A segment of the HN discussion argues the status quo is indefensible: the DoD sitting on 218 million addresses (more than all of Latin America) while legacy corporations like Ford, GE, and Halliburton hold entire /8 blocks is a historical accident that should be unwound. They view the current allocation as a form of digital land grant that no longer reflects actual need.
This camp points out that clawing back every legacy /8 would yield only about 18 months of global growth at current consumption rates. They argue the industry should have fully migrated to IPv6 a decade ago, and that debating IPv4 ownership is rearranging deck chairs while the actual solution has been sitting there unused.
The iptrust.co post frames IPv4 as physical land with deeds, highlighting the MIT-to-Amazon transaction as a banker's reference comparable and documenting prices rising from near-zero pre-2011 to ~$47/address today. The implicit argument: IPv4 is a tradeable commodity with brokers, commissions, and a real market, and engineering teams need to budget accordingly.
Practitioners running production dual-stack note that real-world IPv6 adoption has crawled from 30% to only 45% over five years. They argue that treating the IPv4 market as temporary is a planning error — architectures and budgets need to assume IPv4 scarcity pricing will persist indefinitely.
The iptrust.co post making the HN front page this week does something embarrassingly simple: it maps the IPv4 address space onto a flat grid and labels the plots. Each /8 block — 16,777,216 addresses — becomes a county. Each /16 becomes a neighborhood. Each /24 becomes a house. Then it draws the deeds.
The result is uncomfortable to look at. The US Department of Defense owns thirteen /8 blocks — roughly 218 million IPv4 addresses, more than the entire allocation of Latin America combined. MIT held 18.0.0.0/8 until 2017, when it sold half to Amazon in a transaction that bankers now use as a reference comparable. Ford, HP (via the Compaq and DEC acquisitions), GE, Eli Lilly, Halliburton, and Prudential each sit on a /8. Apple still has 17.0.0.0/8, which it has repeatedly declined to sell. Xerox PARC, Interop, and a handful of universities round out a group that was lucky enough to ask ARIN for addresses before anyone realized they'd become a commodity.
The HN thread (251 points, 180+ comments at time of writing) splits predictably. One camp wants the legacy /8s clawed back and redistributed. Another points out that the 'reclamation' would net you roughly 18 months of global growth and that we should have been on IPv6 a decade ago. The third camp — mostly network engineers who actually operate dual-stack production — notes that IPv6 adoption in the real world has crawled from 30% to about 45% over five years, and that pretending the IPv4 market will go away is a planning error, not a principle.
IPv4 addresses have become a tradeable asset class, and most engineering teams have not internalized what that means for their budgets or their architectures. The spot price for a /24 (256 addresses) is sitting around $12,000 as of Q3 2026 — roughly $47 per address — up from about $25 in 2021 and essentially zero before 2011. Brokers like Hilco, IPv4.Global, and Prefix Broker charge 8–12% commissions on transfers, which now move through ARIN, RIPE, and APNIC with the formality of a title company.
The scarcity is real but asymmetric. AWS, Google Cloud, and Azure collectively hold the three largest commercial IPv4 inventories outside the DoD, and all three now charge for public IPv4 attachment — AWS at $0.005/hour per IP since February 2024, which pencils out to $43.80/year. That's not a surcharge; that's AWS pricing against the secondary market, with a small arbitrage on top. The cost structure of 'spin up a VM with a public IP' has quietly become a cost structure of 'rent an address from your cloud's inventory,' and that line item is now visible enough that FinOps teams are starting to flag it.
The land analogy is useful here precisely because it exposes the dumb thing engineers keep doing. We treat IPv4 like water — a utility that shows up at the tap — when it's actually behaving like Manhattan real estate, with legacy owners, zoning (RIR policy), title transfers, and a brokerage market that extracts rent. The reason you're paying for NAT gateways, CGNAT translation layers, and 'elastic' IP attachment is not that networking is hard; it's that someone, somewhere, is paying the rent on the underlying address, and that cost is being passed down the stack to you with a cloud markup on top.
IPv6 is the obvious answer and has been for 25 years. The HN comments correctly point out that if you're running greenfield on Hetzner, OVH, or any modern CDN, you're probably already dual-stack and the IPv4 problem is largely invisible to you. The painful cases are the ones nobody wants to touch: legacy enterprise VPNs with hard-coded private ranges that collide on merger, IoT fleets whose firmware will never speak v6, and SaaS products whose customers' firewalls still require you to publish a static v4 egress list. Those are the systems paying the tax.
Three concrete moves worth making this quarter. First, audit your public IPv4 footprint and treat it as a line item. If you're on AWS, enable Public IPv4 Insights (the dashboard added in late 2024) and figure out which workloads actually need a public v4 versus which are legacy defaults. The common pattern is dozens of NAT gateways and load balancers burning v4 addresses that could trivially consolidate.
Second, if you're building a product that terminates connections from the public internet — API, SMTP, anything a customer firewalls against — publish your IPv6 ranges now, even if nobody's using them. The downstream cost of adding v6 after your IP allowlist is baked into customer runbooks is painful. Doing it preemptively is a few hours of DNS and ALB configuration.
Third, stop designing systems that assume more public IPv4 is available on demand. The RIR free pools are empty, waitlists exist, and the only way to add v4 is to buy on the transfer market or rent from a cloud — both of which price against the same scarcity curve. If your architecture needs 'another public IP' as a scaling knob, you've built a slow bomb.
The iptrust map is a nice piece of explanatory journalism, but the real takeaway isn't that the DoD got lucky in 1983. It's that the IPv4 scarcity story is now baked into cloud pricing, M&A integration, and every network migration — and that most engineering orgs are still treating it as a networking team's problem rather than a budget line and architectural constraint. The teams that come out ahead over the next five years are the ones that stop renting the house, buy their own v6 land, and only reach for a v4 lease when a specific customer requires it.
From a Georgist perspective, IPv4 is actually a form of land (it's a finite "natural resource", although that's not quite proper), and you could make a case for "land rent" being taxed on the addresses. This is the same as radio broadcast frequencies, web addresses, act
The only thing that surprised me was that Canada was so small compared to other Western countries. But I suppose that might be an artifact. The map shows IPv4 proportional to land area, but a "fair" allocation would have IPv4 proportionate to population, not land area, and I'd assume
IPv6 got some real traction in the last decade. From around 10% adoption up to 50%. It's not dead, like many are claiming. It works fine, just keep implementing it.
The fact that we have yet to fully deprecate IPv4 should be considered a great failure in the tech industry's competence.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
India: 47MChina: 358MBoth countries have roughly the same population. The IPv4 problem has been swept under the rug with CGNAT and a normal person doesn't even notice it. But it sucks. Forget having a dedicated IP, even on a broadband connection, you don't get a public IP that's truly