IPv6, the solution that took 30 years to roll out

Field note. Most players still connect over IPv4. Some on mobile networks are dual-stack. Understanding the transition explains some weird CGNAT-related disconnect patterns.

IPv4 has 4.3 billion addresses. IPv6 has 340 undecillion. That's 3.4 × 10^38, or about 7.9 × 10^28 addresses per person on Earth. If we built a Dyson sphere and addressed every atom in it, we'd still have addresses to spare.

The standard was finalized in 1998. So why isn't the whole internet on it yet?

This article explains what IPv6 is, why it's better, why the transition has dragged, and what it means for ordinary users.

The headline number

IPv6 uses 128-bit addresses. That's 2^128 = 340,282,366,920,938,463,463,374,607,431,768,211,456 possible addresses.

If you assigned one IPv6 address to every grain of sand on Earth, you'd use ~5 × 10^22, leaving 99.9999999999999% of the IPv6 space available.

The number is absurd by design. The IETF (who designed IPv6 in the mid-1990s) wanted to make sure they'd never have to do this again.

What an IPv6 address looks like

2001:0db8:85a3:0000:0000:8a2e:0370:7334

Eight groups of four hex digits separated by colons. Each hex digit is 4 bits; each group is 16 bits; eight groups times 16 bits is 128 bits.

You can shorten the representation:

• Leading zeros in each group can be dropped: 0db8 → db8.
• One run of all-zero groups can be collapsed to ::: 2001:db8::1.

So 2001:0db8:0000:0000:0000:0000:0000:0001 becomes 2001:db8::1.

When writing IPv6 addresses with ports, the address is wrapped in brackets:

[2001:db8::1]:8080

Otherwise the colon in the port would be ambiguous with the colons in the address.

Special IPv6 addresses

A few you'll see:

• ::1: localhost (equivalent of IPv4's 127.0.0.1).
• ::: "any address" or unspecified.
• fe80::/10: link-local. Auto-assigned. Only works on the local network segment.
• fc00::/7: unique local addresses. The IPv6 equivalent of private addresses (RFC 1918 in IPv4). Used inside private networks, not routed on the public internet.

What IPv6 changes beyond size

It's not just "IPv4 with more bits." A few design improvements:

Simpler header. IPv4 has many optional header fields, complicating router processing. IPv6's header is fixed-size, easier for routers to handle quickly.

No NAT (in theory). With trillions of addresses, every device can have a real public IP. NAT becomes unnecessary for address conservation. (In practice, networks use NAT-like translation for other reasons, but the address-shortage motivation goes away.)

Stateless autoconfiguration (SLAAC). Devices can self-assign IPv6 addresses without a DHCP server, using the network prefix and their MAC address (or a randomized version). Makes networks simpler to deploy.

Multicast first-class. IPv6 makes multicast (one-to-many delivery) a built-in primitive, simplifying protocols that need it.

Built-in security (IPsec) optional but expected. Originally mandatory; later loosened. Still well-supported.

These are real improvements. Whether they justify the transition cost is the entire debate.

Why adoption was slow

Several reasons interlock:

No backward compatibility. A pure IPv4 device cannot talk to a pure IPv6 device. You need a translator (dual-stack hosts, gateways, etc.). Every device, app, network, and service has to be updated.

NAT bought us time. With NAT, IPv4 scarcity is a slow squeeze, not a cliff. Operators had no urgent reason to migrate.

Chicken-and-egg. Content providers don't enable IPv6 if few users have it. Users don't get IPv6 if few sites need it. Each waits for the other.

Operational complexity. Running dual-stack (both IPv4 and IPv6) is harder than running one. Each network change has twice the configuration. Easier to defer.

Cost without revenue. IPv6 adoption costs effort; it doesn't directly produce revenue. Hard to prioritize.

Some hardware doesn't support it. Older home routers, some embedded devices, some legacy systems. Replacing these takes time.

Despite all this, adoption has crept up year over year. By 2026, ~40 percent of global internet traffic is IPv6. In some countries (India, Germany, US) it's over 50 percent.

What it looks like in practice

If your ISP supports IPv6 (most major ones do by 2026), your devices typically get both an IPv4 and an IPv6 address. Modern operating systems prefer IPv6 when both are available; this is called "happy eyeballs."

For a typical user, IPv6 is invisible. Sites and services work; you don't notice which protocol is in use. You can check at sites like test-ipv6.com.

For service operators, supporting IPv6 means:

• Configuring AAAA records alongside A records.
• Ensuring the service listens on both stacks.
• Testing both paths regularly.

Subnetting in IPv6

A subtlety. IPv6 uses CIDR notation like IPv4 (2001:db8::/32). Common allocations:

• /32: a large network operator's allocation from their regional registry.
• /48: typical for a single organization or sub-allocation.
• /56: typical for a residential customer assignment.
• /64: a single subnet. Stateless autoconfig requires /64 minimum.

In IPv6, you don't worry about address conservation in subnetting. A single /64 has more addresses than the entire IPv4 internet. Allocate generously.

The transition mechanisms

A few approaches networks have used:

Dual stack. Run both IPv4 and IPv6 simultaneously. Every device has addresses in both. Most common approach.

6to4, 6rd, Teredo. Tunnel IPv6 inside IPv4. Older transition tech, used when one side lacks native IPv6.

NAT64 / DNS64. A gateway translates IPv6 traffic to IPv4 for clients that are IPv6-only. Used by mobile networks running IPv6-only internally.

464XLAT. A combination of techniques used by mobile carriers to give IPv6-only customers compatibility with legacy IPv4 services.

For ordinary users, none of this matters. Your devices and your ISP figure it out.

Privacy concerns and SLAAC

When IPv6 was first designed, the proposed address-generation method used the device's MAC address. This meant your IPv6 address was unique to your device, persistent across networks. Privacy nightmare: trivially trackable.

The fix: privacy extensions (RFC 4941). Devices generate random IPv6 addresses periodically. The "permanent" address (used for inbound services) and the temporary address (used for outbound connections) are different.

Modern operating systems do this by default. Privacy extensions are why your IPv6 address can look different each time you load a page.

IPv6 and game hosting

For game hosting:

Most games support IPv6 now, but many haven't been audited for IPv6 issues. Quirks happen.

Minecraft Java fully supports IPv6. Server and client both. Some plugin / mod stacks may have bugs but the core is solid.

Valve games partial. Steam supports IPv6 for many things; game traffic is sometimes IPv4-only depending on the game.

Valheim, Vintage Story, Eco: IPv6 support varies. Test before assuming.

SRV records still work but use AAAA targets when appropriate.

Some players are on networks with broken IPv6 (rare, but real). Make sure your game server has IPv4 reachability too as a fallback.

The practical recommendation: for game servers, dual-stack. Don't go IPv6-only yet.

Why IPv6 won't disappear

The Internet Architecture Board has made clear that IPv4 is being phased out, not replaced incrementally. As of 2026, the trajectory is clear:

• Major ISPs deploy IPv6 universally.
• Most cloud providers offer IPv6 by default.
• Content providers ramp IPv6 usage.
• IPv4-only services become increasingly anomalous.

The transition will take another decade or more, but it's not in doubt.

For network engineers in 2026: knowing IPv6 is increasingly part of the baseline skill set. For users: it's becoming transparent.

A final perspective

IPv6 is the technically correct fix for IPv4 scarcity. Its slow adoption isn't because of technical flaws; it's because of inertia, NAT, and the fundamental difficulty of replacing infrastructure used by billions.

The transition is happening. It's just on internet-scale time, not human-impatience time.

In 20 years, the IPv4 era will look like a curious historical period. For now, we live in the slow transition between two designs, dealing with the complexity of both.

Coming up

Next: NAT in depth. We've mentioned it; now we'll cover how it actually works, what it lets you do, and what it breaks.