Per IP vs Per GB Proxy Pricing: Which Costs Less

Scraping heavy pages from targets that tolerate hundreds of requests per address? Buy per IP. Pages are big, addresses last, and the byte meter is punishing you for something you are not consuming much of: identity.

Scraping light pages from targets that ban an address after a handful of requests? Buy per GB. You are consuming enormous identity and almost no bandwidth, and a per-IP plan will bill you for thousands of addresses to move a few gigabytes.

Cannot predict either? Neither unit is safe, and the answer is a third one: flat monthly pricing by concurrent thread, which is flat along both axes. SparkProxy sells that shape, from $75 a month for 100 threads up to $440 for 1,000, unlimited bandwidth on every tier.

The next section explains why those three sentences are the whole decision.

The framing that makes this comparison tractable is to stop treating the units as prices and start treating them as inventories.

Per GB is not a price for bandwidth. It is a price for IP diversity, metered in bytes. Residential and mobile pools are expensive because assembling and maintaining millions of consumer addresses is expensive. Bandwidth is the meter the industry settled on, not the thing you are buying. You can prove this to yourself: nobody charges $3 per gigabyte for transit, because transit costs a fraction of a cent. The $3 is the address.

Per IP is not a price for an address. It is a price for exclusivity over time. When you rent a proxy for a month, you are buying the right to be the only one using that address, or one of a small number, for that period. Bytes are usually thrown in because the vendor already pays for the port.

Once you see the units that way, the comparison stops being "which is cheaper" and becomes a question about your workload: how much identity does it consume per unit of data? Everything else follows.

Our explainers on what bandwidth means in proxy services and residential proxy pricing per GB cover each meter in isolation if you want the mechanics first.

Here is the arithmetic, in full, so you can run it against whatever two plans you are actually comparing.

Let N be the number of distinct addresses your job needs in a month, G the gigabytes it moves, P_ip the monthly price of one address, and P_gb the price of one gigabyte.

• Per-IP bill: N × P_ip
• Per-GB bill: G × P_gb

They are equal when N × P_ip = G × P_gb, which rearranges to:

N / G  =  P_gb / P_ip

The left side is addresses per gigabyte, a property of your workload. The right side is a ratio of two published prices, a property of the market. Compare them and you have your answer, with no forecasting involved.

Run it on real September 2026 rates. Webshare's static residential tier lists $0.30 per proxy at 20 units. Decodo's residential subscription lists $3.50 per GB at the 10 GB tier. So:

P_gb / P_ip  =  3.50 / 0.30  =  11.7 addresses per gigabyte

Need more than about 11.7 distinct addresses to move each gigabyte, and per GB wins. Need fewer, and per IP wins. Swap in IPRoyal's published $5.25 per GB at 10 GB against the same $0.30 address and the threshold moves to 17.5. Compare against Webshare's datacenter tier at $0.0299 per proxy and the threshold jumps past 100, which is the arithmetic reason datacenter proxies feel absurdly cheap when they work at all.

One quotable line falls out of this: count addresses per gigabyte, not gigabytes. The gigabyte number on its own tells you nothing about which plan to buy.

You need two inputs, and both come from a couple of hours of testing rather than a spreadsheet guess.

Input one: T, requests per address before the target pushes back. Run a single address at your normal pacing and keep going until status codes change, challenges appear, or latency spikes. That count is the address's useful life for this target. Our guide to managing per-IP request limits covers how to run this cleanly.

Input two: S, average response size in megabytes. Sum len(response.content) across a few hundred real fetches and divide. Include the failures, because per-GB plans bill them.

Then:

addresses per gigabyte  =  1024 / (T × S)

Four worked examples, using the 11.7 threshold from the Webshare and Decodo rates above:

Two things jump out. Heavy pages push you toward per IP, hard, because bytes are the meter you are hammering. Tight per-address tolerance pushes you toward per GB, hard, because addresses are the inventory you are burning. The middle row is close to the line, which is where most real ecommerce work sits and why this argument gets had so often.

Every row below was read off that vendor's own page on 23 September 2026. Check the current figure before you buy.

The spread inside the per-GB column is the detail buyers miss. At the 10 GB tier, two published residential rates differ by 50 percent for a product doing the same job. Volume changes it again: IPRoyal's published custom rate at 10 TB is below its own 10 GB rate by a factor of three. Per-GB pricing is a volume curve, so a per-GB quote at your real monthly volume is a different number from the one on the pricing widget.

Plans are chosen on today's workload and paid for on next quarter's. Here is how each unit responds when something moves.

Read the columns, not the rows. Per IP is flat against everything except the target's tolerance for an individual address. Per GB is flat against everything except bytes. So the buying rule is: identify the axis of your workload you cannot forecast, then pick the unit that is flat along it.

The retry point deserves its own sentence, because it is the most expensive thing on that table and almost never modelled. On a per-GB plan, a 20 percent block rate is a 20 percent price increase. Blocked responses are still bytes. Our guide to reducing proxy bandwidth costs has the mitigations, and the biggest one is boring: stop downloading images and fonts you never parse.

Read the bandwidth column on any per-IP pricing page before you treat it as flat.

Webshare's datacenter tiers, read in September 2026, show $0.0299 per proxy alongside a 250 GB bandwidth figure. That is not a per-IP plan in the sense the break-even formula assumes. It is a per-IP plan with a byte ceiling, and once you cross the ceiling the economics change to whatever the overage terms say. Their static residential tiers show bandwidth as "Up to Unlimited", which is a range rather than a promise and worth a support ticket before you size around it.

This matters because the break-even formula only holds while both plans behave like their unit. A per-IP plan with a hard byte cap is really a per-GB plan wearing a per-IP coat, and you should price it that way: divide the plan cost by the included gigabytes and compare that effective rate against the per-GB column.

Do the same in reverse for per-GB plans that cap concurrent sessions. The published rate tells you the price of a gigabyte, not the rate at which you are permitted to consume one, and those are different constraints. We take apart the marketing around unmetered claims in are unlimited bandwidth proxies worth it.

Both units above bill something that varies with your workload. A third unit bills something that varies only with a decision you make: how many requests you run in parallel.

SparkProxy prices this way. Starter is $75 a month for 100 threads and 5 whitelist slots, Core $140 for 250 threads and 10 slots, Boost $240 for 500 threads and 15 slots, Plus $440 for 1,000 threads and 25 slots. All four carry unlimited bandwidth and 30 days validity, against a pool of 1M+ datacenter IPs across 80+ countries including 50,000+ US addresses, with random rotation per request and 5-minute sticky rotation available. Higher Pro and Pro+ thread counts exist in the fair usage policy without a published price, so treat those as a quote conversation.

Two honest caveats. The fair usage policy attaches a speed ceiling per tier, 25 Mbps on Starter through 150 Mbps on Plus, and a ceiling is not a guaranteed rate. And $75 a month is a genuinely bad deal for a job that makes 20,000 requests, where a per-GB plan with a $10 floor is obviously correct.

What the thread unit buys you is a bill that does not move when the workload does. Retries are free. Page weight is free. Address churn is free, because rotation happens inside the pool. The variable you are left managing is concurrency, which is the one variable you control directly rather than discover in production. If that framing is new, understanding concurrent connections covers how thread counts translate to throughput.

Four terms change the effective rate more than most tier discounts do.

Traffic expiry. Per-GB subscriptions usually reset monthly, so unused gigabytes evaporate. IPRoyal's page states that traffic never expires on its residential product, which is a materially different offer at the same nominal rate if your volume is lumpy. Ask whether unused traffic rolls over before you compare two per-GB prices.

What the meter counts. Request bytes as well as response bytes? Headers? TLS handshake overhead? Compressed or uncompressed? A vendor metering uncompressed response bodies against one metering wire bytes will differ by a factor of three or more on text-heavy pages. Get this in writing, then verify it against your own byte counts in the first week.

Replacement policy on per-IP plans. Addresses go stale. When one starts returning challenges, can you swap it, how many swaps per month, and is there a cooldown? A plan with no replacement allowance is a plan whose effective address count falls all month.

Minimum commitment and overage rate. The per-GB overage rate is frequently higher than the in-plan rate, which turns a small forecasting miss into a large invoice. The per-IP equivalent is a minimum term, where the 90-day price is attractive until the target changes and you hold addresses you cannot use.

If you are buying at any real volume, all four of these are negotiable, and our notes on negotiating proxy volume discounts cover which ones vendors actually move on.

The one pattern worth carrying away: workloads that consume identity should pay for identity, and workloads that consume bandwidth should pay for bandwidth. When you cannot tell which you are, that uncertainty is itself the answer, and a flat unit is the cheapest way to buy it. For the wider cost picture, how much proxies cost puts these numbers next to the rest of the market.