What Is Proxy Pricing? A Guide to How Proxies Are Billed

The disclosure, since this is a pricing article written by a vendor: we are Geonode and we sell proxies, mostly priced by traffic. That biases us, and the honest way to handle it is to be specific about where traffic pricing is the wrong model and you should buy from someone else. That section is below and it is not decorative — if your workload is high volume through few addresses, or seasonal and lumpy, our pricing model will cost you more than the alternative and we would rather you knew that before the invoice than after. The arithmetic in this article works the same regardless of who you buy from; run it on us too.

The Four Pricing Models

The first two account for most of the market, and they are not variations on a theme. They are different products with different economics, and a workload that is cheap under one can be ruinous under the other.

Per-request pricing is worth flagging separately because it is often the right answer and is rarely compared against the others. A managed scraping or SERP API bills per successful result and absorbs the proxy cost, the browser rendering, the retries and the maintenance. It looks expensive per unit and is frequently cheaper in total, because the units you are comparing are not equivalent — one includes an engineer's ongoing time and the other does not. We went into this trade-off for one specific case in our comparison of rank tracker APIs.

Per-GB and Per-IP: The Two That Matter

Per-gigabyte pricing charges for data through the proxy, with no limit on how many addresses you use. Residential and mobile pools are almost always sold this way, and the reason is structural: the provider's own cost is bandwidth on somebody else's connection, so bandwidth is what they meter.

The consequence people miss is that page weight becomes a line item. A text-heavy page might be 50 KB; the same page rendered in a headless browser with images, fonts and tracking scripts might be 3 MB. That is a sixtyfold difference in what you pay for the identical piece of information. On per-GB pricing, blocking images and unnecessary resources is not an optimisation, it is the difference between a viable budget and an unviable one.

Per-IP pricing charges for addresses reserved for a period, usually monthly, with unlimited or very generous traffic through them. Datacentre and ISP proxies are typically sold this way, because the provider's cost is the address and the server, not the bytes.

The consequence here is the mirror image: volume is free and addresses are expensive. If ten addresses can carry your whole workload, per-IP pricing means your cost is fixed regardless of how hard you push them. If you need five hundred distinct addresses to avoid rate limits, the model collapses.

There is one important market anomaly worth knowing about, and it is ours, so treat it accordingly: datacentre proxies priced by traffic rather than per IP. Most of the market sells datacentre access per address. Selling it by the gigabyte inverts the usual trade-off — you get unlimited addresses and pay for what you move, which suits workloads that are spread thin across many endpoints and penalises workloads that are concentrated. Whether that is good news depends entirely on the shape of your usage, which is the theme of this whole article.

Unlimited Plans and What They Really Mean

"Unlimited" in this industry means something specific and it is worth reading the terms.

Unlimited traffic, limited concurrency. The most common structure. You can move as much data as you like, but only through a capped number of simultaneous connections. That cap is the real product. A plan with unlimited traffic and 50 threads has a hard throughput ceiling regardless of what the headline says.

Unlimited traffic, fair-use policy. Less common and requires more care. Somewhere in the terms is a threshold beyond which the provider reserves the right to throttle or renegotiate. If the threshold is not written down, you are buying a number the vendor can change.

Unlimited within a pool subset. Occasionally "unlimited" applies to a smaller or lower-quality slice of the network. Read which pool you get.

The arithmetic for when unlimited makes sense is straightforward. Take the monthly flat rate, divide by your best per-GB rate, and you have the break-even volume. At $1,800/month against $0.27/GB, that is roughly 6.7 TB. Below that, metered is cheaper. Above it, unlimited wins and the margin grows.

The trap is buying unlimited for predictability rather than for volume. Predictability is worth something, but if you are moving 500 GB a month, paying for 6,700 GB of headroom to avoid an unpredictable invoice is an expensive form of budgeting. Set a spend cap on a metered plan instead.

Working Out Which Model Fits Your Workload

Two numbers decide it: how much data you move, and how many distinct addresses you need. Estimate both before looking at any price list.

Estimating data. Take a representative request, measure the response size, multiply by your request count, and then multiply by a fudge factor for retries and overhead. If you are using a headless browser, measure the browser's actual traffic rather than the HTML size — the difference is routinely an order of magnitude and it is where budgets go wrong.

50,000 requests/day x 400 KB average = 20 GB/day = ~600 GB/month

Estimating addresses. This one people guess at, and the guess is usually far too high. The right method is empirical: run your workload from a single address and find the request rate at which you start being limited. That number, divided into your required throughput, is roughly how many addresses you need. Most people discover they need tens, not thousands.

Then compare:

Two worked examples make the divergence concrete.

Workload A: 600 GB/month spread across hundreds of addresses. On per-GB pricing at $0.50/GB, that is $300. On per-IP pricing, several hundred addresses at even $1.25 each is more, before you have moved a byte. Per-GB wins clearly.

Workload B: 600 GB/month through ten addresses against one target. On per-GB at $0.50, still $300. On per-IP, ten ISP addresses at $1.25 is $12.50 with traffic included. Per-IP wins by a factor of twenty-four.

Same volume. Same monthly total under one model. A twenty-four-fold difference under the other. That is why "which provider is cheapest" is an unanswerable question without knowing the shape of your usage, and why any comparison table of headline per-GB rates is close to meaningless on its own.

The Costs That Are Not on the Price List

Budgets fail on these more often than on the rate.

Retries. A 20% failure rate means you pay for 20% more traffic than you receive useful data from. On aggressive targets, failure rates are considerably higher and the entire retry volume is billable.

Browser rendering. Covered above and worth repeating because it dominates. A headless browser fetches images, fonts, analytics scripts and video preloads. Blocking resource types you do not need can cut bandwidth by 70–90% for typical pages, which is the single largest lever available on a metered plan.

Failed requests still cost. Blocked responses, redirects, error pages and challenge pages all transfer data. You pay for the CAPTCHA you did not solve.

Traffic expiry. Many providers expire unused traffic monthly. If your usage is seasonal, you are buying capacity you burn. This is a real cost and it never appears in a comparison table.

Minimum commitments. The attractive per-GB rate frequently requires a volume tier you have to reach. Check what you pay at your volume, not at the volume that unlocks the headline rate.

Engineering time. The largest hidden cost by a wide margin. Building and maintaining a pipeline is an ongoing commitment, not a one-off. This is the number that makes managed per-request APIs look reasonable when they seemed expensive per unit.

Why Residential Costs More Than Datacentre

Understanding the underlying economics helps you tell a fair price from a suspicious one.

Datacentre proxies run on servers the provider controls, in facilities with wholesale bandwidth. The marginal cost of another gigabyte is low, and the marginal cost of another address is a registry fee. This is why datacentre is cheap — around $0.14/GB at the low end of the market — and why per-IP pricing is the norm there.

Residential proxies route through ordinary consumer connections. The provider does not own those connections and must compensate whoever does, usually through an app or SDK that pays users or provides them something in exchange. That compensation is a genuine per-gigabyte cost that does not fall with scale the way datacentre bandwidth does. Market rates run roughly $0.79 to $7.00 per gigabyte, and the spread reflects pool size, geographic coverage and how the network was assembled.

Mobile proxies cost most because carrier data is expensive and addresses are shared among many subscribers by the carrier itself.

The important inference: an unusually cheap residential offer should prompt a question, not a purchase. The provider's own cost is real. If someone is well below the market, either they have a volume advantage they will explain, or the network was assembled in a way that did not involve paying anyone — which is an ethical problem, a stability problem, and after the enforcement action Google's threat intelligence team announced against a group of interlinked resellers in January 2026, demonstrably a legal one. Ask how the pool is sourced. The answer, and the willingness to give one, tells you a great deal.

Our Numbers and Where They Are Wrong for You

From our pricing page, checked September 2026. Verify before budgeting.

New accounts get 1 TB of residential traffic free, which is enough to run a real workload and get actual numbers rather than estimates. Use it for that.

Now the part that costs us sales.

Concentrated high-volume workloads are cheaper elsewhere. If you move a lot of data through a small number of addresses, per-IP pricing is the model that fits, and providers who sell datacentre access per address will beat us substantially. Our traffic-priced datacentre offering is an advantage for spread-out usage and a disadvantage for concentrated usage. Workload B above is the case: ten addresses, high volume, and we are the wrong shop.

Seasonal and bursty usage loses money with us. Our traffic does not carry over month to month. If you scrape hard for two weeks a quarter and idle otherwise, look for a provider whose traffic does not expire. You will pay less and worry less.

Very low volume should not be a purchase at all. If you need a few gigabytes a month, use the free tier and do not set up a subscription. We would rather have an accurate reputation than a small recurring charge.

Very specific geographic requirements should be tested before price is considered. No provider covers every city equally. If your work depends on one particular location, coverage beats rate, and the trial is where you find out.

Red Flags in Proxy Pricing

Patterns worth a second look, regardless of vendor.

No published pricing at all. A sales call before a number usually means the number depends on what they think you will pay. Some enterprise pricing is genuinely bespoke; a consumer-scale product hiding its rate is a different signal.

Prices far below the market with no explanation. As above. The underlying cost of residential bandwidth is real, and being well under it requires either a structural advantage the vendor can name, or a sourcing practice they would rather not discuss.

"Unlimited" with no thread or concurrency figure. The concurrency cap is the actual product. If it is not stated, it is not a commitment.

Pool size quoted as the headline feature. "150 million IPs" is a marketing number and unverifiable. What matters is how many are available in the location you need, at the time you need them, and how they behave against your target. Test rather than count.

No trial, or a trial too small to test anything. A hundred requests proves nothing about success rate over time. If a provider will not let you run a realistic sample, ask why.

Terms that are vague about what counts as billable traffic. Do failed requests count? Do retries? Does overhead? These should be stated, and where they are not, assume the answer that favours the vendor.

Guarantees of undetectability. No provider can promise this and the ones that do are the ones to avoid. The honest claim is that residential addresses raise fewer flags than datacentre ones, not that anything is invisible.

People Also Ask

How much do proxies cost?

Datacentre bandwidth starts around $0.14/GB, or a few dollars per address per month on per-IP pricing. Residential runs roughly $0.79 to $7.00/GB across the market, falling with volume. Mobile costs more again. The wide residential spread reflects real differences in coverage and how the network was assembled.

What is the difference between per-GB and per-IP proxy pricing?

Per-GB bills for data transferred with unlimited addresses; per-IP bills for addresses held with generous or unlimited traffic. Which is cheaper depends entirely on the shape of your usage — the same 600 GB monthly workload can cost $300 under one model and $12.50 under the other, depending on how many addresses it needs.

Are unlimited proxy plans worth it?

Only above the break-even volume. Divide the flat rate by your best metered rate to find it — at $1,800/month against $0.27/GB, that is about 6.7 TB. Below that you are paying for headroom. Also check the concurrency limit, which is usually the real constraint rather than traffic.

Why are residential proxies so expensive?

Because the provider does not own the connections and must compensate the people who do. That is a genuine per-gigabyte cost that does not fall with scale the way datacentre bandwidth does. A residential price far below the market is a reason to ask how the pool was sourced.

How do I estimate how much proxy traffic I need?

Measure a representative request rather than guessing, multiply by your request volume, and add a margin for retries and overhead. If you use a headless browser, measure the browser's actual traffic — it is routinely ten to sixty times the HTML size, and that gap is where most budget estimates fail.

How can I reduce my proxy costs?

Block images, fonts and media in headless browsers; use raw HTTP requests wherever rendering is unnecessary; cache aggressively so you never fetch the same thing twice; reduce retries by fixing the cause rather than raising the retry count; and check whether datacentre bandwidth works for the target before defaulting to residential.

Do failed proxy requests still cost money?

Yes, under per-GB pricing. Blocked responses, redirects, error pages and challenge pages all transfer data and all bill. This is why success rate is a cost metric and not only a quality metric, and why a pool with a 60% success rate is far more than 40% worse than one at 95%.

Is a scraping API cheaper than buying proxies?

Sometimes, and more often than the per-unit prices suggest. A managed API's price includes the proxies, the rendering, the retries and the ongoing maintenance. Raw proxies do not include the engineer. Compare total cost of ownership rather than unit rates, particularly for standard jobs like rank tracking where a managed service has done the hard part already.

Wrapping Up

Proxy pricing looks confusing because the industry sells several different products under one word, and the units are not interchangeable. A gigabyte and an IP address measure different things, and any comparison that puts them in the same column is comparing nothing.

The work that makes the decision easy takes ten minutes: estimate your monthly data, estimate how many distinct addresses you actually need by testing where rate limiting starts, and then price your workload under each model. The results frequently differ by an order of magnitude in one direction or the other, which means the cheapest provider for your neighbour may be the most expensive for you.

Then look past the rate. Retries, browser overhead, expiring traffic and engineering time are all real costs that never appear on a pricing page, and browser overhead in particular is usually the biggest single lever anyone has. Getting that one right will save you more than switching vendors.