Where residential proxy IPs actually come from
Every few weeks someone in our comments asks the same thing. If a residential proxy costs a few dollars a gigabyte, who is giving up their home connection, and do they know? It’s a fair question, and the honest answer is that it depends entirely on the vendor. Two providers can sell you a product with the same label, the same dashboard and the same “real home IPs” claim. One built its pool from people who consented under a clear agreement. The other bought traffic from apps whose users never read a screen of legal text.
I run Singapore proxy and cloud-phone farms and manage large account fleets, so I sit on both sides of this. I buy and test pooled residential traffic for some jobs, and I run my own hardware for others. This article covers how residential proxies are sourced, how each source behaves, why that changes what you see in practice, and how detection systems read the result. It’s the long companion to the video, so expect mechanics and edge cases rather than a summary.
One caution up front. I’m not going to name vendors and rank them. I haven’t run a controlled test I could show you for every provider, and a verdict without a test is just an opinion. What I can give you is the structure of the market, so you can run the tests yourself and reach your own conclusions.
What “residential” actually means
“Residential” is a classification, not a physical fact. Detection vendors and IP intelligence databases label an address by looking at who announces it. Every IP sits inside a block registered to an organization, and that block is originated by an autonomous system, an ASN. If the ASN belongs to a consumer ISP that sells home broadband, the address is classed as residential. If it belongs to a hosting company, it’s datacenter. If it belongs to a mobile carrier, it’s mobile. That’s the whole taxonomy at the coarse level.
So when a vendor says an IP is residential, the claim is about the network it lives in. It says nothing about whether a person plugged it in for you. A router in a family flat, a phone on a train, and a rented server in an ISP’s own rack can all sit inside the same consumer ASN. To the database they look the same.
This matters because there are really two questions hiding in the word. The first is classification: will a detection system label the address as consumer? The second is provenance: how did traffic from that address end up available to a customer, and who agreed to it? Classification is what affects your accounts on Monday. Provenance is what affects how long the address stays usable, how many other customers share it, and whether you’re comfortable with the supply chain.
A few further properties ride along with the ASN. Consumer ISPs typically hand out dynamic addresses that change when the router reconnects. They often sit behind carrier-grade NAT, which means many subscribers share one public address. Reverse DNS on the address tends to look like a customer pool rather than a server name. Geolocation databases place it at city or region level, not at a street. All of that is expected of home traffic, and a good pool inherits it for free because the traffic really does leave from those networks.
Source one: SDKs embedded in apps
The largest share of pooled residential supply, as far as the industry describes itself, comes from software development kits. An SDK is a library an app developer drops into a mobile app, desktop program or game. The library runs a small background client that lets the proxy company route some customer traffic through the device’s connection. In return, the developer gets paid, typically per active user, and the app often removes ads or unlocks a feature for people who opt in.
Mechanically it’s simple. The device opens an outbound connection to the vendor’s gateway and keeps it alive. When a customer’s request arrives at the gateway, the gateway sends it down that connection, the device makes the request to the target site, and the response travels back the same way. The target sees the device’s home or mobile IP. Because the connection is outbound, no port forwarding or router setup is needed, which is why this model scales to millions of devices without anyone touching a router.
The quality of the arrangement depends on the consent layer. In the good version, the app shows a plain prompt: this app shares idle bandwidth, here’s what that means, here’s the button to decline, and the app still works if you do. In the bad version, the language is buried in a terms page, or the sharing is bundled with something the user can’t refuse. Security researchers have documented both patterns over the years, and a vendor’s blog post won’t tell you which one you’re buying.
What you can observe from the customer side is a specific texture. SDK pools are dominated by phones and laptops that are online in bursts. Devices come and go when apps are closed, batteries die, or people walk out of wifi range. That’s why sticky sessions from this source have a maximum lifetime and why the exit IP can vanish mid-session. Uptime per address is low. Volume across the pool is huge. Those two facts are the same fact seen from different ends.
Source two: extensions and free VPN apps
The second source is related but distinct: browser extensions and free VPN or utility apps that route other users’ traffic through your connection as the price of the free tier. The user installs something that looks like a VPN, an ad blocker or a coupon tool, and part of the deal is that their connection becomes an exit node for someone else.
Some of these disclose it clearly, describing a peer-to-peer network where you share bandwidth in exchange for free service. Others have been caught doing it quietly through updates. The pattern security teams describe is an extension that behaves normally for months, then ships a version that adds a proxy component. Because browser stores auto-update extensions, the user never sees a decision point.
Two failure modes matter to you as a customer. First, the traffic exits from whatever machine the user happens to be on, which may be a work laptop or a shared computer. Second, the IP’s reputation is shaped by everything else flowing through it. If the same exit served scraping, ticket bots and credential stuffing an hour ago, that address carries the consequences when your session arrives. You have no view of that history.
There’s a legal and ethical exposure here that I take seriously. If a pool includes people who never understood they were an exit node, then traffic you send appears to come from their household. If a site treats that traffic as abuse, the household is what gets blocked, and in the worst cases it’s what gets questioned. That’s a reason to prefer sources with explicit informed opt-in, and to ask vendors about it directly instead of treating it as someone else’s problem.
There’s a practical signal too. Pools that lean on extensions tend to skew toward desktop browsers and toward countries with heavy free VPN usage. If a vendor’s Singapore coverage comes mostly from this source, it will be thin, and you’ll see the same few addresses repeat across sessions.
Source three: payout apps
The third source is apps built specifically to pay people for sharing bandwidth. You install a small program on a home computer, a spare phone or a Raspberry Pi, leave it running, and it earns credits based on the traffic it carries. In one sense this is the most transparent of the pooled sources, because the whole product is the sharing. Nobody installs it by accident.
The supply behaves differently from SDK supply. Devices are usually on fixed home broadband, left on around the clock, and run by people who have thought about it. Uptime per address is much better, and many exits are wired residential lines rather than phones. The catch is that a portion of participants run many instances to multiply the payout, sometimes several per household, or on cloud servers with a spoofed identity. Vendors fight this with checks on ASN, latency and device fingerprints, and the fight never ends. A pool from this source can therefore contain addresses that are less residential than advertised.
I run a version of this idea myself, in effect. My own hardware exists to be an exit, but I’m the one who owns it, and the SIMs and lines are in my company’s name. That’s the cleanest provenance there is, and also the most expensive per gigabyte to build, which is why marketplaces exist at all.
For the person paying, the interesting property of this source is the spread. A typical payout-app pool has a wide quality range: some excellent, stable home lines, and a tail of questionable ones. You can’t tell from the dashboard which you got. You find out by measuring, and I’ll cover how in the worked example below.
Worth saying plainly: participant earnings are small, and I’m not describing this as a way to make money. It’s a description of how supply gets built.
Source four: ISP allocations and static residential
The fourth source is addresses that really are allocated by a consumer ISP but don’t come from a household. Sellers call these static residential or ISP proxies. The seller works with an ISP, or is an ISP customer, and gets blocks of consumer-classed addresses assigned to servers in a data center. The traffic runs on datacenter hardware with datacenter uptime, but the ASN in the lookup belongs to a consumer network.
This is why the label is slippery. The address is genuinely classed as residential by ASN. It isn’t a home. It doesn’t rotate when a router reboots, it has the bandwidth of a server, and its latency profile looks like a server’s. For a single long-lived account that needs one stable identity, that can be exactly what you want. For an account that’s supposed to look like a person on a phone, it’s a mismatch, and some detection stacks look for it.
What gives these addresses away, when anything does: the reverse DNS or WHOIS entry may show a business customer instead of a subscriber pool, neighboring addresses in the block are all proxies, and the address answers on ports that home routers don’t. Detection vendors keep lists of blocks that have been seen hosting proxies. Once a whole block is tagged, every address in it carries the tag regardless of how clean the ASN looks.
Static ISP addresses are also often shared by very few customers, which is a real benefit. One or two users per address means you control much of the reputation. I’d choose this source when stability matters more than blending into a crowd, and I’d avoid it for anything that’s supposed to move around like a person. Ask how many customers share each address and whether the blocks are dedicated to the seller.
Source five: hardware the operator owns
The fifth source, and the one I know best, is hardware the operator owns. A residential farm is real connections in real premises with real routers. A mobile farm is modems or phones with carrier SIMs, connected to a host machine that exposes each one as a proxy port.
Mobile deserves its own explanation because the IP behavior is different. Mobile carriers use carrier-grade NAT, so hundreds or thousands of subscribers share a small set of public addresses. When you rotate a mobile modem, it detaches and reattaches, and the carrier hands it another address from a shared range. Because so many real people already sit behind each address, platforms are reluctant to block one outright, and that’s the property people pay for.
Running it is not glamorous. You need USB hubs with real power delivery, because a bus that browns out drops a dozen modems at once. You need SIMs on plans that allow the usage, and you need to read the terms, since many consumer plans forbid resale. In Singapore there are also licensing questions about reselling telecom services, which is a conversation to have with a lawyer before you scale, not after. Modems need monitoring for stuck sessions, carrier throttling and repeated IPs. Rotation isn’t random: you often see the same few addresses cycling back.
The advantage is provenance. I know whose line it is, what it’s used for and who else shares it. The disadvantage is cost and labor, and it doesn’t scale like an SDK does. A farm of a few hundred ports is a serious operation. A pooled vendor advertises tens of millions of addresses. Those numbers aren’t comparable, because one is ports I can count and the other is devices seen over some period, counted by a method the vendor chooses.
Consent and the grey zone
Let’s talk about the uncomfortable part. Not every residential IP in the market arrived by consent, and I’d be misleading you if I implied otherwise. There are three tiers worth separating.
The first is informed opt-in: the person knew, agreed, and can leave. Payout apps and clearly disclosed SDKs live here when they’re done properly. The second is technical consent without understanding: a checkbox in a long agreement, or a feature bundled with a free app. That’s defensible in many jurisdictions and ethically thin. The third is no consent at all: devices compromised by malware, routers with default passwords enrolled into a network, or apps that do something other than what they claim. Law enforcement and security firms have taken down proxy networks built this way, and some of them sold access through storefronts that looked like ordinary residential proxy shops.
I don’t touch anything in the third tier, and I’d tell you not to buy from it. Beyond the ethics, there’s a practical problem. An address on a compromised device is an address that will be cleaned, taken down, or investigated. Access can disappear overnight, and if your accounts are tied to it, you inherit the disruption.
How do you tell the tiers apart from outside? Often you can’t, with certainty. What you can do is look for signals. Does the vendor publish how its supply is sourced, in specific terms? Do they describe an opt-in flow you can actually see? Do they run identity checks on customers? Vendors who know their supply is in good shape tend to say so in detail. Vendors with something to hide say “ethically sourced” and stop talking.
I’m deliberately not labelling any provider here. To call one clean or dirty I’d need evidence I can show you, and I don’t have that for the market as a whole.
From pool to product: gateways and rotation
Whatever the source, the customer usually doesn’t connect to a device. You connect to a gateway, and the gateway picks a device. This is called a backconnect setup. You point your browser or antidetect profile at one hostname and port with credentials, and the vendor’s system maps your session to an exit.
Credentials often carry the instructions. A username string might include a country, a city, a session ID and a duration. The same session ID means the same exit for as long as that device stays online. A new session ID means a new exit. Without a session ID, many gateways rotate on every request, which is useless for a logged-in account, since the site sees your login from one address and your next click from another country.
Two behaviors matter for account work. First, sticky sessions aren’t guaranteed. The vendor promises “up to” some number of minutes, and if the underlying device disconnects, you’re moved without warning. From the site’s perspective, your session just changed IP mid-login. Antidetect browsers can’t fix that, because the change happens below them.
Second, geography is a filter applied to a pool, not a guarantee about the pool. Asking for a country narrows the candidates. In a small market like Singapore, the candidates are few, and repeated sessions land on repeated addresses. The usable pool ends up smaller than the headline number by orders of magnitude.
This is also where concurrency creeps in. The gateway may hand the same exit to two customers at different times, and sometimes at the same time. You don’t see that, and neither do they.
None of this is a defect. It’s the shape of the product. It just means you should design around it: use static or long-sticky exits for anything with a login, and keep rotating pools for tasks where identity doesn’t persist.
What the source means for quality
Now connect source to what you observe. I’ll treat these as tendencies, because I’m not going to present numbers I haven’t measured.
SDK pools offer enormous breadth, high churn, mixed device types and often strong geographic diversity. Expect sessions to drop more often, and expect to see phones and laptops in odd time zones. Reputation varies, because the same addresses serve many customers doing many different things.
Extension and free VPN pools skew toward desktop, skew toward countries with heavy free VPN use, and carry reputation risk from unknown activity on the same device.
Payout-app pools have better uptime, a wired home mix, and a tail of fake or multi-instance nodes.
Static ISP addresses give server stability, a consumer ASN, very few users per address, and block-level exposure if the block is known.
Owned hardware has known provenance, small scale, and carrier-specific quirks like repeating addresses.
Reputation deserves a closer look. An IP reputation score is built from what’s been seen coming from the address: spam, login attempts, scraping, payment failures. A residential address shared across a pool’s customers inherits a blend of all of them. That blend is why two accounts created on the “same quality” of proxy can have very different fates. It has nothing to do with the browser fingerprint and everything to do with who was there before you.
There’s a subtler cost too. Pools that recycle addresses quickly mean an address you used yesterday might be handed to someone else today, and your warmed account may land on a stranger’s history the next time it logs in. This is why serious account work pins an exit and keeps it, rather than sampling from a pool each time.
How detection reads an IP
Detection looks at more than the IP’s label, and this is where the source shows through. Here’s what a reasonably capable stack examines.
ASN and IP type come first. Datacenter ASNs are treated with suspicion for signups and logins. Consumer and mobile ASNs get more benefit of the doubt.
Reputation history is next. Commercial feeds track addresses seen in abuse. Proxy detection vendors go further and maintain lists of addresses known to belong to proxy networks, learned by operating as customers, by scanning, and by correlating traffic across many sites. A residential address that appears in one of those lists is tagged even though its ASN is clean.
Then there’s network behavior. Latency and timing can reveal a relay. If the IP geolocates to Singapore but the round trip behaves like a path through another continent, that’s a hint. Open ports and banners on the address can reveal a proxy server sitting there.
Consistency across signals matters just as much. The browser’s timezone, language and WebRTC-exposed addresses should agree with the IP. DNS requests should resolve through the same place. A Singapore IP with a browser reporting a European time zone and a resolver in another country is a mismatch that costs trust, however good the IP is.
Finally, session continuity. A logged-in account that changes country between two requests looks like a hijack. A stable address, even a datacenter-classed one, can be less alarming than a rotating residential one.
What detection can’t see directly is provenance. It can’t read the consent form. It sees the symptoms of a weak supply chain: churn, crowding, tagged blocks, odd latency. That’s why the source matters to you even when the classification looks perfect. I’m keeping this at the level of concepts, because the point is defensive: knowing what a clean stack has to line up.
Worked example: evaluating a Singapore pool
Here’s how I’d evaluate a pooled residential product for Singapore, as a method rather than a set of results. I won’t report numbers, because I only do that for tests I ran and can show, and the video demonstrates the process, not a verdict on a specific vendor.
Step one is to sample. Buy a small allowance and request sessions with the country set to Singapore. Log each exit IP over a few hundred requests, across a few days and different times, against a neutral IP echo service.
Step two is to measure repetition. Count unique addresses against total sessions. If you drew three hundred sessions and saw sixty unique addresses, the effective pool is small regardless of the advertised size. Repetition tells you how crowded the exits are.
Step three is to classify. Run each unique address through two or three IP intelligence sources and record the ASN, the type, and any proxy or VPN flags. Note what share is consumer or mobile and what share is hosting or flagged. Different databases disagree, and the disagreement is itself data, because an address one source flags and another doesn’t is one a detection vendor may also flag tomorrow.
Step four is to check consistency. For a handful of addresses, check reverse DNS, geolocation against your target city, and round-trip time from a machine in Singapore. A Singapore address with a long round trip from a Singapore machine deserves a question.
Step five is to check stickiness. Request a session with a ten minute duration and poll the exit IP every thirty seconds. Note when it changes. Repeat several times. That gives you the real distribution, not the marketing one.
Step six is to test in your stack. Load the exits into an antidetect profile and confirm that WebRTC, DNS and timezone line up before any account touches it.
Keep the log. A month later, run it again. Pools drift, and a good pool in March can be a crowded one by September.
Questions to ask a vendor
Before you buy, ask questions a well-run supplier can answer specifically. Vague answers are information.
Where does the supply come from? A useful answer names categories: SDK partners, a peer app with its own users, ISP allocations, owned hardware. “Real users worldwide” isn’t an answer.
How is consent collected, and can I see the screen? A vendor with a proper opt-in can show it. Ask what happens if a user declines, and whether the app still works.
How many unique addresses in Singapore did you see last month, and how do you count? Watch the definitions. Some count every address ever seen, others count addresses online right now, and the two differ by a lot.
What’s the maximum sticky session, and what happens on disconnect? If the answer is that the exit may change, plan for it.
Do you verify customers, and what’s prohibited? A vendor that lets anyone buy anything will have a pool that carries the reputation of anyone.
Do you allow mixed use on the same exits? Shared exits mean shared history.
What do the terms say about the pool itself? Read the acceptable use policy. Many forbid things that account work brushes against, and you should know that before you’re mid-project.
What do I get if the exit is flagged on day one? The refund and replacement policy shows how much the vendor trusts its own pool.
And ask yourself one too: what’s the account worth? If it’s a throwaway, a rotating pooled exit might be fine. If it’s a client asset or a year of warm-up, a pinned dedicated exit with known provenance is cheap insurance against what you can’t see.
I can’t tell you which vendors will answer well, because I haven’t tested them all. I can tell you these questions separate the serious ones from the rest quickly.
What to change on Monday
Here’s what I’d do differently, in order.
First, stop treating “residential” as a quality grade. It’s a classification. Write down what you need from the IP: stability, blending, geography, or throwaway breadth. Then pick the source that gives you that. Stability points to static ISP or owned hardware. Blending points to mobile or a good home pool. Breadth points to SDK pools.
Second, pin exits for anything with a login. One account, one exit, kept for as long as the account lives. Don’t sample a rotating pool for warmed accounts.
Third, run the measurement routine from the worked example on any pool before it touches an account, and again monthly. Keep the logs.
Fourth, line up the rest of the stack with the IP. Timezone, language, DNS, WebRTC and the device profile should describe the same person in the same place. A perfect IP doesn’t rescue an inconsistent fingerprint.
Fifth, separate accounts by exit so a burned address doesn’t take its siblings with it. No two accounts that matter should share one exit.
Sixth, read the terms of everything: the proxy vendor, the carrier plan, the platform. Your real constraint is the strictest of them.
And the limits. Nothing here guarantees that accounts survive. Platforms change detection, vendors change supply, and a clean IP is one input among many. I’m describing how to cut avoidable mistakes, not promising an outcome.
If you want to watch these tests run on screen, including the repetition count and the sticky polling, that’s what the video covers. The process is more useful than any verdict, because it still works after the vendor landscape shifts.
If this kind of breakdown is useful, there’s more on antidetect browsers, proxies, cloud phones, account isolation and warm-up on the Multi Account Ops home page, with new videos and articles as we finish testing them.
Get new guides and videos first — join the Telegram channel.