IPv6 passed half the internet. IPv4 just had its busiest half-year.

Two things happened to the IPv4 market in the first half of 2026, and most of the commentary has picked one and ignored the other.

The first is that IPv6 finally crossed half the internet. The second is that the IPv4 transfer market moved 5,016,064 addresses across 596 transactions — more than two and a half times the volume of the same period a year earlier — while the average price per address fell to $20.04, down 35.7% year on year.

Read separately, those look like a protocol winning and an asset class dying. Read together, they are a description of a market that has finally started working properly.

Volume up, price down, and why that is not a contradiction

The instinct on seeing a third wiped off the average price is that demand has collapsed. Demand did not collapse. Transactions rose from 420 to 596, and the address count rose from roughly 1.95 million to just over 5 million. More buyers closed more deals than in any comparable period we have data for.

What changed is the mix. The single most common trade in the first half of 2026 was a /24 — 182 of the 596 deals, about 30.5% of the market. But the volume growth came from the other end: large, clean prefixes moving in size. A /16 does not price like a /24, and it never has. When the large end of the market wakes up, the blended average falls even if nothing any individual seller holds has become less valuable.

This is the number that gets misread most often in our conversations with clients. An average price across every block size is a statement about what traded, not about what your block is worth. A holder of a clean /22 reading a headline about a 35% decline is reading a statistic that mostly describes other people’s transactions.

The market is now genuinely global

On 4 February 2026, AFRINIC ratified its Number Resources Transfer Policy. The policy establishes the conditions under which IPv4 blocks and ASNs may move between AFRINIC members, and between AFRINIC and other regions where reciprocal policies exist. It also keeps a firm hand on the outcome: some resources are restricted from leaving the region, transfers require AFRINIC’s written approval, and transfers arranged outside approved channels are not recognised.

That last clause matters more than the headline. A ratified policy is not an open door; it is a defined one. Anyone treating the African region as newly unrestricted supply is going to spend a lot of time on filings that do not clear.

The regional picture explains why. AFRINIC recorded no transfers at all in the first half of 2026 — the policy was ratified in February, and a policy is not the same as a functioning market. Meanwhile ARIN and RIPE between them carried 563 of the 596 transactions. LACNIC recorded eleven, at the highest average price of any region, which is what thin liquidity looks like rather than what strength looks like.

RIPE deserves a note of its own. Its 24-month hold on transferred space means a block cannot be bought and flipped, which removes a class of speculative participant from the market and keeps supply tighter than the raw transaction count suggests. It is the reason RIPE’s average sat above ARIN’s despite ARIN closing more deals. If you are choosing where to source from, that structural detail is worth more attention than the price difference itself. We cover the specifics on our RIPE broker page and ARIN broker page.

IPv6 crossed 50%, and IPv4 got busier anyway

Google’s measurement of IPv6-capable clients crossed 50% for the first time in spring 2026. The Internet Society’s Pulse project marked the crossing at 50.10% on 28 March; by late April it was sitting there consistently.

Then the useful complication: APNIC Labs, measuring the same internet on the same date with a different weighting, put the figure at 42%.

Neither number is wrong. They answer slightly different questions — what fraction of clients can use IPv6, weighted how, and sampled from where. The eight-point gap is a useful reminder that “half the internet” is a headline rather than a measurement.

More to the point for anyone holding or needing address space: IPv6 capability is not IPv4 obsolescence. As APNIC’s George Michaelson put it in his write-up of the milestone, the internet now runs as a two-protocol network, with IPv4 persisting through direct connectivity, NAT and carrier-grade NAT alongside IPv6. Dual-stack is not a transitional inconvenience that ends on a date. It is the operating state, and it requires IPv4 on both ends of every connection that has not gone IPv6-only.

Which is precisely why the transfer market got busier in the same half-year that IPv6 passed 50%. Networks growing today still need reachable IPv4, and the only place left to get it is from someone who already has it.

What this means if you are holding space

The honest read is that the market got more efficient, not weaker. More participants, more closed transactions, tighter spreads, and now a policy path in all five regions. That is a better market to sell into than the one that existed two years ago, even though the headline average is lower.

Three things follow from the data:

• Stop pricing off the blended average. Your block prices off its size, its registry, whether it can move to where the buyer is, and its history. A clean, contiguous, well-documented prefix does not trade at the same number as the aggregate. Neither does one with blocklist entries attached.
• The lease case has strengthened, not weakened. At the current spread, an address bought outright pays back in roughly 34 months against prevailing lease rates. That maths cuts both ways: it is why buyers with a dated requirement increasingly lease rather than purchase, and it is why holders who expect to need their space back should look hard at a term lease before selling.
• Timing is a policy question, not just a price question. RIPE’s hold period, ARIN’s review of the recipient, LACNIC’s conditions and AFRINIC’s new approval requirement all put dates in your calendar that have nothing to do with what a buyer will pay. A deal agreed without checking those dates is a deal that sits.

And if you are on the other side of it — growing, and short of addresses — the same numbers read as an invitation. There is more supply reaching the market than at any point since exhaustion, at a lower average price, with a formal transfer path in every region for the first time. The seven acquisition models exist because the right structure depends on whether your constraint is capital or a deployment date.

What we are watching for the rest of 2026

IPv4.Center’s own forecast puts the year-end average at $19.54, essentially flat against the first half. We would treat that as a reasonable central case and watch three things that could move it:

1. Whether AFRINIC transfers actually start. Ratification in February, zero transfers by June. The first cleared inter-RIR transfer out of the region will tell us far more than the policy text does.
2. The ARIN draft policy lowering in-region usage requirements from a /22 to a /24. If it passes, a meaningful number of smaller organisations become eligible recipients, which pushes demand into exactly the block size that already dominates deal count.
3. Whether the large-prefix activity that drove H1 volume continues. If it does, expect the blended average to stay soft while individual /24 pricing holds up. That divergence will confuse a lot of headlines.

If you hold IPv4 space and want to know what it is actually worth — not what the average says — that is a conversation, not a calculator. Tell a consultant what you hold, roughly how it is used, and what your constraint is, and you will get a structure, a timeline and a number back.