The strangest commodity market on the internet just had its busiest spring on record. Roughly 29.7 million IPv4 addresses changed hands in the first half of 2026, by the registry transfer logs. That is more than half of 2025's full-year total and nearly 30 percent ahead of the five-year average pace. Extrapolated to December, 2026 lands near 59.4 million transferred addresses, which would edge past 2025's all-time record of 58.1 million.
Prices tell the opposite story. CircleID's half-year analysis of publicly priced transactions puts the average at $20.04 per address, down 35.7 percent from the first half of 2025 and roughly a third of the market's late-2021 peak. Volume up, price down: a reset, not a collapse. This post walks through the half-year numbers, what actually sets the price of an IPv4 address, and why nearly 30 million moved addresses matter even if you never plan to buy one.
The half-year in numbers
Two datasets describe this market, and they measure different things. Registry transfer statistics count every inter-organization transfer recorded by the five regional internet registries, including private deals and acquisitions where no price is ever disclosed; that is where the 29.7 million figure comes from. Marketplace tallies count only deals with published prices, a much smaller but price-transparent slice: 596 transactions covering just over five million addresses in the first half, worth an estimated $58.5 million.
| H1 2026 priced-transaction snapshot | Value |
|---|---|
| Transactions with published prices | 596 |
| Addresses in those deals | 5,016,064 |
| Estimated deal value | $58.5 million |
| Average / median price per address | $20.04 / $20.00 |
| Most traded block size | /24 (30.5 percent of deals) |
| Estimated lease rate | $0.59 per address per month |
The /24 dominates deal flow for a structural reason: 256 addresses is the smallest slice of IPv4 space that operators will accept into the global routing table, which makes it both the minimum unit of trade and the unit with the largest pool of buyers. Almost anyone running infrastructure can deploy a /24 tomorrow; very few organizations can absorb a /16.
From $60 to $20: how the price found its floor
An IPv4 address cost nothing for the internet's first three decades: you asked your registry, justified the need, and received space. Exhaustion ended that between 2011 and 2019 as the registries ran dry one by one, and a transfer market took over. Prices climbed for a decade and topped out near $60 per address in late 2021, an era of cheap capital, pandemic-driven cloud expansion, and speculative buying.
Then the inputs reversed. Cheap money disappeared, and address blocks are exactly the kind of yield-free asset that suffers when capital gets expensive. In February 2024 AWS began charging $0.005 per hour for every public IPv4 address, about $43.80 per year, which put an official rent on an address and pushed enormous amounts of hoarded and idle space back toward the market. Legacy holders, some sitting on blocks allocated in the 1980s and 1990s, sold into the softness. CircleID's analysis of the stalled transition notes that most transferred blocks today were allocated more than a decade ago: the market is almost entirely reuse, since only APNIC and AFRINIC retain even modest unallocated pools.
The crash was not uniform, and the way it split is the most interesting price signal in the market. xTom's price history has large blocks trading near $52 per address in late 2023 while /20 to /24 blocks fetched around $36, a 44 percent premium for size. By mid-2024 every block size had converged near $33. Then large blocks fell off a cliff: below $13 by late 2025, a ten-year low, with /16s slipping under $20 in May 2025 for the first time since 2019. The premium inverted. Buyers who can write a seven-figure check for a /16 are rare; buyers for a /24 are everywhere.
The first half of 2026 looks like a floor forming. Brander Group reports large, clean /16 to /18 blocks that recently traded near $9 now clearing $13 to $16, with smaller prefixes commonly moving between $18 and $24. The brokers disagree about what comes next: Brander Group and IPv4.Global read the volume surge as the start of a recovery into year-end, while CircleID's model projects a second-half average of $19.54, slightly softer still. Notably, both camps put the small-block floor right around $20.
Who is buying at the floor
The demand side has changed shape. Hosting providers are the steady core of it: Brander Group pointed to Hostinger and Hetzner earlier this year as examples of buyers acquiring space for immediate deployment rather than speculation. AI infrastructure is the new category: GPU data center buildouts still need front-end IPv4 for APIs, VPN gateways, and customer endpoints, and xTom lists it among 2026's significant demand drivers alongside the US government's $42 billion BEAD broadband program, which is projected to lift IPv4 demand by 10 to 20 percent as new rural networks come online.
Buyers are also pickier than they were during the boom. Well-documented blocks with clean histories command a 10 to 15 percent premium over space with a spotty past. A cheap block that arrives on spam blocklists, or that geolocation databases still place with its previous owner on another continent, costs more to rehabilitate than the discount saved.
The lease-or-buy math
A liquid lease market now runs alongside the sales market, and it is the cleanest way to see how the asset is valued. CircleID estimates the going lease rate at about $0.59 per address per month against a $20.04 purchase price, so buying pays for itself in roughly 34 months of use. Flip it around and the same numbers say a holder who leases out inventory grosses about 35 percent of the asset's value per year.
A 35 percent gross yield on a supposedly safe infrastructure asset is the market telling you it does not believe the asset is safe. The yield prices in the risk that values fall further, the risk that a block arrives with an abuse history that takes months to launder out of reputation systems, and above all the long-run possibility that IPv6 finally makes the whole market obsolete. Lease rates barely moved through 2025, holding in xTom's observed band of $0.38 to $0.50 across RIPE and ARIN space even as sale prices halved: the cost of using an address decoupled from the price of owning one.
IPv6 crossed 50 percent, and the IPv4 market barely noticed
On March 28, 2026, Google's connectivity measurements recorded 50.10 percent of users reaching Google over IPv6, the first time the number has crossed half, eighteen years after the measurement began. It is a genuine milestone, and it moved the IPv4 market not at all.
The reason is arithmetic. Dual-stack networking means a service still needs IPv4 to reach the other half of the internet, so crossing 50 percent changes almost nothing about who must hold addresses. The adoption curve also hides enormous variance: France and India sit above 70 percent while several large economies remain below 20, and the global line still dips every weekday as traffic shifts from IPv6-heavy home and mobile networks to enterprise networks where deployment lags. Geoff Huston's assessment, quoted in CircleID's address-crisis piece, is that NAT removed the urgency years ago and that the endgame may be fragmentation rather than a clean transition.
Put together, that stall is what defines the price band. The ceiling on an IPv4 address is the cost of engineering around it: carrier-grade NAT per subscriber, $43.80 per year in cloud rent, or going IPv6-only where the audience allows. The floor is the lease yield at which holding inventory beats selling it. Between those bounds, price is a negotiation over block size, registry, and reputation. The total allocated pool, meanwhile, is fixed for practical purposes: it declined by just 0.01 percent in 2025.
29.7 million moved addresses are 29.7 million geolocation updates
There is a reason IP data people watch transfer statistics as closely as brokers do. Every transfer is a discontinuity: a block that belonged to a European carrier on Monday can front a Singapore hosting company by Friday, changing its country, network type, connection class, and abuse profile overnight. The market's structural drift toward /24s makes it worse, since a single /16 sold off in pieces becomes up to 256 independent blocks with 256 separate fates. Databases that lean on stale registry records keep placing traffic with the seller for months, which is how payments get declined in the wrong country and region-locked content ends up on the wrong side of the wall.
Tracking that churn is core to what Ipregistry does: transfer logs, BGP origin changes, and operator geofeeds are folded into the dataset continuously, so a block that changed hands last week does not geolocate like it is 2019. You can try the API on your own traffic with 100,000 free lookups to get started.
Keep users in the loop Ship release notes that get read. Try Noticeable