How Much Do Proxies Cost? Real Prices by Proxy Type

Every figure below is each vendor's own published rate, read from their public pricing page as of September 2026. These pages reprice often and most run promotional discounts, so treat them as a snapshot and check the vendor's site before you budget.

Two things jump out of that table. The spread inside a single category is larger than the spread between categories: residential runs from $1.40 to $6.00 per GB depending on who you ask and how much you commit. And the per-IP datacenter market has effectively two price points, roughly three cents and roughly a dollar fifty, which are not the same product.

You cannot compare a $2.99 plan to a $75 plan until you know what each one meters.

The unit should follow the shape of your workload, not the sticker price. A rank tracker pulling 3 million lightweight SERP pages a month and a QA team opening 200 sessions a day sit at opposite ends of this table, though both call themselves "proxy buyers". If concurrency is the term you are least sure about, concurrent connections in proxies explains what a thread ceiling actually limits.

Datacenter IPs are the cheapest exits in the market because the underlying resource is cheap: a subnet in a hosting facility, not a contract with a household or a SIM card. They are also the most detectable, since the ASN is visibly commercial.

There are two sub-markets, and the price gap between them is a quality gap.

Bulk shared pools sell at a few cents per IP per month. Webshare's published ladder runs $2.99 for 100 proxies down to $1,076.40 for 60,000, which is $0.0179 per IP at the top. You are sharing those IPs with other customers, and you inherit whatever they did on your target yesterday. That is fine for uptime checks and generic crawling, and unpredictable for anything with a login. The trade-offs are covered in detail in shared vs dedicated datacenter proxies.

Dedicated IPs sell for roughly $1 to $2 per IP per month. Bright Data's published datacenter ladder is $1.80 per IP at 10 IPs, $1.45 at 100, $1.40 at 500 and $1.30 at 1,000, as listed in September 2026 with a discount banner on the page.

The third shape is the one most scrapers should price out: a rotating gateway with unlimited bandwidth, sold by concurrency rather than by IP or byte. SparkProxy's published plans work this way:

All four are unlimited bandwidth on 30-day validity, drawing from 1M+ datacenter IPs across 80+ countries including 50,000+ US datacenter IPs, reached through one host at gateway.sparkproxy.io on port 11000 for HTTP and HTTPS, 11002 for sticky sessions and 13000 for SOCKS5. Two larger tiers, Pro at 1,500 threads and Pro+ at 2,000, exist in the fair usage policy but are quoted rather than published, so no price is printed here. The speed figures are ceilings, not guaranteed throughput, which matters when you size a nightly window. For the wider landscape of how these models are structured, see datacenter proxy pricing models.

ISP proxies are datacenter-hosted machines announcing IP space registered to a consumer ISP. You get residential-looking ASN data with datacenter latency and a static address that survives for the billing period.

Published rates in September 2026 sit between $0.30 and $1.80 per IP per month, which is a 6x spread for a category that sounds uniform. Webshare lists static residential at $6.00 for 20 proxies, $270 for 1,000 and $2,250 for 10,000. Bright Data lists $1.40 per IP at 10, $1.00 at 100 and $0.90 at 1,000. IPRoyal advertises "from $1.80/proxy". Oxylabs shows ISP "starts from $16" without a per-unit breakdown on the page.

The reason the spread is so wide is that "ISP proxy" describes a routing fact, not a quality tier. Two providers can both announce consumer ASN space while one holds clean, low-abuse ranges and the other resells subnets that every major retailer already scores as suspicious. Price is a weak signal for that. A 200-request test against your actual target is a strong one. If you are still deciding between this tier and plain datacenter, ISP proxies vs datacenter proxies compares them on the axes that change the buying decision.

Residential pools route through real consumer devices, which is why they cost 20 to 100 times more per byte than datacenter traffic. The provider has to acquire and compensate the exit, and that acquisition cost is the floor under every per-GB price in the market.

The published range as of September 2026 is roughly $1.40 to $6.00 per GB:

Check each vendor's page for current rates before you commit, since several of these numbers were displayed alongside a promotional discount.

Notice what the ladders have in common: the discount is bought with commitment, not negotiation. Bright Data's own ladder charges 60% more per gigabyte at pay-as-you-go than at its $1,999 tier. Webshare's charges 150% more at 1 GB than at 3,000 GB. If your volume is genuinely uncertain, the honest way to read those tables is that you will pay the top of the range, because unused gigabytes on most networks expire at the end of the month rather than rolling over.

The second thing to understand is that the per-GB rate is only half the equation. Your bill is rate multiplied by bytes, and bytes are almost entirely under your control. That is the subject of the next two sections, and what bandwidth means in proxy services covers what gets counted.

Mobile exits route through 4G or 5G carrier networks. Because carrier-grade NAT puts thousands of real subscribers behind the same address, blocking a mobile IP has real collateral damage, so sites are slow to do it. You pay for that reluctance.

Two models exist. Dedicated port rental gives you exclusive use of one modem with an API endpoint to force a rotation, priced per month. IPRoyal publishes mobile "from $117/month" as of September 2026, and single-port rentals across the market commonly land between $50 and $300 per month depending on country and carrier, with US and Western European ports at the top of that range. The alternative is a metered mobile pool sold per gigabyte, which usually prices above residential.

Mobile is worth its price for a narrow set of jobs: mobile app traffic, platforms that treat carrier IPs differently from residential, and account work where a single sticky identity has to survive for weeks. It is a poor fit for bulk collection, where you pay a large premium per byte for an exit limited by a radio link. What a mobile proxy is covers the mechanics if you are evaluating the tier for the first time.

A scraping API bundles the proxy, the retry logic, the fingerprint handling and often a headless browser into one billed request. You stop paying for bandwidth and start paying for outcomes.

SparkProxy's published credit model shows how the arithmetic works. A plain fetch costs 1 credit, a JavaScript render costs 5, and a screenshot or PDF costs 10. Plans are $49 for 250,000 credits at 50 concurrent, $99 for 1,000,000 at 100 concurrent, $249 for 3,000,000 at 200 concurrent, and $599 for 8,000,000 at 400 concurrent. There are 1,000 free credits with no card, which is enough to answer the only question that matters before you buy: does it get through on your target.

Convert that to a unit price on the $99 Growth plan:

The comparison to make is not API price against proxy price. It is API price against proxy price plus the engineering time to build and maintain what the API includes. If one engineer spends two days a month on block handling, that is real money against a $99 plan. Web scraping API vs self-managed proxies works through where each side wins.

Per-GB and per-month prices are not comparable until you convert both to cost per 1,000 successful pages. The conversion needs one number you have to supply yourself: average bytes per page.

Do not guess it. Run your own crawler against 100 real target URLs, sum the response sizes, and divide. The figures below are worked examples at three page weights so you can see the shape of the curve, not measurements of anything:

The lever nobody sells you sits in the left column. Blocking images, fonts, media and third-party analytics in a headless browser routinely takes a page from the bottom row toward the top one, and that single change moves a residential bill by roughly an order of magnitude. No amount of vendor negotiation gets you a 10x rate cut. Request filtering can.

On a flat unlimited plan the same table inverts, because volume is the denominator rather than the multiplier:

The "out of reach" cell is the honest part. A flat plan does not scale without limit: 10 million pages a month at 250 KB each is 2.5 TB, and Starter's 25 Mbps ceiling cannot move that inside a month even running continuously. Concurrency and the speed cap, not billing, are what stop you. Size the tier to the throughput you need, then read the price off it.

Here is the calculation that decides most buying decisions in this category, and it takes one line.

A $75 flat plan buys the same spend as 30 GB of residential at $2.50/GB, or 43 GB at $1.75/GB. At 250 KB per page, 30 GB is about 125,000 pages.

So: if your target accepts datacenter IPs and you pull more than roughly 125,000 pages a month, the flat plan is cheaper, and it stops getting more expensive as you grow. Below that volume, per-GB residential is cheaper and you should not be paying a monthly minimum at all.

The condition carries all the weight. Run the test before the arithmetic: send a few hundred real requests through a datacenter exit at your actual target and record the success rate. If datacenter IPs get blocked, the cheapest per-GB plan beats the cheapest flat plan by an infinite margin, because the flat plan returns zero usable pages. Residential vs datacenter proxies sets out which target types tend to fall on which side.

Four of these move budgets more than the headline rate does.

Retries are billed. On per-GB and per-request models, a blocked response costs you the same bytes as a successful one, plus the TLS handshake. At a 70% success rate you pay for about 1.43 attempts per usable page, which turns a $2.50/GB rate into an effective $3.58. Flat unlimited plans are indifferent to this, which is a real and rarely priced advantage on hostile targets.

Unused volume expires. Most per-GB plans reset monthly with no rollover, so a lumpy workload pays for peak capacity every month and uses it once a quarter.

Overage rates exceed plan rates. Going past a bundled allowance is typically billed at the pay-as-you-go rate, which on published ladders is the most expensive tier available.

Geo and platform surcharges. Some networks price certain countries or certain target categories above the base rate. Read the rate card, not the landing page.

Then the softer ones: engineering time for rotation and retry logic, seat limits that force a tier upgrade for organisational rather than technical reasons, whitelist slot limits if you authenticate by IP, and KYC approval that can delay a start by days.

Work through this in order. Stop at the first line that fits.

1. Fewer than about 10,000 requests a month? Buy nothing yet. Free tiers and trials cover this, including 1,000 free scraping API credits with no card.
2. Does your target serve normal responses to a datacenter IP? Test 200 requests before deciding. If yes, and you are above roughly 125,000 pages a month, a flat unlimited datacenter plan is the cheapest option available and stays cheapest as you grow.
3. Do you need a stable identity per account or session? You are buying ISP or static residential per IP. Count the sessions you need, multiply by the per-IP rate, and ignore bandwidth entirely.
4. Does the target block datacenter ASNs outright? You are buying gigabytes. Cut page weight first, then shop the per-GB rate, and commit only to volume you will actually consume.
5. Is the target a mobile app, or a platform that treats carrier IPs differently? Mobile ports, priced per month, and accept that this is the expensive tier.
6. Are you spending more on block handling than on proxies? Price a scraping API per successful request and compare it against proxy spend plus engineering hours.

Whatever you shortlist, buy the smallest possible unit first and run one week of your real workload through it. Success rate against your target is the only number that determines whether any of the prices above were a good deal.