Proxy pricing explained: per GB, per IP, and the cost per page that matters

Proxy pricing pages are written in three different units, and most of them are designed so you never convert one into another. One provider quotes per gigabyte, the next per IP per month, a third sells a monthly plan with an allowance that vanishes on renewal. All three look cheap on their own page.

This post walks through each model, what it rewards and what it punishes, and then turns all of them into the one number that lets you compare: what a page that actually loaded cost you.

How much do proxies cost?

It depends on the unit, and the unit depends on the proxy type. Rotating residential proxies are usually sold by the gigabyte. Static ISP and datacenter proxies are usually rented by the IP for a fixed term. Here, residential is $5.50/GB for a single gigabyte, falling to $1.75/GB from 1,000 GB. A datacenter or ISP IP for 30 days starts at $2.10.

Those numbers cannot be compared with each other, and a provider’s headline cannot be compared with another provider’s headline in a different unit. The rest of this post is about getting both into the same currency: money per page.

Per GB pricing: how residential proxies are billed

On a per-gigabyte plan you pay for traffic, not for addresses. You get a gateway, and every new connection through it comes out of a different home IP from the pool. Use one IP or ten thousand, the bill is the same; the bytes are what count.

Which bytes is the first question to ask any per-GB provider. Our meter counts request bytes plus response bytes, headers included, as our terms say. Some providers count less, some count more, and a few do not say. On small pages the headers are a real share of the total, so the rule matters more than the headline rate.

The second question is how the rate falls with volume. Ours is a ladder set by the size of each order:

• 1 GB at $5.50/GB
• 10 GB and up at $3.95/GB
• 100 GB and up at $2.65/GB
• 1,000 GB and up at $1.75/GB

It is pay as you go: the minimum top-up is $5, the balance does not expire, and nothing renews behind your back. What you give up on our residential line is choice of place. The pool does not let you pick a country, so if you need to appear from somewhere specific, that is a job for a static IP.

Per IP pricing: how ISP and datacenter proxies are billed

On a per-IP plan you rent addresses for a term, and the country is picked at checkout. The price is fixed however hard you use the address, within whatever traffic the plan includes. Our ISP and datacenter lines use the same plans and the same prices, so the figures below cover both.

• Shared, $2.10 for 30 days. Up to 3 customers use the address at once, and it includes 1 GB of traffic. Beyond that you pay $0.35/GB if you add it with the order or $0.40/GB if you add it later.
• Dedicated, $3.20 for 30 days. The address is yours alone and has no traffic meter.
• Premium, $4.80 for 30 days. Dedicated, no traffic meter, plus a fresh IP nobody has used before and the option to swap the IP yourself.

Longer terms cost less per day: 60 days takes 5% off, 90 days takes 10% off, 365 days takes 20% off. Renting more addresses at once lowers the price per IP too. The catch with any per-IP plan is the opposite of the per-GB catch: you pay the same whether the IP works on your target or not, so a blocked address is money spent on nothing.

Monthly subscriptions and gigabytes that expire

A lot of the market sells residential traffic as a monthly plan: a fixed fee buys an allowance of gigabytes for the month, and whatever you do not use is gone at renewal. The headline rate per gigabyte on those plans assumes you use every one of them.

You rarely do. Divide the plan fee by the gigabytes you actually used, not the ones you bought. Use half the allowance and you paid double the advertised rate. Use a quarter and you paid four times it. Before signing up, check three things: whether unused traffic rolls over, whether the plan renews automatically, and whether you can buy a smaller top-up when you run out, or have to jump to the next plan up.

Pay as you go avoids the whole question, which is why we sell residential that way. The trade is that a subscription’s headline rate can look lower than a pay-as-you-go one. It is lower only if your usage is steady enough to spend the allowance every month.

What do “unlimited” proxy plans actually limit?

Something. Every plan has a cost to the provider, and an offer that names no limit has moved it somewhere you have to look for. Read the plan and the terms for these:

• What is unlimited. Bandwidth, IPs, and concurrent connections are three different things. An offer can be generous on one and tight on the other two.
• Concurrency caps. A cap on simultaneous connections limits how many pages you can fetch in a month just as surely as a gigabyte meter does.
• Port speed. A throttled connection is a meter measured in time instead of bytes.
• Fair-use clauses. “Unlimited, subject to fair use” means there is a limit and the provider decides where it is.
• Shared addresses. Unlimited traffic on an IP shared with strangers is unlimited traffic on an IP that may already be flagged.

For the record, ours: dedicated and premium ISP and datacenter IPs have no traffic meter. More threads are a paid add-on, and we say so on the product page instead of in a footnote.

Per GB vs per IP proxies: which is cheaper?

Per IP, whenever the target accepts it and you move more than a little traffic. One dedicated datacenter IP for 30 days costs the same as 0.58 GB of residential at the single-gigabyte rate. Past that, the flat price wins.

The catch is the “whenever the target accepts it”. A datacenter address that gets 403s on every page has a cost per page of infinity, and no flat price fixes that. Whether yours is accepted is a ten-minute test, not a guess: the residential vs datacenter test answers it for rotating versus static, and ISP vs datacenter answers it for the two static lines.

How to work out proxy cost per request, and per page

This is the number worth comparing. Two formulas, one per billing model:

• Per GB: cost per successful page = (price per GB × GB per page) ÷ success rate. GB per page means request plus response, measured on your target; success rate is the share of requests that came back with the page you wanted.
• Per IP: cost per successful page = what the addresses cost for the month ÷ successful pages they fetched in it.

The success rate is where cheap offers go to die. A lower rate per gigabyte with half the requests blocked costs more per page than a higher rate where they all load. Retries make it worse: here, every retry you make is billed as its own request, and most per-GB meters work the same way.

Cost per request is the same idea with the success rate left out. On a per-GB plan it is simply bytes times rate, which is why small API calls are cheap per request but expensive per useful byte: the headers can weigh as much as the answer. On a per-IP plan the extra request costs nothing until the site starts rate-limiting the address.

A worked example, with an assumed page weight

The page weight and success rate here are assumptions for illustration, not typical values. Measure your own: the bandwidth measuring guide has a script that does it from your own connection for free, and the bandwidth calculator does the multiplication.

Say each page weighs 250 KB on the wire, 90% of requests succeed, and you need 20,000 successful pages a month. That is 5.56 GB of traffic including the failures, so the smallest residential order is 6 GB at $4.60/GB, or $27.60. Per successful page that is $1.28 per thousand.

If the same site accepts a dedicated datacenter IP, the month costs $3.20 and the same 20,000 pages come to $0.16 per thousand. If it does not accept it, residential is the only column that has a number in it.

Cost per 1,000 pages at different page weights

Same arithmetic, a few page weights. The per-IP columns assume one address fetches 50,000 pages in 30 days, about one every 52 seconds, and that the target accepts it. Every page is assumed to succeed; divide by your own success rate.

A few things fall out of the table. The dedicated IP does not care how heavy the page is, because nothing is metered. The shared IP starts cheap and climbs once the month outgrows its allowance. Residential is priced per byte, so the page weight decides everything, which is why measuring it first is the cheapest thing you will do all week. And at the light end, a small JSON response, residential beats both static IPs outright: 50,000 pages a month is not enough traffic for a flat price to pay off. You still buy at least the smallest order, 1 GB, but whatever the month does not use stays on your balance, and the balance does not expire.

The per-IP columns lean on their assumption as hard as residential leans on page weight. If one address can only fetch half as many pages before the site starts rate-limiting it, its cost per 1,000 pages doubles and you need a second address; the residential columns do not move, because they never depended on how hard one IP was worked. So the test on your target has two outputs, not one: the success rate, and how many pages an hour one address gets before the 429s start. Both go into the sum.

Costs that are not on the pricing page

• Retries and redirects. Each is its own request and its own bytes. Stop retrying 403s; they rarely change on the second ask.
• Browsers. A page loaded in a headless browser pulls scripts, fonts and images through the proxy. Block what you do not parse, or measure the browser rather than the HTML.
• Expiry and minimums. A large minimum purchase or an allowance that resets turns unused traffic into a fee.
• The meter itself. If you cannot check what you were billed for, you cannot check the price either. The bill audit guide shows how to reconcile a usage export against your own logs.

How to compare two proxy offers

1. Write down the unit each one bills in, and what its meter counts.
2. Measure your page weight on the real target.
3. Run a small test on the cheapest option and record the success rate.
4. Turn each offer into cost per 1,000 successful pages at your real volume, including expiry and minimum purchase.
5. Buy the smallest amount of the winner and measure again.

Our own prices are on the pricing page, and the 365-day discount of 20% only makes sense after step 5. If you would rather see this arithmetic run across the market, the cheap proxy providers roundup does it provider by provider.