“Unlimited residential proxies” is one of the most searched phrases in the market, and one of the least precise. It almost always means one thing: the plan does not meter bandwidth. Something else is metered instead, and that something decides what the plan is worth to you.
Most residential proxy plans bill by the gigabyte, which makes the cost of every request visible. An unlimited plan hides that cost inside a flat fee, and the provider protects itself by limiting a different resource. This guide covers which resources get limited, how to read the terms, and how to compare an unlimited quote with a metered one on the only number that matters: cost per successful request.
Why residential bandwidth is never truly unlimited
A residential exit is a real home connection. The operator pays the people or partners who supply those connections, usually in proportion to the traffic that flows through them. Every byte you send costs the provider money, whatever the plan says.
That means a flat fee only works for the provider if the average customer uses less than the fee covers. Heavy users have to be slowed down, capped or moved to a smaller slice of the network. None of this is dishonest by itself. It just means “unlimited” describes the billing, not the supply.
What unlimited plans limit instead
When bandwidth is not the meter, one or more of these usually is:
- Concurrent connections. A cap on threads or simultaneous sessions sets your maximum throughput. Our post on unlimited concurrent proxy connections covers why that cap matters at scale.
- Speed. A per-port or per-account bandwidth ceiling in Mbps. Your requests still go through, just more slowly once you are busy.
- Ports or IP count. You get a fixed number of gateway ports or addresses, and each one is shared across your whole workload.
- Pool size. Unlimited traffic is sometimes routed through a smaller, dedicated subset of the network rather than the full pool.
- Targeting. Country targeting may be included while city, state or ASN targeting is not.
- Time. Plans priced per day or per month, with a fair use threshold that triggers throttling or a plan review.
- Excluded targets. Some flat-rate plans exclude the sites that cost the most to serve.
A plan rarely uses all of these, but it always uses at least one. Find it before you compare prices.
The fair use clause is the real plan
The headline says unlimited. The fair use clause says what happens when you behave like a customer who believed it. Read it for three things:
- The trigger. Is there a stated threshold in gigabytes, requests or hours, or is it at the provider’s discretion?
- The consequence. Throttling, a forced upgrade, a move to a different pool, or suspension.
- The notice. Whether you are told before the change or find out when your success rate drops.
A discretionary clause with no stated threshold means the effective limit can change at any time. That is not automatically a reason to walk away, but it is a reason to test at the volume you will really run, not at a trial’s light load.
How a smaller or busier pool shows up
The most expensive limit is the one that never appears on the invoice. If unlimited traffic is routed through a smaller IP pool, each address serves more requests from more customers. Sites notice addresses that work too hard.
On an undefended page you will not see the difference. On search engines, social sites and anything behind a managed challenge, a heavily used address is challenged sooner and blocked more often. Your bill stays flat while the share of requests that return usable data falls. A flat fee spread over fewer successes is a higher price, even though nothing on the pricing page changed.
This is why an unlimited plan has to be measured against your real targets with content checks, not status codes. A challenge page usually arrives as a normal 200 response and looks like traffic served.
Pricing results, not plans
The comparison that works for any billing model is cost per successful request:
cost per success = what you paid ÷ requests that met your pass rule
For a metered plan you can estimate it before buying, with your payload size, the per-GB rate and a measured success rate. Our bandwidth estimation guide covers the payload side. For an unlimited plan, you need two measurements: how many requests the plan’s limits let you send in a month, and how many of those pass.
A worked example, with made-up round numbers:
- Unlimited plan. A flat $600 a month. At its concurrency cap your job sends 4 million requests, and say 60 of every 100 pass. That is 2.4 million successes, or about $250 per million.
- Metered plan. $4 per GB, pages averaging 40 KB. The same 4 million requests are about 153 GB, roughly $610. Say 90 of every 100 pass. That is 3.6 million successes, or about $170 per million.
The unlimited plan has the lower sticker price and the higher cost per result. Change the inputs and the answer flips: with heavy pages, a full browser and easy targets, a flat fee can be far cheaper. The point is that neither the word “unlimited” nor the rate card tells you which case you are in. The guide to cost per successful request explains the retry tax that drives this.
When an unlimited plan makes sense
Flat-rate pricing is a good fit when:
- Your pages are heavy, for example full browser loads with images and scripts, so bandwidth would dominate a metered bill.
- Your targets are lightly defended, so a busier pool costs you little in failures.
- Your volume is steady and stays well inside the concurrency and speed caps.
- You value a fixed monthly cost over the lowest cost per result.
It is a poor fit when your targets are search engines, social platforms or sites behind a managed challenge, when you need city or ASN targeting, or when your volume spikes. In those cases the hidden limit tends to bind exactly when you need the capacity.
A checklist before you buy
- Find the limit that replaced bandwidth: threads, speed, ports, pool or time.
- Read the fair use clause for its trigger, its consequence and its notice.
- Ask whether unlimited traffic uses the full pool or a subset.
- Run a trial at your real concurrency, against your real targets, for more than a day.
- Count successes with a written pass rule per target, not status codes.
- Divide what you would pay by those successes, and do the same for every metered alternative.
- Only then compare headline prices.
The true cost page of Atlas runs step 6 for every network it measures, putting the rate card beside the cost after failed requests, so you can see how far the two drift apart before you sign anything.