---
title: "What is CGNAT (carrier-grade NAT)?"
url: https://proxy.wiki/glossary/carrier-grade-nat/
type: Glossary Term
author: "proxy.wiki editorial"
published: 2026-09-06
updated: 2026-10-21
site: proxy.wiki
topics: ["Proxy fundamentals"]
license: CC BY 4.0; quote freely with attribution to https://proxy.wiki/
---

# What is CGNAT (carrier-grade NAT)?

> Carrier-grade NAT (CGNAT) is address translation inside an operator's network, so that many subscribers share one public IPv4 address.

Carrier-grade NAT is the reason a mobile proxy’s address is shared with strangers, and nearly everything that is good and bad about mobile proxies follows from that. Operators deploy it because they cannot give every subscriber a public IPv4 address of their own. It is standard on mobile networks, is also used on fixed broadband, and is usually abbreviated CGNAT or CGN.

## What happens to a packet on the way out

```
device                     carrier's translator             destination
100.72.14.9 : 51544  --->  rewrites address + port  --->  sees 203.0.113.20 : 40112
(shared space, RFC 6598)   keeps the mapping table         (public, shared by many
                                                             subscribers at once)
```

The device holds an address the internet cannot route to. The translator replaces the source address and source [port](/glossary/port/) with one of its own public pairs, remembers the mapping, and reverses it for the reply. The destination sees only the right-hand pair and has no way to tell one subscriber behind it from another. A home router may perform a second translation before the carrier’s, so one connection can cross two NATs in a row; the subscriber controls the first and cannot touch the second.

RFC 6598 reserved `100.64.0.0/10`, 4,194,304 addresses, as Shared Address Space for the link between the translator and the customer’s equipment, because numbering that link from the RFC 1918 private ranges caused operational problems.

## Is my own connection behind it?

Compare the address on your interface with the address a remote service reports back. The interface address alone tells you most of what you need:

| Address on your interface | Defined by | What it suggests |
| --- | --- | --- |
| 100.64.0.0/10 | RFC 6598 | Carrier-grade NAT, almost certainly; the range exists for nothing else |
| 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16 | RFC 1918 | A NAT somewhere upstream: your own router, the carrier’s, or both |
| A public address that matches what remote services report | Registry allocation | No translation on this path |
| A public address that differs from what remote services report | Registry allocation | Something upstream rewrites it: a proxy, a VPN, or a NAT you do not control |

A `traceroute` whose first hops beyond your router show `100.64.x.x` addresses points the same way. So does being unable to accept any inbound connection without the carrier’s help, because a listener behind the translator has no public pair for anyone to reach.

If you classify addresses in code, Python’s `ipaddress` treats this range as neither private nor global. On Python 3.15.0, `ipaddress.ip_address("100.64.0.1")` returns `is_private False` and `is_global False`, while `10.1.2.3` returns `is_private True`. The documentation states the exception explicitly. A check written as `if not addr.is_private` will treat a CGNAT address as public.

## What it changes for proxies

Blocking a shared address hits every subscriber behind it, so targets are reluctant to do it and more likely to reach for softer measures. That reluctance is the asset you pay for with a [mobile proxy](/glossary/mobile-proxy/); nothing in the proxy software produces it. The [guide to proxy servers](/guides/what-is-a-proxy-server/#the-four-types-and-why-the-distinction-exists) puts it the same way, and [residential vs datacenter vs ISP vs mobile](/guides/residential-vs-datacenter-proxies/#the-decision-procedure) covers when that asset is worth its price.

The same sharing cuts the other way. A target that counts requests per address puts you in a queue with every stranger on that public address, so a [rate limit](/glossary/rate-limiting/) can be reached before you send anything unusual. Two requirements in RFC 6888 shape the rest:

- **Paired pooling (REQ-2).** A CGN must by default map all of one subscriber’s sessions to the same external address. Connections from one device therefore leave from one public address while its mappings last, which is why a mobile exit keeps its address the way a [sticky session](/glossary/sticky-session/) does, until the mappings expire or the carrier reassigns it.

- **Per-subscriber port limits (REQ-4).** A CGN must support capping the external ports each subscriber can hold. The cap is the operator’s setting and is not published to you, so a single mobile device has a ceiling on simultaneous connections that you discover by hitting it.

## What it does not hide

Sharing an address conceals the subscriber from the destination, not from the carrier. To answer abuse reports, RFC 6888 lists what an operator would need to log for each mapping: the transport protocol, a subscriber identifier, the external address, the external port and a timestamp. RFC 7422 describes allocating fixed port ranges per subscriber so that far less needs logging at all. Either way the carrier can map a public address and port at a given moment back to one line. Attribution on a shared address needs the source port and the time, not just the address, which is worth knowing if you ever have to read an abuse complaint.

## Frequently asked questions

### How can I tell whether my connection is behind CGNAT?

Look at the address on your own interface. An address in 100.64.0.0/10 is shared address space reserved for carrier-grade NAT. A private RFC 1918 address means some NAT is upstream, which may be your router, the carrier, or both. Then compare with the address a remote service reports: if your router's WAN side holds a private or 100.64 address, the carrier is translating.

### Can I port forward or host a server behind CGNAT?

Not through ordinary router settings, because the public address and port belong to the carrier's translator, not to you. RFC 6888 requires CGNs to implement a protocol that lets subscribers request mappings explicitly, but whether your carrier enables it is up to them. A tunnel to a host with a public address is the usual workaround.

### Does CGNAT hide me from my mobile carrier?

No. The carrier performs the translation and can map an external address and port at a given time back to a subscriber. Sharing an address hides you from the destination, not from the network doing the sharing.

### Why are mobile proxies harder to block than datacenter proxies?

Because the visible address is shared by many real subscribers. Blocking it removes paying customers of the target along with you, so targets weigh that cost and often choose softer measures. The protection comes from that collateral damage, not from anything the proxy itself does.

## Sources

1. [RFC 6598: IANA-Reserved IPv4 Prefix for Shared Address Space, 100.64.0.0/10](https://www.rfc-editor.org/rfc/rfc6598.html)
2. [RFC 6888: Common Requirements for Carrier-Grade NATs (CGNs), pooling, port limits and logging](https://www.rfc-editor.org/rfc/rfc6888.html)
3. [RFC 4787: NAT Behavioral Requirements for Unicast UDP, the definition of Paired pooling](https://www.rfc-editor.org/rfc/rfc4787.html)
4. [RFC 7422: Deterministic Address Mapping to Reduce Logging in Carrier-Grade NAT Deployments](https://www.rfc-editor.org/rfc/rfc7422.html)
5. [RFC 1918: Address Allocation for Private Internets](https://www.rfc-editor.org/rfc/rfc1918.html)
6. [Python documentation: ipaddress, the is_private exception for 100.64.0.0/10](https://docs.python.org/3/library/ipaddress.html)
