How Many Proxies Do I Need?

This question has two completely different answers depending on whether you are buying *identities* or *throughput*. Accounts need identities and the count is fixed. Scraping needs throughput and the count falls out of a calculation. Mixing the two up is why people either massively overspend or get throttled on day one.

Web scraping

With rotating residential proxies you do not count individual IPs — the pool rotates automatically across a large address space. Instead you size by bandwidth. But you still need to know how many *concurrent exits* your job implies, because that determines whether the pool can support it at all.

The concurrency formula

You find the denominator experimentally: send requests through one exit at increasing rates until the success rate degrades, then take about half of that figure. If a target tolerates 300 per hour per IP and you need 60,000 per hour, you need roughly 200 exits working at once. A rotating pool provides that without you renting 200 addresses; a static setup would not.

The bandwidth estimate

Separately, work out what you will transfer. Fetch twenty representative pages, measure the average response size, and multiply.

The last two rows are the same job. Blocking images, fonts and media in the fetcher is the difference between €288 and €62 a month, and it takes a few lines of configuration. Do that before you shop for a cheaper per-GB rate.

Multi-accounting

Use one dedicated IP per account — a static ISP proxy — so accounts stay isolated. The number of proxies equals the number of accounts you run. This one is not a heuristic you can tune: sharing an address across accounts recreates exactly the linkage the setup exists to prevent, as covered in the multi-accounting guide.

At €1.80 per IP, twenty accounts is €36 a month. Weighed against the cost of losing twenty warmed accounts at once, this is the cheapest insurance in the stack.

Sneakers & ticketing

Plan roughly one IP per checkout task or queue position. If you run 50 tasks, aim for around 50 ISP IPs, or enough residential bandwidth to support 50 concurrent sessions. The ratio holds because the tasks run *simultaneously* — sharing an address across two tasks means two entries from one household at the same instant, which is a straightforward filter for the site to apply.

For sneaker drops latency matters as much as count, so ISP is usually the right type. For ticketing queues, where you may sit waiting for a long time, count matters more than raw speed.

Quick reference

Sizing mistakes that cost money

1. 1Buying bandwidth before measuring page weight. Your estimate will be wrong by a factor of five in either direction.
2. 2Renting static IPs for a rotating workload. Ten IPs is ten identities, and the rate limit will find you.
3. 3Sharing account IPs to economise. The saving is a few euros; the exposure is the whole portfolio.
4. 4Sizing for peak instead of average. Buy for the normal month and top up for the spike — bandwidth carries over.
5. 5Ignoring retries in the estimate. A 15% failure rate is 15% more bandwidth than your page count suggests.