How many proxies do I need?

A proxy count and a capacity limit are different numbers. Of 710 plan rows, 239 publish an IP count and 238 publish a concurrency maximum, but only 77 publish both. Another 225 publish none of the four capacity fields for IPs, concurrent sessions, threads or ports. The plans behind the distinction include oxylabs-datacenter-dedicated-pay-per-ip-20-ips, which sells 20 IPs without a published concurrency cap, and litport-mobile-us-shared-device-month, which publishes 100 threads without an IP count. Oxylabs source, retrieved 2026-07-25T15:31:32Z · LitPort source, retrieved 2026-07-25T16:03:55Z · Full observed plan data

There is no universal “one proxy per task” rule in this dataset. The required number depends on whether “proxy” means an assigned IP, a rotating gateway, a port, a thread allowance or a simultaneous session. Vendors sell all five shapes.

First decide which unit must be sized

An assigned-IP plan gives a count directly. oxylabs-datacenter-dedicated-pay-per-ip-20-ips publishes 20 IPs for $40 per month. The row does not publish a concurrent-session or thread maximum, so it answers the address-count question but not the workload-capacity question. Oxylabs source, retrieved 2026-07-25T15:31:32Z

A thread plan can expose capacity without exposing a purchased address count. proxyrack-datacenter-usa-100-threads publishes 100 threads and unmetered transfer, while its ip_count is not published. Treating that row as “100 proxies” would convert a workload limit into an address claim the vendor source does not make. ProxyRack source, retrieved 2026-07-25T16:03:55Z

A port plan is a third shape. hydraproxy-usa-mobile-standard-30d publishes one port, unlimited concurrent connections, 30-day access and a $59.95 checkout. One purchased port therefore does not imply one allowed simultaneous request in that row. It also does not publish how many distinct addresses the buyer will observe. HydraProxy source, retrieved 2026-07-25T17:34:30.296Z

These are not edge cases. The dataset contains 267 per-IP rows and 38 concurrency-priced rows. Of the per-IP rows, 205 publish an IP count, while only 87 publish a concurrency value. None of the 38 concurrency-priced rows publishes an IP count. Full observed plan data, with evidence retrieved from 2026-07-25T15:31:32Z through 2026-07-26T02:15:19Z

Size fixed identities by assigned IP count

If the workload requires a distinct, stable assigned IP for each simultaneous identity, the starting calculation is:

required assigned IPs = peak simultaneous identities that must not share an IP

That calculation applies only after the requirement for separate IPs is established. It is not a claim about how many requests each IP can handle. webshare-free-proxy-server, for example, publishes 10 static IPs and a $0 checkout, but no session or thread maximum in the collected evidence. Webshare source, retrieved 2026-07-25T16:33:08.221Z

The other extreme shows why a large pool number is not a capacity answer. pia-s5-socks5-enterprise publishes 30,000 IPs for a $1,350 list checkout, but its row does not publish a concurrent-session or thread maximum. The count is real and the missing capacity value is also real. PIA checkout, retrieved 2026-07-25T17:41:11.108Z

Size rotating access by parallel work

For a rotating gateway, start from peak simultaneous requests, then compare that number with the plan’s published concurrency, thread or port rule. litport-mobile-us-shared-device-month publishes 100 threads for $49.99 per month but no IP count. proxyrack-datacenter-usa-100-threads also publishes 100 threads without an IP count. These rows can be candidates for a 100-thread requirement, but the dataset does not contain target tests proving that 100 useful responses will complete in parallel. LitPort source · ProxyRack source, both retrieved 2026-07-25T16:03:55Z

Do not multiply a rotating pool’s marketed address count by an assumed request rate. Pool scope, rotation and stickiness affect which addresses appear, while concurrency controls how much work can be attempted at once. The current rows keep these as separate fields because one is not a substitute for the other. Full observed plan data, evidence retrieved from 2026-07-25T15:31:32Z through 2026-07-26T02:15:19Z

Missing capacity should stop the calculation

Only 77 rows publish both an IP count and a concurrency maximum. There are 162 rows with an IP count but no concurrency value, 161 with a concurrency value but no IP count, and 310 with neither. A missing value means “Not published,” not zero and not unlimited. Full observed plan data, evidence retrieved from 2026-07-25T15:31:32Z through 2026-07-26T02:15:19Z

Before buying, write down four workload values: peak simultaneous identities, peak simultaneous requests, required stable-session duration and measured traffic for the billing period. Match each value to the plan’s IP, concurrency or threads, sticky-session and bandwidth fields. If a required field is absent, ask the vendor for the number or choose a row that publishes it.

The answer is therefore a unit-specific calculation. Buy enough assigned IPs for identities that truly need separation, or enough published threads, sessions or ports for the peak parallel workload. Do not turn one of those numbers into another, and do not use a large marketed pool as evidence of measured capacity.