%, Deutsche Telekom 81%. The laggards are the mirror image. Italy is at 19% because Telecom Italia is at 0.7%; Spain at 13% because Orange Spain is at 0.3% and Telefónica only began in 2023; South Korea at 19% with KT at 14%; Russia at 5%. The African region as a whole is at 7%. China has three numbers: its government reports 865 million users and 77%, APNIC's advertising-delivered sample finds about 50%, and Google's figure of 7% reflects only the few users who reach Google from inside the country. India, the single largest contributor to the world's rise, has been flat to slightly down since the end of 2023. The growth of 2025 and 2026 came instead from China, Indonesia, where Telkomsel went from 8% to 59% in two years, Türkiye, where the incumbent went from nothing to 32% in a year, Brazil, Japan, and Western Europe.

The content side lags the user side by a wide margin and always has. A 2025 study found that only 12.5% of the top hundred thousand websites were fully reachable over IPv6; a July 2026 crawl of the top million web hosts found 32% answering over it. Cloudflare reported 29% of requests to its dual-stacked sites arriving over IPv6 in 2025; the same network's requests from people, with automated traffic set aside, ran 41% in the companion report's September 2026 measurement, the difference being crawlers and scripts, which live in data centers and mostly on IPv4.

The economics have changed shape without changing direction. The price of an IPv4 address rose from about 11 dollars in 2011 to between 45 and 60 at the end of 2021, then fell: the mean was about 22 dollars in January 2026, and large blocks were under 13, because supply came onto the market. A lease runs about forty cents an address a month. Thirty-three million addresses changed hands in 2025. The registries' waiting lists are long, twelve to twenty months at ARIN and over a year at RIPE, which says new networks still want IPv4 badly enough to wait. Against this, the cloud providers made IPv4 a visible line item for the first time: since 1 February 2024 Amazon has charged half a cent an hour for every public IPv4 address, about 44 dollars a year, explaining that its own cost to acquire one "has risen more than 300% over the past 5 years," and Google Cloud moved to the same rate the same day.

The tools for running without IPv4 at all exist and are spreading slowly. Apple has required every App Store app to work on an IPv6-only network since June 2016. A 2020 standard, RFC 8925, lets a network tell a capable device to skip IPv4 entirely, which Apple's and Google's operating systems honor; the RIPE and IETF meeting networks run this way, with most devices taking the offer. Microsoft shipped the missing piece for Windows laptops, a translator for wired and wireless networks, as a public preview in June 2026. Microsoft's own cloud, as of July 2026, still cannot run a virtual machine without an IPv4 address. One government has set a date to turn the old protocol off: the Czech Republic resolved in January 2024 that its central services will stop answering over IPv4 on 6 June 2032, the twentieth anniversary of World IPv6 Launch. France's regulator reports 87% of residential customers enabled at the end of 2024 and operators forecasting near-completion by the end of 2027, without a mandate, through a published annual ranking.

The forecasters have stopped giving dates. Howard, in 2018, fitted an adoption curve and said "in just four years, global IPv6 penetration will reach 80%"; it reached 40. Huston's October 2024 column fits a straight line to APNIC's series and gets "late 2045" for completion, with the disclaimer that the only assumption is that nothing changes, and then says what he thinks:

This is where we are, caught in a partial state of transition to IPv6 that is taking on an unfortunate air of permanence!

Geoff Huston, "The IPv6 Transition," October 2024

His keynote that month was titled "Whatever Happened to IPv6?", and its abstract describes an internet that "has settled into a dual-stack state that is permanently 'in transition.'" The Internet Society's Mat Ford, marking the first day above half in April 2026, wrote of the day every measure agrees: "I wouldn't like to predict when that will be!"

9. The next ten years

What follows is our prediction, reasoned from the record above and from the companion report's measurements, and labeled as a prediction. We will be glad to be wrong in the direction of faster.

The line keeps rising, and keeps slowing. The last decade's growth came from a short list of decisions by very large carriers, and most of those carriers are now above ninety percent and have nothing left to add. What remains is a list of incumbents that have shown no sign of moving, a continent at seven percent, and a user base that is no longer growing much in the rich countries, where the number of networks has been falling since 2020. A straight line at the 2023 to 2025 pace puts Google's December average above 50% in 2027 and near 72% in 2036. A straight line at the 2026 pace puts 50% in 2028 and near 60% in 2036. We expect the lower line or a little under it, since the yearly gain has shrunk every year since 2022: the press will report the crossing of 50% from a weekend in 2026, the first full month above half will come in 2027 or 2028, and the figure in 2036 will be around 60%. We would be surprised by 75% and astonished by 90%.

The transition does not complete, and the word stops fitting. No country turns IPv4 off at national scale by 2036. The Czech date of 2032 is the one to watch, and its first status report, due in 2027, will show whether a government can hold such a date; if it holds, others will copy the form. But dual stack remains the state of the internet through the decade, and by its end the people running networks will have stopped calling it a transition. Huston's phrase, an air of permanence, is the one we expect to age best.

What could make it faster. We see six forces, in rough order of how much we would bet on them. First, cloud pricing: a recurring charge per IPv4 address is the first cost that reaches software developers rather than network operators, and it acts on exactly the side, the content side, that has lagged for twenty years; if the charge spreads and rises, the top-million figure of 32% could pass half. Second, a few incumbents flipping: Telefónica has started, and a decision at Telecom Italia, KT, or China Telecom would move a country's figure by tens of points in a year, as Türk Telekom's did. Third, IPv6-only networks becoming ordinary inside companies, now that Windows can join one, which lowers the cost of running a network without IPv4 and trains a generation of administrators who never had to. Fourth, China's stated plan to run single-stack IPv6 and remove translation, which, if its own figures are right, moves the largest national user base. Fifth, switch-off dates: if the Czech model spreads to the EU or to US federal services, the deadline finally has teeth it never had. Sixth, and least likely in the next few years, a return of scarcity pricing: if the supply of IPv4 addresses coming onto the market dries up, the price could climb back toward its 2021 peak and change the arithmetic for providers that have been leasing.

What holds it where it is. The same forces that held it for twenty years. Address sharing still works, and the sharing ratios keep rising without a visible crisis. IPv4 is cheap again, about twenty dollars to own and under five a year to lease, which is cheap next to re-engineering anything. Nothing switches off. Enterprises are shielded from scarcity and see cost without benefit. The content side stays on IPv4 because two thirds of hosts are IPv4-only, so no access network loses anything by waiting, and access networks that wait give content no reason to move. And measurement itself saturates: a country's share is bounded by its carrier mix, so once a carrier is at 95% its users contribute nothing further, and the world's figure becomes a weighted vote of the laggards.

What it means for a peer. The companion report found that the IPv6 that has arrived has arrived first on closed devices, phones and tablets on cellular networks, and slowest on the open devices that could use a reachable address to be reached: 48% of requests from closed devices came from IPv6 addresses in September 2026, against 29% from open ones. We expect that gap to narrow but not close. Fixed-line and office networks are where the open devices live, and they are the slow half of every country's figure; France shows that a residential network can reach 87%, and we expect the rich countries' fixed lines to follow over the decade, but offices to lag them, and the world average for open devices to reach perhaps 40% by 2036. Two peers drawn at random from open devices would then both have IPv6 about one time in six, against one in twelve today. That is a better path for a growing minority of pairs, not a replacement for the IPv4 path, and peer-to-peer software should plan on carrying both, and the whole apparatus of traversal that IPv4 needs, for the entire ten years.

The chart that opened this article will keep rising, slowly, and the ceremonies will not be repeated, because the people who held them have learned what the deadlines taught. What the internet got, in place of the transition that was announced, is two protocols running side by side with no end in sight, in which the larger address space arrived first and most completely on the devices least able to use it. That is the condition the companion report measures, and the one this project builds for.

References

1. Kirkpatrick, S., Stahl, M., and Recker, M. Internet Numbers. RFC 1166, July 1990. https://www.rfc-editor.org/rfc/rfc1166.txt
2. APNIC. IP addressing in China (2004). Apster, December 2004. https://apnic.net/about-apnic/publications-archive/brochures-2/ip-addressing-in-china-2004
3. Stanford University Networking Systems. IP Addressing Changes at Stanford. 1999, archived copy. https://web.archive.org/web/2003id_/http://www.stanford.edu/group/networking/NetConsult/ipchange/indexold.html
4. CNET News.com. United Nations ponders Net's future. March 2004, archived copy. https://web.archive.org/web/2004id_/http://news.com.com/United+Nations+ponders+Nets+future/2100-1028_3-5179694.html
5. Microsoft IPv6 team. Mythbusters #2: Stanford has more IP addresses than China. 14 May 2007. https://learn.microsoft.com/en-us/archive/blogs/ipv6/mythbusters-2-stanford-has-more-ip-addresses-than-china
6. Siddiqui, A. MIT goes on IPv4 selling spree. Internet Society, 31 May 2017. https://www.internetsociety.org/blog/2017/05/mit-goes-on-ipv4-selling-spree/
7. Internet Society. State of IPv6 Deployment 2018. 6 June 2018. https://www.internetsociety.org/resources/2018/state-of-ipv6-deployment-2018/
8. APNIC. Delegated address statistics, extended format. https://ftp.apnic.net/stats/apnic/delegated-apnic-extended-latest
9. Clark, D., Chapin, L., Cerf, V., Braden, R., and Hobby, R. Towards the Future Internet Architecture. RFC 1287, December 1991. https://www.rfc-editor.org/rfc/rfc1287.txt
10. Fuller, V., Li, T., Yu, J., and Varadhan, K. Classless Inter-Domain Routing (CIDR). RFC 1519, September 1993. https://www.rfc-editor.org/rfc/rfc1519.txt
11. Egevang, K., and Francis, P. The IP Network Address Translator (NAT). RFC 1631, May 1994. https://www.rfc-editor.org/rfc/rfc1631.txt
12. Srisuresh, P., and Egevang, K. Traditional IP Network Address Translator. RFC 3022, January 2001. https://www.rfc-editor.org/rfc/rfc3022.txt
13. Huston, G. Observations on the Management of the Internet Address Space. RFC 1744, December 1994. https://www.rfc-editor.org/rfc/rfc1744.txt
14. Partridge, C., and Kastenholz, F. Technical Criteria for Choosing IP The Next Generation (IPng). RFC 1726, December 1994. https://www.rfc-editor.org/rfc/rfc1726.txt
15. Bradner, S., and Mankin, A. The Recommendation for the IP Next Generation Protocol. RFC 1752, January 1995. https://www.rfc-editor.org/rfc/rfc1752.txt
16. Carpenter, B. IPng White Paper on Transition and Other Considerations. RFC 1671, August 1994. https://www.rfc-editor.org/rfc/rfc1671.txt
17. Deering, S., and Hinden, R. Internet Protocol, Version 6 (IPv6) Specification. RFC 1883, December 1995; RFC 2460, December 1998; RFC 8200, July 2017. https://www.rfc-editor.org/rfc/rfc8200.txt
18. National Telecommunications and Information Administration. Transcript, IPv6 Workshop. 28 September 2010. https://www.ntia.gov/files/ntia/publications/transcript_ipv6event_09282010.pdf
19. Salus, P. Summary of Steve Deering's invited talk, USENIX 1997. ;login:, April 1997. https://www.usenix.org/publications/library/proceedings/ana97/summaries/deering.html
20. Dyson, E. ABC News online chat, 26 August 1999, as forwarded to the ICANN comments list. https://archive.icann.org/en/comments-mail/icann-current/msg00643.html
21. Lea, G. MS volte face embraces next-gen Internet. The Register, 21 March 2000. https://www.theregister.com/2000/03/21/ms_volte_face_embraces_next/
22. Government of Japan, IT Strategic Headquarters. e-Japan Strategy, 22 January 2001, and e-Japan Priority Policy Program, 29 March 2001, archived copies. http://web.archive.org/web/2004id_/http://www.kantei.go.jp/foreign/it/network/0122full_e.html
23. Converge Digest. Catalyst keynote: Microsoft says IPv6. 8 July 2003. https://convergedigest.com/catalyst-keynote-microsoft-says-ipv6/
24. Ladid, L. IPv6 FORUM Founding Members. 6bone mailing list, 7 May 1999, SixXS archive. https://sixxs.net/archive/docs/ipv6-history/6bone-mailinglist/1999-May/002136.html
25. IPv6 Forum. About the IPv6 Forum (archived 2001) and home page (archived March 2000 and December 2004). https://web.archive.org/web/20011029100907id_/http://www.ipv6forum.org/navbar/ipv6forum/about.htm
26. Nokia Networks. IPv6: Benefits. Archived 28 January 2001. https://web.archive.org/web/20010128013600id_/http://www.nokia.com/ipv6/benefits.html
27. Sun Microsystems. Solaris IPv6: Renewing the Net. Archived 15 August 2000. https://web.archive.org/web/20000815070547id_/http://www.sun.com/software/solaris/ipv6/
28. Microsoft. Internet Protocol Version 6. Archived 24 August 2003. https://web.archive.org/web/20030824170330id_/http://www.microsoft.com/ipv6
29. Miyake, K. IPv6 comes to life in Japan. Computerworld, 8 June 2001. https://www.computerworld.com/article/1411441/ipv6-comes-to-life-in-japan.html
30. Reuters. Japan throws weight behind new Net standard. NBC News, 5 April 2004. https://www.nbcnews.com/news/amp/wbna4669525
31. MacTech, citing Reuters and USA Today. Addresses quickly running out on the Web. 10 May 2001. https://www.mactech.com/2001/05/10/addresses-quickly-running-out-on-the-web
32. Hardy, I. Tackling the net's numbers shortage. BBC News, 26 October 2003, archived copy. https://web.archive.org/web/2004id_/http://news.bbc.co.uk/2/hi/technology/3211035.stm
33. McLaughlin, A. Bad Journalism, IPv6, and the BBC. CircleID, 7 November 2003. https://circleid.com/posts/bad_journalism_ipv6_and_the_bbc
34. APNIC. Myths: the myth of address shortage. https://apnic.net/about-apnic/publications-archive/brochures-2/myths
35. Stenbit, J. Internet Protocol Version 6 (IPv6). US Department of Defense memorandum, 9 June 2003, archived copy. http://web.archive.org/web/2004id_/http://www.defenselink.mil/nii/org/cio/doc/IPV6.pdf
36. Nextgov. DOD moving to IPv6. 13 June 2003. https://www.nextgov.com/people/2003/06/dod-moving-to-ipv6/248720/
37. Evans, K. Transition Planning for Internet Protocol Version 6 (IPv6). OMB memorandum M-05-22, 2 August 2005. https://obamawhitehouse.archives.gov/sites/default/files/omb/assets/omb/memoranda/fy2005/m05-22.pdf
38. Hain, T. A Pragmatic Report on IPv4 Address Space Consumption, with Huston, G., response. The Internet Protocol Journal 8(3), September 2005. https://ipj.dreamhosters.com/wp-content/uploads/issues/2005/ipj08-3.pdf
39. Huston, G. The Changing Foundation of the Internet: Confronting IPv4 Address Exhaustion. The Internet Protocol Journal 11(3), September 2008. https://ipj.dreamhosters.com/wp-content/uploads/issues/2008/ipj11-3.pdf
40. European Commission. An unlimited source of Internet addresses to be on stream in Europe by 2010. Press release IP/08/803, 27 May 2008. https://ec.europa.eu/commission/presscorner/api/files/document/print/en/ip_08_803/IP_08_803_EN.pdf
41. Leibson, S. IPV6: How Many IP Addresses Can Dance on the Head of a Pin? EDN, 28 March 2008, archived copy. https://web.archive.org/web/20140101000000/http://www.edn.com/blog/980000298/post/820024082.html
42. ARIN public policy mailing list, thread on the grains-of-sand comparison, September 2011. https://lists.arin.net/pipermail/arin-ppml/2011-September/022992.html
43. Google. IPv6 adoption statistics and data file. https://www.google.com/intl/en/ipv6/statistics.html and https://www.google.com/intl/en_ALL/ipv6/statistics/data/adoption.js
44. RIPE NCC. Stenographer's transcript, RIPE 57 plenary, 30 October 2008. https://www.ripe.net/meetings/ripe-meetings/ripe-57/steno/thursday-amal-2pm/
45. Nextgov. Pentagon dropped the ball on Internet upgrade, audit says. 7 April 2015. https://www.nextgov.com/digital-government/2015/04/pentagon-dropped-ball-internet-upgrade-audit-says/109534/
46. Computerworld. The Pentagon's groundbreaking IPv6 project hasn't broken much ground yet. 6 April 2015. https://www.computerworld.com/article/1648396/the-pentagons-groundbreaking-ipv6-project-hasnt-broken-much-ground-yet.html
47. US Government Accountability Office. Internet Protocol Version 6: DOD Needs to Improve Transition Planning. GAO-20-402, 1 June 2020. https://www.gao.gov/products/gao-20-402
48. Office of Management and Budget. Scorecard press release, 1 July 2008. https://georgewbush-whitehouse.archives.gov/omb/pubpress/2008/070108_scorecard.html
49. RIPE NCC. Stenographer's transcript, RIPE 60, 4 May 2010. https://ripe60.ripe.net/steno-transcripts.php%3Fsteno=Main-100504-AM2.html
50. Department of Telecommunications, India. IPv6 roadmap update, APNIC 33, February 2012. https://conference.apnic.net/33/pdf/n-ram-pp-on-ipv6-roadmap-updates_1330324098.pdf
51. Kundra, V. Transition to IPv6. OMB memorandum, 28 September 2010. https://obamawhitehouse.archives.gov/sites/default/files/omb/assets/egov_docs/transition-to-ipv6.pdf
52. Route Fifty. Graded on a curve: how feds lead the way on IPv6. 1 October 2012. https://www.route-fifty.com/digital-government/2012/10/graded-on-a-curve-how-feds-lead-the-way-on-ipv6/280468
53. Ministry of Science, ICT and Future Planning, South Korea. IPv6 Expansion Roadmap, March 2014. https://ipv6forum.com/dl/presentations/IPv6 Roadmap Korea.pdf
54. General Office of the CPC Central Committee and State Council. Action Plan for Promoting Large-Scale Deployment of IPv6, November 2017. https://www.gov.cn/zhengce/2017-11/26/content_5242389.htm
55. APNIC Blog. China reports major IPv6 growth. 4 December 2025. https://blog.apnic.net/2025/12/04/china-reports-major-ipv6-growth/
56. Vought, R. Completing the Transition to Internet Protocol Version 6 (IPv6). OMB memorandum M-21-07, 19 November 2020. https://www.whitehouse.gov/wp-content/uploads/2020/11/M-21-07.pdf
57. Number Resource Organization. Free Pool of IPv4 Address Space Depleted. 3 February 2011. https://www.nro.net/ipv4-free-pool-depleted
58. ARIN. ARIN IPv4 Free Pool Runs Dry. 24 September 2015. https://www.arin.net/vault/about/media/releases/20150924/
59. The Register. Microsoft pays Nortel $7.5m for IPv4 addresses. 24 March 2011. https://www.theregister.com/2011/03/24/microsoft_ip_spend/
60. The Register. IPv4: the last 200 days. 25 January 2011. https://www.theregister.com/2011/01/25/ipv4_last_200_days/
61. Internet Society. Successful World IPv6 Day Demonstrates Global Readiness for IPv6. 9 June 2011. https://elists.isoc.org/pipermail/chapter-delegates/2011-June/008103.html
62. Internet Society. World IPv6 Day: FAQ. Archived 5 June 2011. https://web.archive.org/web/20110605220319id_/http://www.worldipv6day.org/faq/index.html
63. Colitti, L. World IPv6 Day begins 24 hours from now, with the post-event update. Google, 6 June 2011, archived copy. https://web.archive.org/web/20110613002609id_/http://googleblog.blogspot.com/2011/06/world-ipv6-day-begins-24-hours-from-now.html
64. Lee, D. Exciting Results from World IPv6 Day. Facebook Engineering, 8 June 2011, archived copy. https://web.archive.org/web/20110704164341id_/https://www.facebook.com/note.php?note_id=10150198443513920
65. Malan, R. World IPv6 Day: Final Look and "Wagon's Ho!" Arbor Networks, 9 June 2011, archived copy. https://web.archive.org/web/20110612135439id_/http://asert.arbornetworks.com/2011/06/world-ipv6-day-final-look-and-wagons-ho/
66. Keranen, A., and Arkko, J. Some Measurements on World IPv6 Day from an End-User Perspective. RFC 6948, July 2013. https://www.rfc-editor.org/rfc/rfc6948.html
67. Murphy, K. World IPv6 Day fails to kill the internet. The Register, 9 June 2011. https://www.theregister.com/2011/06/09/ipv6_fails_to_kill/
68. Rashid, F. Y. World IPv6 Day Ends, Everyone Goes Back to IPv4. eWeek, 9 June 2011. https://www.eweek.com/networking/world-ipv6-day-ends-everyone-goes-back-to-ipv4/
69. Bly, J. What happened on World IPv6 Day. ARIN, 23 June 2011. https://www.arin.net/blog/2011/06/23/what-happened-on-world-ipv6-day
70. Marsan, C. D. 'IPv6 brokenness' problem appears fixed. InfoWorld, 27 July 2011. https://www.infoworld.com/article/2272257/ipv6-brokenness-problem-appears-fixed.html
71. Ellis, L. What happened on World IPv6 Day. Multichannel News, 20 June 2011. https://www.nexttv.com/news/what-happened-world-ipv6-day-126912
72. Internet Society. World IPv6 Launch Solidifies Global Support for New Internet Protocol. 17 January 2012. https://www.internetsociety.org/news/press-releases/2012/world-ipv6-launch-solidifies-global-support-for-new-internet-protocol/
73. Internet Society. World IPv6 Launch: FAQ. Archived 6 June 2012. https://web.archive.org/web/20120606id_/http://www.worldipv6launch.org/faq/
74. Sharwood, S. Techies beg world to join the 1% on IPv6 launch day. The Register, 5 June 2012. https://www.theregister.com/2012/06/05/ipv6_launch_day/
75. van Beijnum, I. World IPv6 Launch gets 27 percent of pageviews on IPv6. Ars Technica, 6 June 2012. https://arstechnica.com/information-technology/2012/06/world-ipv6-launch-gets-27-percent-of-page-views-on-ipv6/
76. Daigle, L. On June 6th, World IPv6 Launch Will Change The World Forever. Internet Society, 16 May 2012. https://www.internetsociety.org/blog/2012/05/on-june-6th-world-ipv6-launch-will-change-the-world-forever/
77. Kline, E. IPv6: countdown to launch. Google, 17 January 2012, archived copy. https://web.archive.org/web/20120118004339id_/http://googleblog.blogspot.com/2012/01/ipv6-countdown-to-launch.html
78. Cerf, V. World IPv6 Launch: Keeping the Internet growing. Google, 5 June 2012. https://blog.google/inside-google/company-announcements/world-ipv6-launch-keeping-internet/
79. Nygren, E. 10 Years Since World IPv6 Launch. Akamai, 6 June 2022. https://akamai.com/blog/trends/10-years-since-world-ipv6-launch
80. Kruse, M. Happy Launchiversary, IPv6! Celebrating 4 Years Since World IPv6 Launch. Internet Society, 6 June 2016. https://www.internetsociety.org/blog/2016/06/happy-launchiversary-ipv6-celebrating-4-years-since-world-ipv6-launch/
81. Chirgwin, R. Worst. Birthday. Ever. IPv6's party falls flat. The Register, 10 June 2018. https://www.theregister.com/2018/06/10/network_roundup_june_6_2018/
82. McCarthy, K. Vint Cerf: it's time to get with the program. The Register, 6 June 2018. https://www.theregister.com/2018/06/06/vint_cerf_ipv6_program/
83. Internet Society. World IPv6 Launch, archived site. https://www.worldipv6launch.org/
84. Marsan, C. D. Leading Internet prognosticator still pessimistic about IPv6. InfoWorld, 13 June 2012. https://www.infoworld.com/article/2287740/leading-internet-prognosticator-still-pessimistic-about-ipv6-2.html
85. Bernstein, D. J. The IPv6 mess. About 2002. https://cr.yp.to/djbdns/ipv6mess.html
86. Bush, R. IPv6 Transition & Operational Reality. APNIC 24, 6 September 2007. https://conference.apnic.net/24/program/plenaries/apnic/presentations/bush-ipv6-op-reality.pdf
87. Marsan, C. D. Biggest mistake for IPv6: It's not backwards compatible, developers admit. Network World, 25 March 2009. https://www.networkworld.com/article/789774/lan-wan-biggest-mistake-for-ipv6-it-s-not-backwards-compatible-developers-admit.html
88. Grundemann, C. A Chat with Scott Bradner. 2012. https://chrisgrundemann.com/index.php/2012/chat-scott-bradner/
89. LACNIC Blog. Interview with Radia Perlman, 12 June 2013. https://blog.lacnic.net/en/i-believe-in-making-things-more-user-friendly-more-than-i-believe-in-training-people/
90. Gilligan, R., and Nordmark, E. Transition Mechanisms for IPv6 Hosts and Routers. RFC 1933, April 1996. https://www.rfc-editor.org/rfc/rfc1933.txt
91. Huston, G. Is the Transition to IPv6 a "Market Failure?" The ISP Column, September 2009. https://ispcol.potaroo.net/2009-09/v6trans.html
92. Howard, L. The Cost of IPv4-IPv6 Transition. Rocky Mountain IPv6 Task Force, April 2013. https://www.rmv6tf.org/wp-content/uploads/2013/04/1-The-Cost-of-IPv4-IPv6-Transition.pdf
93. Howard, L. Prisoner of IPv4. NANOG 76, June 2019. https://prod.nanog.revsys.com/events/nanog-76/content/3702/
94. Fioccola, G., Volpato, P., Palet Martinez, J., Mishra, G., and Xie, C. IPv6 Deployment Status. RFC 9386, April 2023. https://www.rfc-editor.org/rfc/rfc9386.html
95. Huston, G. In Defence of NATs. 2017, slide transcription. https://slidetodoc.com/in-defence-of-nats-geoff-huston-chief-scientist/
96. Huston, G. IP Addressing through 2021. The ISP Column, January 2022. https://ispcol.potaroo.net/2022-01/addr2021.html
97. Huston, G. IP Addresses through 2024. The ISP Column, January 2025. https://ispcol.potaroo.net/2025-01/addr2024.html
98. Cerf, V. Vint Cerf Answers Your Questions About IPv6 and More. Slashdot, 25 October 2011. https://interviews.slashdot.org/story/11/10/25/1532213/vint-cerf-answers-your-questions-about-ipv6-and-more
99. Huston, G. Flailing IPv6. The ISP Column, December 2010. https://ispcol.potaroo.net/2010-12/6to4fail.html
100. Huston, G. IPv6 Performance. The ISP Column, November 2015. https://ispcol.potaroo.net/2015-11/v6perf.html
101. Troan, O., and Carpenter, B. Deprecating the Anycast Prefix for 6to4 Relay Routers. RFC 7526, May 2015. https://www.rfc-editor.org/rfc/rfc7526.txt
102. Internet Society. Case Study: T-Mobile US Goes IPv6-only Using 464XLAT. 13 June 2014. https://www.internetsociety.org/resources/deploy360/2014/case-study-t-mobile-us-goes-ipv6-only-using-464xlat/
103. Aben, E. Measuring World IPv6 Launch: Comparing IPv4 and IPv6 Performance. RIPE Labs, 21 June 2012. https://labs.ripe.net/author/emileaben/measuring-world-ipv6-launch-comparing-ipv4-and-ipv6-performance/
104. Czyz, J., Allman, M., Zhang, J., Iekel-Johnson, S., Osterweil, E., and Bailey, M. Measuring IPv6 Adoption. ACM SIGCOMM 2014. https://faculty.cc.gatech.edu/~mbailey/publications/sigcomm14_ipv6.pdf
105. Ghosh, M. Reliance Jio boosts India past 20% IPv6 capability. APNIC Blog, 7 February 2017. https://blog.apnic.net/2017/02/07/reliance-jio-boosts-india-past-20-ipv6-capability/
106. Despres, R. IPv6 Rapid Deployment on IPv4 Infrastructures (6rd). RFC 5569, January 2010. https://www.rfc-editor.org/rfc/rfc5569.txt
107. Colitti, L. IPv6 at Google. APRICOT 2010 slides, March 2010. https://conference.apnic.net/29/pdf/IPv6-at-Google.pdf
108. Hinden, R. IP Version 6 (IPv6): Past, Present, and Future. June 2018. https://www.ripe.net/documents/1412/Bob_Hinden_-_IPv6_PastPresentFuture.pdf
109. LACNIC Blog. Vint Cerf: Internet success factors and challenges for the next 50 years. 13 October 2021. https://blog.lacnic.net/en/vint-cerf-internet-success-factors-and-challenges-for-the-next-50-years/
110. Ford, M. 18 Years Later, IPv6 Reaches Majority. Internet Society Pulse, 21 April 2026. https://pulse.internetsociety.org/en/blog/18-years-later-ipv6-reaches-majority/
111. APNIC Labs. IPv6 measurement, per economy and per network. https://stats.labs.apnic.net/ipv6
112. Thottungal Valapu, S., and Heidemann, J. Towards a Non-Binary View of IPv6 Adoption. ACM IMC 2025; arXiv 2507.11678, July 2025. https://arxiv.org/html/2507.11678v1
113. Common Crawl. Is one vantage point enough? IPv6 across the top million web hosts. 17 July 2026. https://commoncrawl.org/blog/is-one-vantage-point-enough-ipv6-across-the-top-million-web-hosts
114. Cloudflare. Radar 2025 Year in Review. December 2025. https://blog.cloudflare.com/radar-2025-year-in-review
115. Huston, G. IP Addresses through 2025. The ISP Column, January 2026, republished by ARIN. https://www.arin.net/blog/2026/01/22/ip-addresses-through-2025/
116. ARIN. IPv4 Waiting List distribution, 7 October 2026. https://www.arin.net/announcements/20261007/
117. Barr, J. New: AWS Public IPv4 Address Charge. AWS News Blog, 28 July 2023. https://aws.amazon.com/blogs/aws/new-aws-public-ipv4-address-charge-public-ip-insights/
118. Apple. Supporting IPv6-only Networks. Apple Developer News, 4 May 2016. https://developer.apple.com/news/?id=05042016a
119. Colitti, L., and Linkova, J. IPv6-Only Preferred Option for DHCPv4. RFC 8925, October 2020. https://datatracker.ietf.org/doc/rfc8925/
120. Microsoft. IPv6 for Azure Virtual Network. Microsoft Learn, updated July 2026. https://learn.microsoft.com/en-us/azure/virtual-network/ip-services/ipv6-overview
121. SIDN. Czech government to withdraw IPv4 support in 8 years. 2024. https://sidn.nl/en/news-and-blogs/czech-government-to-withdraw-ipv4-support-in-8-years
122. Arcep. Barometer of the Transition to IPv6, 2025 edition. https://arcep.fr/fileadmin/reprise/observatoire/ipv6/Arcep_2025_Barometer_of_the_Transition_to_IPv6.pdf
123. LACNIC Blog. Interview with Lee Howard, 28 June 2018. https://blog.lacnic.net/en?p=10809
124. Huston, G. The IPv6 Transition. The ISP Column, October 2024. https://ispcol.potaroo.net/2024-10/ipv6-transition.html
125. Huston, G. BGP in 2025. The ISP Column, January 2026. https://ispcol.potaroo.net/2026-01/bgp2025.html
126. ftorrent. Open Address, Closed Device. October 2026. https://docs.ftorrent.com/open-address-closed-device.html