What Are Residential Proxies and How Do They Work?

A residential proxy is an intermediary server that sends your HTTP request out through an IP address that some internet service provider assigned to a real home. The website you are contacting sees a Comcast subscriber in Denver or a Telefónica subscriber in Madrid. It does not see your server, your office, or the cloud region you are actually running from.

Everything that people find confusing about residential proxies — the per-gigabyte pricing, the inconsistent latency, the arguments about ethics — comes from that one design choice. Once you understand what has to happen for a stranger's home connection to carry your traffic, the rest of the market makes sense.

What actually makes an IP "residential"

There is no flag in the IP packet that says "this is a house." The classification is inferred from public registry data. Every routable IP block belongs to an autonomous system, identified by an ASN, and the organisation behind each ASN is published in the regional internet registries — ARIN, RIPE, APNIC and the rest.

AS16509 is Amazon. AS14061 is DigitalOcean. AS7922 is Comcast Cable, AS3320 is Deutsche Telekom. When an anti-bot system wants to know whether a visitor is a person or a server, the first cheap check it runs is the ASN behind the connecting IP. A request from a hosting ASN carries an obvious signal that no amount of header spoofing will hide.

How a request travels through a residential proxy

Providers hide almost all of this behind a single gateway hostname, which makes the system feel simpler than it is. A typical request goes through five hops:

1. Your client opens a connection to the provider's gateway, usually one hostname and port for the whole network, and authenticates with a username that also encodes your targeting options.
2. The gateway parses those options — country, city, ASN, session ID — and picks an exit node from the pool that matches.
3. The chosen exit node is a device on a home connection running the provider's client software. It already holds an open connection back to the provider's infrastructure, because home routers sit behind NAT and cannot accept inbound connections.
4. Your request is relayed down that existing tunnel, and the home device performs the actual TCP connection to the target website.
5. The response travels back the same way. The target logs the household's IP; you never learn it unless the provider tells you.

# Targeting is normally expressed in the username, not in a separate API call
curl -x http://user-country-us-city-denver-session-a7f3c1:[email protected]:7000 \
     https://api.ipify.org?format=json

Hop three is where the latency comes from. A home upload link is asymmetric, often congested by whatever else the household is doing, and it may be a phone tethering over LTE. You are inheriting a consumer connection, so a p50 of a few hundred milliseconds is normal and a long tail is unavoidable. Our live latency rankings show how wide the spread between providers gets on the same test route.

Where the IP addresses come from

A provider cannot buy residential IPs the way it buys datacenter IPs. It has to convince households to route traffic. Four models dominate, and they produce pools with very different behaviour.

The bundled-SDK model is the reason residential proxies are ethically contested. When a flashlight app quietly turns a phone into an exit node, the household never meaningfully agreed to carry a stranger's scraping traffic. Serious providers publish their sourcing and let you audit it; the rest say "ethically sourced" and stop there. We cover the legal side in are residential proxies legal.

Rotating exits versus sticky sessions

Two rotation modes cover nearly every use case. In rotating mode the gateway hands you a different exit for every request, which suits stateless work like fetching a million product pages. In sticky mode you pin a session identifier and keep the same exit for a fixed window, usually somewhere between one and thirty minutes, which is what you need whenever a login cookie or a shopping cart has to survive more than one request.

Getting this wrong is the most common self-inflicted wound in scraping. Rotating mid-checkout invalidates the session; sticking to one IP for ten thousand requests gets that IP rate-limited. The full decision tree is in rotating vs sticky proxy sessions.

What residential proxies cost

Residential bandwidth is billed per gigabyte, and the headline rate falls steeply with commitment. Entry plans sit in the high single digits per GB. Mid-tier commitments land in the low single digits. Large contracts go below a dollar. Because the unit is bandwidth rather than requests, two teams making the same number of requests can see a tenfold difference in their bill depending on whether they download images and fonts.

Blocking media and third-party requests at the browser or client level is the highest-leverage cost control available, and it costs nothing to implement. We walk through the specific techniques in reduce proxy bandwidth costs, and normalise every provider to a single price per 100GB comparison so the tiers are actually comparable.

When you should not use residential proxies

Residential is a premium answer to one specific problem: the target treats hosting IPs as suspicious. If that is not your problem, you are paying a large multiple for nothing. Public APIs with documented rate limits, sites that never challenge server traffic, and internal endpoints all run fine on datacenter IPs at a fraction of the cost.

• Use datacenter when the target does not fingerprint the ASN, and you need throughput above everything else.
• Use ISP (static residential) when you need a residential-looking IP that stays the same for hours or days — account management, ad verification, long-lived sessions. See ISP proxies explained.
• Use mobile only when the target is a mobile app backend or tolerates nothing else, because carrier-grade NAT makes mobile IPs the hardest to ban. See mobile proxies explained.
• Use residential for consumer web targets that block server IPs but do not require a stable identity.

The full side-by-side, including cost per successful request rather than cost per GB, is in residential vs datacenter proxies.

How to evaluate a provider before you commit

Every provider markets on pool size and price. Neither predicts whether your job will finish. Four measurements do, and all four are things you can test during a trial:

1. Success rate on your targets, not on httpbin.org. A pool can be flawless against a neutral endpoint and useless against the site you care about. See proxy success rate explained.
2. Latency distribution, specifically p95 rather than the average, because the tail is what stalls a concurrent crawler. See how to test proxy speed.
3. IP reputation, measured as fraud score and clean rate across a large sample of exits. See proxy fraud score explained.
4. Effective cost, meaning price per gigabyte divided by success rate. A cheap pool that fails a third of its requests is not cheap.

We run exactly these measurements continuously against every provider we track, from both US and EU infrastructure, and publish the raw result on the live benchmark table. If you would rather run the tests yourself, the procedure is documented in how to benchmark proxy providers and in our methodology.