Why Proxy Prices Vary So Much Between Providers

Three forces set a proxy price, and only the first is about packets.

How the addresses were obtained. Leasing hosting address space, leasing consumer address space from an internet provider, and paying end users for the right to exit through their connection are three completely different cost structures. They differ by orders of magnitude, and they explain most of the gap between proxy types.

How much of the surrounding work the vendor does. Keeping ranges clean, screening customers, replacing burned addresses, staffing support, holding certifications and signing data processing agreements all cost money. A vendor who does none of it is genuinely cheaper to run, and you inherit the consequences.

Who the vendor is selling to. The same underlying capacity sold to a procurement department with a security questionnaire costs more than the same capacity sold through a self-serve checkout, because the sales cycle, the legal review and the account manager are all in the price.

Most of the spread you see is the second and third of those, not the first.

1. Address acquisition. Datacenter providers lease or own IPv4 blocks, which carry a monthly holding cost per address that exists whether you send a byte through them or not. ISP providers negotiate with consumer internet providers for addresses inside their registered ranges, which is a relationship business with a much smaller supply and a correspondingly higher price. Residential networks pay end users, usually through revenue-shared software, which converts an asset cost into a per-gigabyte cost. Mobile networks add SIMs, data plans and hardware on top. Those are four different balance sheets, and no amount of comparison shopping makes them the same product.

2. Bandwidth and transit. Wholesale transit in a datacenter is bought by the gigabit and is cheap per gigabyte at scale. Consumer exits are not: every gigabyte has to be paid for at the exit, usually to the person whose connection carried it, plus transit on both legs. The gap between those two numbers is the single largest input to the price difference between datacenter and residential products. Background on the unit itself is in what bandwidth means in proxy services.

3. Subnet quality and churn. A range nobody has hammered is worth more than a recycled one, and the difference is invisible on a pricing page. Providers who rotate out burned prefixes and buy clean ones carry a cost that providers who simply keep reselling degraded ranges do not. When a price looks impossibly low, this is usually where the saving came from. Clean datacenter subnets covers how to check, and what a datacenter ASN is explains why targets classify by range in the first place.

4. Abuse handling and customer screening. Customer screening, use-case review and an abuse team are pure cost with no visible feature attached. They exist to stop a minority of customers from burning the ranges everyone else is using. A provider who screens nobody has lower costs today and dirtier ranges in six months, and you are sharing those ranges.

5. Support and paperwork. Round-the-clock support, named account managers, security questionnaires, data processing agreements and certification audits all get amortised into the unit price. For a two-person team none of that is worth paying for. For a company whose security review can block a purchase, it is the product.

6. Platform engineering. Rotation logic, sticky sessions, per-country exits, usage dashboards, sub-user management and an API are software, and software has a payroll behind it. A vendor selling a list of addresses and a password has a smaller engineering bill than one selling a managed gateway.

7. Margin and route to market. Enterprise vendors carry sales teams, legal review and long procurement cycles. Self-serve vendors carry a signup form. That difference alone can double a price without a single byte moving differently.

Here is the structural fact that makes sense of almost everything else, and it comes straight from driver one and driver two.

For a datacenter proxy, the cost is the address. It is a fixed monthly holding cost, and the marginal cost of pushing one more gigabyte through an address you are already paying for is close to nothing. For a residential or consumer-exit proxy, the cost is the traffic. Every gigabyte is paid for at the exit, so the marginal cost of the next gigabyte is real and positive and never goes away.

That asymmetry has a consequence you can use as a diagnostic: a provider's willingness to sell you unlimited bandwidth tells you what their cost structure is. Unmetered plans are common on datacenter products because the marginal byte is nearly free. Nobody sells genuinely unlimited consumer-exit bandwidth, because doing so would be selling a negative margin to the heaviest user on the platform.

It also explains the shape of the fair usage clauses attached to unmetered plans. When a provider's real constraint is capacity rather than bytes, the limit they impose is usually a speed ceiling or a concurrency cap, not a traffic charge. SparkProxy's own plans are an instance of this and we may as well be transparent about the arithmetic: Starter at $75 a month for 100 threads, Core $140 for 250, Boost $240 for 500 and Plus $440 for 1000, all unlimited bandwidth on 30 day validity, with fair usage speed ceilings of 25, 50, 100 and 150 Mbps respectively, rising to 200 and 250 Mbps on the Pro and Pro+ thread tiers that exist in the policy without a published price. We can price that way because the exits are datacenter addresses whose marginal byte costs us nearly nothing. A residential network cannot, and any that claims to is either metering you somewhere you have not read yet or has not met a heavy user. The general argument is in whether unlimited bandwidth is worth it.

You can bound a vendor's costs using nothing but their competitors' published prices, on the reasonable assumption that the cheapest seller in a category is not running at a loss indefinitely. Two worked deductions, from figures read on each vendor's own page on 23 September 2026.

Deduction one. Somebody is selling a hundred exclusive static addresses for $30 a month and staying in business, so supplying those addresses cannot cost much more than $0.30 each. The $2.45 gap to the $2.75 product is therefore about 89% of that price, and it is not address cost. It is ASN quality, subnet diversity, replacement policy, support, compliance paperwork and margin. Whether that 89% is worth paying depends entirely on whether your targets care about the things it buys.

Deduction two. The same logic on bandwidth: somebody sells consumer-exit gigabytes at $1.30, so at least 84% of an $8.00 pay-as-you-go headline is something other than the byte. Part of that is a genuine no-commitment premium, which is why the same vendor's committed tier is $5.00, and part of it is everything in the seven drivers above.

Treat both as bounds rather than as estimates of anyone's margin. They are still more than any vendor will tell you, and they are enough to stop you assuming that a 9x price difference reflects a 9x difference in what arrives at your socket.

A large share of apparent price variation is not price variation at all. It is unit confusion.

Proxies are sold per address per month, per gigabyte, per concurrent thread and per successful request, and each unit is flat along a different axis. A per-gigabyte plan looks cheap until pages get heavy. A per-address plan looks cheap until your target tightens its per-address tolerance. A per-thread plan looks expensive until you notice it has no meter at all. Comparing the four entry prices tells you nothing, because they are prices of four different things.

The only fair comparison is your own workload priced through every model, which we work through in datacenter proxy pricing models for the flat and metered cases and in residential proxy pricing per GB for the byte-based ones. The rule that falls out is short: find the axis of your workload you cannot forecast, and buy the model that is flat along it.

Low prices are not a warning sign. Low prices with no explanation are.

The four "good" rows are a real business model and there is nothing wrong with buying from it. Several of the cheapest vendors in this market are cheap because they sell through a signup form to people who can read documentation, which is a legitimate way to remove cost. The six "bad" rows are cheap because something was removed that you were going to need. How to spot a fake proxy provider covers the harder end of that, and the economics of the genuinely free tier are in free vs paid proxies.

Price bands in this market correspond fairly reliably to what is included, and knowing the pattern saves a lot of quote-reading.

Budget self-serve. You get addresses, credentials and documentation. You give up account management, negotiated terms, compliance artefacts and often the ability to get a human on a Sunday. Ranges are more likely to be shared and more likely to have history, which is why shared proxies and semi-dedicated proxies exist as named tiers rather than as an accident.

Mid-market self-serve. You get a dashboard, an API, per-country exits, sub-users and a support queue that answers. You give up procurement paperwork and negotiated pricing. For most teams doing most work, this band is the efficient frontier.

Enterprise. You get the questionnaire answered, the data processing agreement signed, an account manager, an availability commitment and an invoice your finance team can process. You give up roughly the price difference, which is real money for a small team and a rounding error for a company whose security review can veto the purchase. Whether the availability commitment means anything is a separate question, covered in how to read a proxy provider SLA and proxy uptime and reliability.

One thing that does not reliably improve with price: success rate on your specific targets. Every vendor publishes a figure in the high nineties, none of those come from an independent audit, ours included, and the ranking on your targets can invert the ranking on price. That is the whole argument for testing rather than shortlisting, and the practical version of it is in what to evaluate when selecting a proxy service.

Two quotes become comparable when you know four things about each. Ask them in writing.

1. What is the unit, and what is included in it? Addresses, gigabytes, threads or successful requests, and whether traffic, geo-targeting and rendering are inside the number or added to it.
2. What is excluded? Tax, overage rates, geo-targeting surcharges, premium exits, and anything that is priced per seat rather than per unit.
3. What happens at the limit? A charge, a throttle, a concurrency reduction or a hard stop. These are four very different outcomes and vendors describe all of them as "fair usage".
4. What happens when something breaks? Replacement policy for burned addresses, response time, and whether a failed request costs you anything. On a per-gigabyte plan a block page is billable traffic, which quietly turns a 20% block rate into a 20% price increase.

Then run the same workload through both for a week, on your own targets, and compare cost per successful request rather than cost per unit. Every vendor in this market has a trial, a free tier or an entry price small enough to make that affordable, which the survey in proxy free trials and refund policies sets out. A week of measurement beats a quarter of comparison tables, including ours.