Residential, datacenter and mobile proxies compared

Here is the answer before the reasoning.

I priced one job across all three last month: about $2 on datacenter, about $60 on residential, about $480 on mobile. Same target, same data, same month. Every one of them returned the job.

I sell the tier on the right. Price everything below accordingly.

The registration is the whole first filter

Every block of addresses on the internet is registered to an organisation, and that registration is public. One lookup returns the holder of the block and the network number it sits in.

That is the entire first stage of tier classification. It costs nothing, it runs before your request body is read, and the result stays cached.

Hosting company on the registration means datacenter. Consumer ISP means somebody’s house. Mobile carrier means a phone.

There is no machine learning in that. It is a database join a junior developer could wire up in an afternoon, which is exactly why it is the cheapest check any site runs and therefore the first one.

Everything people attribute to a proxy being sophisticated is really this one lookup returning a different answer.

Contention, which nobody puts on the pricing page

Datacenter: you either hold a dedicated address or you sit in a shared pool with an undisclosed number of other buyers. The cheap plans are the shared ones. You inherit the address’s history, and I have been handed pool addresses that were already refused by my target on the first request, for things somebody else did before I paid.

Residential: the address belongs to one household. Whether the provider is running three other customers through that same household at the same moment is not something you get told, and asking usually produces a marketing answer.

Mobile: a carrier does not give every phone a public address of its own. Hundreds or thousands of subscribers sit behind one, through carrier grade NAT. Your scraper shares an exit with people checking scores and messaging their families.

Refusing that address costs the site real customers. That is the mechanism, and it is a fact about the carrier’s network design rather than anything the proxy seller built.

The unit you are billed in decides the winner

Datacenter and mobile are both sold per address per month. Residential is sold per gigabyte. Those are not comparable prices until you know how many bytes your job moves, and most people never look.

A crawl I ran recently: 12,000 product pages. Rendered in full, with images and fonts and the tracking scripts, those pages averaged 2.8 MB. The JSON endpoint sitting behind them returned the same fields in 31 KB.

Rendered, that job moves about 33 GB. Against the endpoint, it moves about 0.4 GB.

At $4 a gigabyte, that is $134 versus $1.50. Identical data, identical row count, an eighty times difference in the bill, decided entirely by which URL I pointed at.

So the residential decision is really a question about your own code. Measure your average response size before you argue with anyone about per gigabyte rates.

Datacenter is right more often than I would like

The tier everybody uses and nobody puts in their case study.

Its weakness is real: one lookup classifies it, and any site that wants to refuse hosting ranges can do so with a published list and a few lines of config.

Its strength is that a very large share of collection work happens against targets where nobody is checking anything. Public data portals. Government statistics. Documentation. APIs you hold a key for. Small business sites. Your own infrastructure.

I point my uptime monitoring at datacenter addresses. I own a rack of mobile lines and I do not spend them finding out whether my own pages loaded.

There is a performance argument as well. A datacenter address is a machine on a fat link in a well run facility. Lower latency, more throughput, and far fewer intermittent oddities than any consumer connection will give you. When the tier works, it also works better.

The residential question worth asking your supplier

Residential exits come from consumer connections, which means they come from people.

The supply chain varies. Some pools pay households directly. Some are built from free apps that ship a routing SDK inside them, or browser extensions, or a free VPN funded by selling the exit out the back. Consent in those cases lives in an agreement nobody reads.

I am not accusing any particular supplier. I am saying it is a reasonable question to put to yours, and the shape of the answer is informative on its own:

• where do the exit nodes come from
• what does the person on the other end receive for it
• can they see that it is happening, and can they stop it

A supplier with a clean sourcing story answers in one message and often has a page about it already. If the question makes your account manager go quiet, you have learned something about whose home connection your traffic is running through, and you can decide what that is worth to you.

What a mobile line actually is

Mine cost about $10 a month in carrier data and roughly $1.50 in modem depreciation. Spread the $150 powered hub across the thirty or so ports it feeds, add the $200 box driving them, amortise over three years, and it adds something like 30 cents a month per port.

So my floor is about $12 a line before anything else, and I run 70 to 75 percent gross margin on top of that.

That arithmetic is the honest explanation of the price. There is a physical SIM and a physical modem behind every line, drawing power on a shelf. One line is one address at a time. Forty concurrent identities means forty SIMs and forty modems and enough hub capacity to keep them awake, and the cost climbs in a straight line with no volume trick that bends it much.

What the money buys is the shared carrier address and the crowd of ordinary subscribers on it. What it does not buy is immunity. No address type is unblockable, and anyone using that word is selling a story. I have watched a mobile line get shut out in about ninety seconds because of what the software behind it was doing.

Mobile earns its price on work that needs a small number of durable identities that stay put, rather than a wide pool of disposable ones. That is a narrower category of job than my order book would suggest.

The rule

Start at the bottom. Send a hundred requests through a datacenter address. If the data comes back, you are finished, and the whole investigation cost about two dollars.

If it does not come back, work out which part failed before you upgrade, because the address is one candidate among several and it happens to be the one with a payment link attached.

If you do have evidence that the tier is the problem, meaning the same request succeeds from an ordinary home connection and fails from the hosting address, move up exactly one step. Residential before mobile.

The common failure is skipping the middle. People go from $2 straight to $480 and never discover that $60 would have done it.

What I got wrong

For about a year I recommended mobile as the default. Someone would describe a job, I would say mobile lines, they would buy, and I would invoice. A fair number of them did not need it, and I know because several came back months later and mentioned they had quietly moved half their work back down and nothing changed.

I had one thing in stock and a story that made it the answer to every question. The story was true. It was being applied to jobs it had no business being applied to.

I also priced the wrong dimension once, on a small residential package I bought for a job I was certain was tiny. It was tiny in requests. The target rendered every page in full, and the allowance was gone inside a day.

What no comparison can tell you

I cannot tell you what a specific site does with the registration data it collects. I see how requests from each tier land, and I infer, and I am wrong often enough to say so out loud.

I also cannot rank two providers inside the same tier from the outside. Two residential pools at the same advertised rate can be sourced completely differently and behave completely differently, and the only way to find out is to buy a small amount from each and measure.

None of this is a guarantee about results either. No tier promises a clean run on any target.

The full pricing worksheet, and the real numbers behind the lines I run, are here.

Get new guides and videos first — join the Telegram channel.