IPv8

01 The problem we've been ignoring for 15 years

On February 3, 2011 IANA handed out the last free IPv4 /8 blocks. Since then no more central space has been allocated. The five regional registries ran dry between 2011 and 2020, one after another. Today every usable IPv4 address in the world is held by someone — and the only way to get more is to buy them off that someone.

The patch of these years has been CGNAT: sharing one public IPv4 between dozens or hundreds of subscribers via massive port translation. It works, but it breaks peer-to-peer, complicates geolocation, inflates session latency and turns forensic attribution into a precision sport. ISPs know it, users endure it without realising.

Meanwhile the address market has become a sector with operators, brokers and quarterly quotes. A single IPv4 trades today between $32–$45 per IPv4.Global. A /24 exceeds $10,000. It's not a distortion: it's the equilibrium of a scarce resource treated as negotiable property.

02 Why IPv6 never quite arrived

IPv6 has existed since 1998. Twenty-eight years of being written, standardised, deployed. Yet by 2026 it carries only 45.2% of global traffic according to Google's statistics. Half of the internet plus one is still IPv4. The failure is not engineering — IPv6 is a sound protocol — it's incentives.

The IPv6 transition model demands dual-stack: every device, every application, every router must speak both protocols simultaneously for the duration of the migration. And that duration, absent a real forcing function, is indefinite. Every operator bears the operational cost of dual maintenance without immediate benefit unless their peers have migrated too. With no forcing externality, the rational equilibrium is to stay on IPv4 with CGNAT.

IPv6 breaks compatibility at the packet level. A router that only understands IPv4 cannot forward an IPv6 packet — it doesn't even know the header length. Every hop on the path must be upgraded before the route works. That multiplies the coordination cost by the number of operators involved, which is a big number.

After 25 years of deployment effort IPv6 carries a minority of global internet traffic. The operational cost of the dual-stack transition model, combined with the absence of management improvements, proved commercially unacceptable.

03 What IPv8 proposes

IPv8 changes the game: IPv4 is a proper subset of IPv8. An IPv8 address is 64 bits long and written r.r.r.r.n.n.n.n. The n.n.n.n half is the same 32-bit IPv4 field with identical semantics. The r.r.r.r half encodes the ASN — the number of the autonomous system the IP belongs to.

When r.r.r.r = 0.0.0.0, the IPv8 address is a classic IPv4 address, processed by the IPv4 rules of always. No router, no firmware, no existing IPv4 application needs modification. No flag day. No forced migration.

Each ASN holder receives 4,294,967,296 addresses — as many as the whole IPv4 space. A consumer ISP, a hyperscaler, a university lab: all with room to spare for decades, no CGNAT, no renumbering. Total space goes from 232 to 264 addresses, roughly 18 quintillion. Exhaustion stops being an architectural problem.

The global BGP table also changes. In IPv8, the rule is that the minimum inter-AS injectable prefix is /16. No more /24, /22, /20 inflating BGP4 up to today's 970,000 prefixes. The BGP8 table is bounded by the number of active ASNs, not by prefix proliferation. Today there are ~122,000 ASNs: that's your upper bound. Finite. Manageable.

04 How IPv8 rolls out without breaking anything

The key piece of the transition model is ARP8-driven version selection. When an IPv8 host wants to talk to a neighbour, it sends two probes in parallel: ARP8 at t=0 and ARP4 at t=50ms. Whatever answers decides how the packet is constructed from then on.

• If the neighbour replied to ARP8 → a full 64-bit IPv8 packet is transmitted.
• If it replied to ARP4 → a standard IPv4 packet is transmitted, with no r.r.r.r prefix on the wire.

Consequence: an IPv4-only device never receives a packet with version 8 in the IP header. It can't drop it because the IPv8 sender never constructs one in the version the neighbour doesn't understand. The classic transition error — "router doesn't recognise packet version, drop" — ceases to exist by construction.

To cross IPv4-only cores there's 8to4 anycast: each ASN publishes in its WHOIS8 record an IPv4 anycast address that its border routers announce via normal BGP4. An IPv8 packet needing to traverse an IPv4 transit gets encapsulated in an IPv4 wrapper to that anycast, the IPv4 core forwards it as normal traffic, and the destination's border router decapsulates. Zero preconfigured tunnels. Zero inter-operator coordination.

The migration bill for an IPv8 router fits in four lines of config. That's what makes deployment feasible.

05 What could go wrong with IPv8

IPv8 is a proposal, not a decree. It deserves the same critical reading as any other. These are the real weak points in the design — not to bury it, but to know what it's being evaluated against.

WHOIS8 as critical infrastructure

All north-south traffic validation depends on WHOIS8. Without WHOIS8, routers can't validate BGP8 routes and Zone Servers can't enforce egress control. The draft explicitly calls it a "critical infrastructure service". What in other protocols is an optional security layer, here becomes a prerequisite. If WHOIS8 goes down, it doesn't degrade — it breaks. And historical WHOIS has never been especially robust.

Zone Server concentrates too much

DHCP, DNS, NTP, authentication, route validation, access control, IPv4/IPv8 translation — all on the same platform. Elegant on paper, but multiplies the blast radius of any compromise. An operator who loses the Zone Server doesn't lose a service: they lose the network. Even/odd redundancy helps availability, not trust segregation.

Universal OAuth2 JWT is heavy in practice

Authenticating "every manageable element" via JWT requires key management, rotation, revocation, coherent caching and disaster recovery. A large ISP network has millions of elements. The operational ecosystem for this at that scale doesn't exist — no PKI, no tooling, no skills. The draft doesn't say how the first token is bootstrapped on a virgin device.

/16 minimum breaks current multihoming

Today thousands of companies multihome by announcing their own /24 from two ISPs. IPv8 prohibits that — only /16 or larger can be advertised. Anyone without a /16 is shut out of direct peering and depends on their upstream. For small operators this is a regression: they lose control over their routing.

Fragments IPv6's effort

The few operators who did migrate to IPv6 have invested ~25 years building that ecosystem. IPv8 proposes a third parallel transition. If the market splits between IPv6 and IPv8, neither adoption reaches critical mass. IPv6 at least has published RFCs, mature implementations, and presence in deployed hardware. IPv8 starts from zero.

It's an individual draft, not a WG item

draft-thain-ipv8-02 is submitted individually, with no IETF working group behind it. That means: no community consensus, no structured peer review, no commitment from major operators. Most individual drafts expire without becoming RFCs. It's an interesting proposal; it isn't yet a standard in progress.

None of these objections is fatal. All are tractable. But these are what an experienced operator will put on the table before planning anything. Ignoring them is the surest way for the draft not to advance.

06 What to do now

IPv8 is an individual IETF Internet-Draft, not yet an RFC. To move forward it needs review, experimental implementations and — the hardest part — operational adoption. If you run an AS, do network research, work for an equipment vendor or simply care how the substrate of all this evolves, here are reasonable next steps:

1. Read the draft on the IETF datatracker or in Spanish translation. ~70 pages total, dense but readable.
2. Review the companion specs. Zone Server, WHOIS8, Update8, RINE, Routing Protocols — each has its own individual draft.
3. Write to the author with technical comments: jamie@one.bm.
4. Participate at IETF. Internet Area Working Group is where this draft will seek traction.