Residential vs datacenter: stop buying the expensive one

Residential proxies are the expensive ones. A large share of the people paying for them did not need to, and found out by never testing the cheaper option. This post is about how to tell, in about five minutes, which one a target requires.

The difference that matters

A datacenter IP belongs to a hosting company. When a site looks it up, it sees an ASN owned by a cloud provider or a colocation facility, and it can reasonably guess there is no human behind it.

A residential IP belongs to a consumer ISP. The same lookup says “Comcast” or “Deutsche Telekom”, which is what a real person’s home connection looks like. Blocking it risks blocking paying customers, so sites are far more careful about it.

That is the whole distinction. Speed is not it: datacenter is considerably faster. Reliability is not it either, since datacenter IPs are more stable. You are paying residential rates purely for the ASN the target sees.

How the two are billed here

The two lines are not priced in the same unit, which is the first thing to get straight. Residential is pay-as-you-go by the gigabyte: $2.40/GB for a single gigabyte, falling to $1.50/GB at the top of the ladder. Datacenter is rented by the IP for a term. One dedicated address for 30 days is $1.90, and a dedicated address has no traffic allowance to run out of.

Put those side by side and the break-even is small. A dedicated datacenter IP for 30 days costs the same as 0.79 GB of residential at the single-gigabyte rate, or 1.27 GB at the cheapest rung. If your job moves more than that through one address in a month, and the target lets datacenter through, the flat price wins.

The five-minute test

Do not reason about this. Measure it. Take fifty URLs that are representative of your job and fetch them through datacenter, then through residential, and compare status codes. A datacenter address is yours alone, so its line uses the IP and port you were assigned; residential goes through the shared gateway.

test() {
  ok=0; blocked=0
  while read url; do
    code=$(curl -s -o /dev/null -w '%{http_code}' -x "$1" "$url")
    case "$code" in
      2*) ok=$((ok+1)) ;;
      *)  blocked=$((blocked+1)) ;;
    esac
  done < sample.txt
  echo "$2 — ok:$ok blocked:$blocked"
}

test "http://USER:PASS@IP:PORT"                  datacenter
test "http://USER:[email protected]:8000" residential

The cheapest way to run it is one shared datacenter IP for 30 days, which is $1.25 and includes 1 GB of traffic, far more than fifty pages need. Only buy residential gigabytes if datacenter fails. There is no version of this test that costs more than guessing wrong for a month.

Reading the result

• Datacenter passes everything. Use datacenter. You get a faster job and a monthly price that does not grow with the pages you fetch.
• Datacenter gets 403 or 429 immediately. The site is filtering on ASN. Go residential, or ISP if you need one address to stay put.
• Datacenter works, then degrades after a few hundred requests. This is rate limiting, and IP type has little to do with it. Slow down and spread the load before paying more. See the blocking checklist.
• Both fail. Your problem is not the IP. It is almost certainly headers or TLS fingerprint, and buying pricier proxies will burn money without fixing it.

Cases where datacenter is fine and people assume otherwise

In practice, datacenter tends to be enough more often than reputation suggests:

• Public APIs that authenticate you by key instead of by IP
• Documentation sites, forums, and most content sites
• Government and open-data portals
• Smaller ecommerce sites without a commercial bot-detection vendor
• Anything you are polling on a schedule instead of crawling hard

And the ones where it usually is not:

• Large retail and travel sites with commercial bot protection
• Sneaker and ticket platforms
• Social platforms and their app APIs
• Anything with a login you intend to keep

Cost per page you got

Price per gigabyte and price per IP are both the wrong comparison. The number that matters is what each page you successfully fetched cost you, and you can only get it by running the test above on your own target:

• Residential: average bytes per request (request plus response, which is how we meter), times your rate per gigabyte, divided by the share of requests that succeeded.
• Datacenter or ISP: what the addresses cost for the month, divided by the pages they fetched in it.

A cheap line with a terrible success rate can lose to an expensive one that works. Most of the time, once people measure, the cheaper line wins on this number too, which is why we keep telling people to start there. Measuring bandwidth before you pay shows how to get the bytes-per-page figure for free.

ISP is a third thing, and it is cheaper than you expect

ISP proxies are static addresses that live in a datacenter but whose whois record names a consumer ISP. You get the residential ASN that most filters check, at datacenter speed, and the address does not change until you change it.

That combination is the right answer more often than people realise. If a target blocks hosting ranges and you need to look like the same visitor across a login, a cart and a checkout, rotating residential works against you: every request arriving from a different address is its own signal. A static ISP address does not have that problem. ISP vs datacenter proxies has the test that tells you whether your target needs one.

ISP is rented per IP on the same plans as datacenter. It costs more than datacenter, $2.66 a month at 10 IPs against $1.58, because the address is registered to a home ISP, but it is not priced like residential. The bandwidth is ours, not rented from someone’s home connection, so you are paying for the ASN and not for the pipe. The same sum works for ISP: past 2.13 GB a month through one address, a dedicated ISP IP costs less than that traffic would on residential.

Prices for all three are on the pricing page, and each product page says plainly what the thing is bad at. If you would rather answer this with a flowchart, which proxy type do I need has one.