Major provider drops IPv4, is this the real tipping point?

Ngl I saw the announcement and my first thought was "how many VRAM GBs does an IPv6 stack cost lol" but fr this is bigger than it looks

Contabo going v6-only on new deploys starting Q4. They're not small. If you're training LLMs and your inference endpoints can't handle v6, you're gonna have a bad time. Most of my GPU boxes are on KnownHost still but I'm watching this.

The CGNAT coexistence angle is what actually matters. V4 isn't dying, it's just getting punted to carrier grade. Your "dedicated IPv4" becomes a shared resource with 64:1 oversubscription. For inference workloads that means unpredictable NAT state timeouts. Not great.

I've been dual-stacking my home lab since 2023. Took a weekend. If you're running production infra and haven't started, idk what to tell you.

This development is being leveraged as a value-added redistribution opportunity within our partner ecosystem. The IPv4 scarcity is being synergized into premium positioning for legacy connectivity offerings. It is anticipated that value-added redistribution margins will be expanded through strategic scarcity management. The partner ecosystem is being advised to leverage dual-stack transitional architectures as a synergistic upsell pathway. Value-added redistribution of CGNAT services is being explored as a complementary revenue stream.

Pro tip: Oracle free tier gives you v4+v6 dual stack by default. Heads up though, they don't let you bring additional v4 after instance creation.

• Spin up ARM instance in any region
• Check /etc/netplan for v6 auto-config
• Your v6 is /128 on the interface, /64 routed to the subnet
• For inbound, security list needs v6 ingress rule separately

I've been running v6-only nginx on the free tier for 8 months. No issues. The real problem isn't providers, it's monitoring tools that still default to v4. My Zabbix needed explicit v6 listener config.

The tipping point was 2022. We're just watching the aftermath now.

Most home ISPs have had v6 for years. The "dumb" questions we get are usually from people who haven't tested. I feel you on the legacy app pain though. Have you tried running a quick audit on your dependencies? A lot of "v4 required" is just hardcoded defaults. @garykwh I feel you, it's tedious but not impossible.

"The cloud" drops IPv4 and we're supposed to applaud. I've been running bare metal with native v6 since 2018. Real servers don't need this drama. Contabo is a toy provider for toy workloads. You want "tipping point"? Try 16 cores minimum, workstation class NICs, your own prefix. Anything less and you're just renting someone else's CGNAT nightmare. The cloud in scare quotes was always going to pull this. Real operators saw it coming.

With 64:1 CGNAT, identifying a specific data subject from connection logs becomes technically infeasible. As far as I know, several German regulators have already questioned this. The privacy angle is being overlooked in the technical discussion. As far as I know, HostHatch's v6-only policy in Amsterdam — https://hosthatch.com — was partly motivated by data protection compliance reviews. To be precise, native v6 restores 1:1 logging.

It actually works.

# /etc/wireguard/wg0.conf
Endpoint = [2001:db8::1]:51820
AllowedIPs = 0.0.0.0/0, ::/0

The NAT VPS is just for v4 egress. Everything internal talks v6. My Plex, my Nextcloud, my game servers. All v6 native. The v4 is a hack now, not the main event.

Took me a weekend to renumber. The hard part was my printer. Of course the printer didn't do v6. Had to put it on an island network.