Learn how CGNAT (Carrier-Grade NAT) powers mobile proxies and why mobile IPs are so effective. Understand how to build your own proxy infrastructure using physical devices with SIM cards - the cleanest residential mobile connections available.
What you'll master:
Mobile proxies use real 4G/5G connections from physical devices with SIM cards - the same connections people use daily on their phones. Because mobile IPs come from carrier ASNs, blunt IP blocks risk collateral damageโso many platforms start with softer frictions. Detection still uses device/browser/behavior signals; CGNAT reduces IP-only friction, it doesn't make traffic un-blockable.
Our actual IP pool data shows the massive diversity available through carrier networks:
Customize proxy rotations based on your needs. Note: 10-20 second downtime during rotation.
High IP diversity with minimal repetition, maximizing your reach and reducing detection patterns.
Distinct IPv4s โ rotations ร devices ร days ร (1 โ collision_rate). Track collision_rate = repeated_IPs / total_rotations per carrier/APN/time-of-day; expect higher repeats at night or under heavy load.
Scenario | Est. Collision Rate | Distinct IPs/Month |
---|---|---|
20 modems ร 5-min rotation (288/day/modem) | 10-20% | โ138K-155K |
25 modems ร 5-min rotation | 15-25% | โ162K-184K |
20 modems ร 1-min rotation | 25-40% | โ518K-648K |
10-30 min rotation for natural behavior
5-10 min rotation for session stability
Rotate on challenge spike, session reset, or geo switch
1-2 min rotation for maximum diversity (use sparingly)
Note: 10 s downtime per rotation; scheduler verifies new /32 before resuming.Prefer event-based rotation (challenge spike, session reset, geo switch). Excess churn can raise friction and cut throughput.
Definition: "Distinct IP" = IPv4 seen exactly once in window. "Unique-enough IP" = seen โคN times (we use N=5). Behavior can vary by carrier/APN/time-of-day; we surface metrics in the dashboard.
Yes, in our current POPs per logs; dashboard shows live status.
Every 60 s (โ10 s downtime).
No, counts are distinct observations over time.
10-30 min for natural behavior patterns.
Mobile ranges aren't for direct mail; protects pool reputation.
Numbers sourced from production logs; endpoint /ip-pool-status