← Blog · May 23, 2026 · ipv6, networking
IPv4 vs IPv6: Should You Migrate Your Internal Network in 2026?
Short answer — no, for most networks under 10,000 hosts. IPv4 private space (RFC 1918) gives you 17,891,328 addresses across three reserved blocks. That is enough for almost every company on earth. Dual-stacking your edge for IPv6 reachability — yes. Ripping out IPv4 internally — almost never worth it in 2026.
The available IPv4 private space
| Block | CIDR | Addresses | RFC |
|---|---|---|---|
| Class A | 10.0.0.0/8 | 16,777,216 | RFC 1918 (1996) |
| Class B | 172.16.0.0/12 | 1,048,576 | RFC 1918 |
| Class C | 192.168.0.0/16 | 65,536 | RFC 1918 |
| CGNAT | 100.64.0.0/10 | 4,194,304 | RFC 6598 (2012) |
If you need more than 17.9 million internal addresses, you are AWS, Google, or Cloudflare and you already moved to IPv6 internally years ago.
What IPv6 actually buys you
- 2128 addresses. The standard internal block (
fd00::/8ULA) gives every site a unique/48with 65,536/64subnets, each holding 264 addresses. That is more than the IPv4 internet, per subnet. - No NAT. Every host has a globally unique address (if you use global IPv6, not ULA). Simplifies peer-to-peer apps, VoIP, IPsec.
- SLAAC. Stateless address autoconfiguration — hosts pick their own address from a router-advertised prefix. No DHCP server needed for basic connectivity.
- Cleaner multicast. Built into the protocol, not bolted on.
What IPv6 costs you
- Operational training. Every junior engineer can troubleshoot IPv4. IPv6 has different troubleshooting patterns (neighbor discovery, link-local, extension headers).
- Tooling parity. Most enterprise monitoring, IPAM, and firewall UIs still treat IPv6 as a second-class citizen. Less-than-perfect dashboards, less-than-perfect logs.
- Vendor bugs. IPv6 paths are exercised less. Edge cases are rougher. I have seen production-grade firewalls drop fragmented IPv6 in 2024.
- Operator muscle memory. Typing
fd00:abcd:1234:5678::1is genuinely slower than typing10.1.5.1.
The honest decision matrix
| Your network | Recommendation |
|---|---|
| < 1,000 hosts, single site | IPv4 only internally. Dual-stack at the edge if your ISP gives you IPv6. |
| 1,000 – 10,000 hosts, multi-site | IPv4 with planned 10/8 allocation. Add IPv6 to user-facing services for SEO/reachability. |
| 10,000 – 100,000 hosts | Start the IPv6 migration plan. Dual-stack everything. Greenfield = IPv6-first. |
| > 100,000 hosts | IPv6-first internally, IPv4 only where legacy services need it. NAT64 for IPv6-only clients reaching IPv4 services. |
| Mobile / IoT fleet | IPv6. Carriers already do this — T-Mobile US has been IPv6-only with 464XLAT since 2014. |
| Cloud-native, Kubernetes-heavy | IPv6 dual-stack. The pod-IP exhaustion problem in IPv4 Kubernetes is real at scale. |
What you should definitely do, regardless
- Make every public-facing service dual-stack. An AAAA record next to your A record. Google has been ranking IPv6-reachable services since 2008. Apple and US federal sites enforce IPv6-only as of 2025.
- Test IPv6 reachability on your office WAN. If it's not there, your "dual stack" isn't actually dual.
- Plan your internal IPv4 space at least three jumps ahead. Picking
192.168.1.0/24for a 50-person office is fine. Picking it for a company that might one day acquire another company that also uses192.168.1.0/24is a future merger headache.
The state of IPv6 adoption
Google measures the percentage of users reaching it over IPv6. As of early 2026, global IPv6 adoption is ~45%, up from 30% in 2022. Country leaders: France (~75%), India (~73%), Germany (~70%), USA (~52%). Country laggards: most of Africa and Latin America are still under 15%. China is at ~30% and rising fast.
For your users, IPv6 reachability genuinely matters. For your data center, internally, in 2026 — IPv4 with disciplined allocation is still the pragmatic default.