Proxies are priced in three ways: per gigabyte transferred, per IP address per month, or per port (one dedicated mobile modem/SIM) per month — and which model fits depends on the shape of your traffic far more than on the headline number. A task that moves a lot of data through a few IPs is cheap per-IP and ruinous per-GB; a task that touches a million pages once each is the reverse. This guide explains what each model actually charges for, how to estimate your own consumption before you buy, and the costs that do not appear on the pricing page.
The three billing models
| Per GB | Per IP / month | Per port / month | |
|---|---|---|---|
| Typical products | Rotating residential; some mobile pools | Datacenter; ISP (static residential); dedicated datacenter | Dedicated mobile (one SIM/modem); some 4G/5G rentals |
| What you pay for | Bytes in + out through the proxy | The address itself, usually unlimited traffic | Exclusive use of a physical connection, usually unlimited traffic |
| Cost driver | Page weight × request count | How many addresses you need concurrently | How many simultaneous identities you need |
| Scales well for | Many light requests across many IPs | Heavy traffic through few, stable IPs | Few high-trust identities; on-demand rotation via reconnect |
| Scales badly for | Media-heavy pages, downloads, video | Needing thousands of distinct IPs briefly | Anything needing many IPs at once |
| Order of magnitude | Low single-digit dollars per GB, falling with volume | Datacenter: ~$1–3 per IP; ISP: several times that | Tens of dollars per port; more for premium carriers/countries |
Prices move and vary by provider and country, so treat the last row as relative scale, not a quote. What does not change is the shape: per-GB grows with bytes, per-IP grows with address count, per-port grows with identity count. The types themselves are compared in Residential vs datacenter vs mobile; this article is about what they cost to run.
Estimating traffic per request (the number everything hinges on)
Per-GB billing counts every byte, both directions, of every request — including redirects, retries, and the images, scripts and fonts a page pulls in. The dominant term is what you fetch:
| What you fetch | Typical size | Requests per GB |
|---|---|---|
| API JSON response | 2–50 KB | ~20,000–500,000 |
| HTML only (no assets), product/listing page | 50–300 KB | ~3,500–20,000 |
| HTML with compression negotiated | often 3–5× smaller than uncompressed | proportionally more |
| Full page in a browser (HTML + CSS + JS + images) | 1–5 MB | ~200–1,000 |
| Search-engine result page (HTML) | 100–500 KB | ~2,000–10,000 |
| Image or file download | whatever it is | size-bound |
The single biggest lever on per-GB cost is not loading what you do not need: request HTML only rather than driving a browser, send Accept-Encoding: gzip, br, and in browser automation block images, fonts and media at the network layer. A browser session that renders full pages can cost 10–30× more traffic than fetching the same HTML directly. Where anti-bot forces a browser (see why some targets require one), blocking assets typically halves the bill.
Measure rather than guess: fetch 100 representative pages through the proxy and read the provider dashboard's traffic counter, or sum len(response.content) plus request size in your script (dashboard numbers include protocol overhead and will run somewhat higher).
Worked examples
1. Scraping 500,000 product pages a month, HTML only, protected e-commerce site
Rotating residential is required by the target. At ~150 KB per compressed page: 500,000 × 0.15 MB ≈ 75 GB/month. Per-GB pricing applies; the cost is 75 × the per-GB rate at your tier. Rotation is per request, so the number of IPs is irrelevant to the bill — only bytes matter. If you drove a browser and loaded assets, the same job would be ~1 TB and cost 10–15× more.
2. Fifty marketplace seller accounts, each logged in daily
Each account needs its own persistent IP (see why accounts must not rotate). Traffic per account is small — a few sessions a day — so per-GB would be cheap in bytes but cannot provide stable addresses. This is a per-IP job: 50 ISP proxies. Cost = 50 × per-IP monthly rate. If the platform is mobile-first and bans are the main risk, it becomes 50 mobile ports (or fewer, if the platform tolerates several accounts behind one carrier IP with distinct fingerprints — a judgment call covered in the ISP proxy guide).
3. Price monitoring: 20,000 pages, checked hourly, tolerant target
480,000 fetches a day. Per-GB residential would be ~70 GB/day — expensive. But a tolerant target does not require residential; a few dozen datacenter IPs on unlimited-traffic per-IP plans handle it for a flat fee. Here the type decision drives the cost more than the model: choosing residential where datacenter suffices is the classic overspend.
4. Ad verification in 15 cities
Low volume, high geo precision, must look like a local resident. Rotating residential with city targeting, per-GB, tiny bill because pages are few. Per-IP ISP proxies would need coverage in each city, which may not exist. Per-GB wins on flexibility, and the volume keeps it cheap.
The costs that are not on the pricing page
- Retries and failures. Every 403, CAPTCHA page and timeout you retry is billed traffic on per-GB plans. A 20% failure rate is a 20% surcharge. Improving success rate (right IP type, consistent fingerprint) is a direct cost cut.
- Minimum commitments and tiers. Per-GB rates fall with volume but often come with monthly minimums or expiring bundles; unused GB at month end is money spent. Size the plan to measured need.
- Concurrency limits. Some plans cap simultaneous connections; hitting the cap means buying a higher tier even if bytes are within budget.
- Bandwidth of your own side. Per-IP “unlimited” is unlimited at the proxy; your server's egress may not be.
- Burned IPs. A per-IP or per-port address that gets flagged on your target platform is a sunk month; ask about replacement policy before you rely on it (checklist in what to check before buying ISP proxies).
- Time. Slower proxies mean longer jobs; on residential/mobile, throughput is bounded by the exit connection (measure it — method here). If your pipeline is time-boxed, slower can mean needing more concurrency, i.e. a higher tier.
A quick decision procedure
- Does the target tolerate datacenter IPs? If yes: per-IP datacenter, done. Test it before assuming it does not.
- Do you need persistent identities? If yes: per-IP (ISP) or per-port (mobile), one per identity. Rotating pools cannot do this.
- Otherwise you need residential/mobile for stateless volume: estimate bytes = requests × size per request; minimise size (HTML only, compression, asset blocking); price it per GB.
- Add a failure margin (10–30% depending on how protected the target is) to the byte estimate.
- Compare against the per-port alternative if the identity count is small: a handful of mobile ports with reconnect-rotation can beat per-GB for medium-volume, high-trust work.
Then run a small paid trial and measure actual consumption for a day before committing to a tier. Provider rankings help with the who; the reliability ranking notes which vendors offer which billing models.
Frequently asked questions
Why are residential proxies billed per GB?
Because the provider pays for someone else's home bandwidth (via SDK partnerships or leases) with no fixed pipe to amortise, so it passes bytes through as the cost unit. Datacenter and ISP proxies run on the provider's own servers, where a flat per-IP price works.
How much traffic does scraping use?
Roughly the size of what you fetch times how many times you fetch it: 50–300 KB per HTML page, 1–5 MB per fully rendered browser page, a few KB per API call. Measure 100 real requests to calibrate.
Is per-IP or per-GB cheaper?
Per-IP for heavy traffic through few stable addresses; per-GB for light traffic across many addresses. The crossover depends on page weight and request count — compute both for your numbers.
What does a mobile proxy port mean?
Exclusive use of one physical SIM/modem connection. You get one carrier IP at a time, unlimited traffic, and can request a new IP by reconnecting. Priced per port per month; the unit is an identity slot, not bytes or addresses.
How do I reduce per-GB proxy costs?
Fetch HTML instead of rendering pages where possible, enable compression, block images/fonts/media in browser automation, cut retries by improving success rate, and never route non-essential traffic (analytics, third-party scripts) through the proxy.
Do failed requests cost money?
On per-GB plans, yes — whatever bytes moved (including CAPTCHA pages and error bodies) are billed. On per-IP/per-port plans, no extra bytes charge, but a burned IP is a lost month.