What the two things actually are
A datacenter address belongs to a hosting provider. It is allocated in blocks, it lives in a range that anyone can look up, and it is what you get when you rent a server. It is fast, because the machine sits in a well connected facility, and it is cheap, because creating another one costs the provider almost nothing.
A residential address belongs to a consumer connection: a home broadband line, sometimes a mobile network. Traffic exits through a real subscriber connection, which is why it looks like a person. It is slower, it is far more expensive, and the supply is finite, because there are only so many households.
That difference in origin is the whole story. Everything below follows from it.
How a site tells them apart
Address ranges are published. Anyone can check which organisation owns a block and whether that organisation is a hosting company. A site that wants to reduce automated traffic starts there, because it is the cheapest check available: if the request comes from a known hosting range, treat it with more suspicion, show a challenge, or refuse.
This is why a datacenter address can be blocked before a single other signal is examined. The block is not detecting your crawler. It is detecting that you are not a household.
The comparison that matters
| Item | Datacenter | Residential |
|---|---|---|
| Speed | Fastest available, low latency, high throughput | Slower, and it varies with the household connection |
| Cost per request | Low | Several times higher |
| Block rate on defended sites | High, often before any other check | Low |
| Block rate on ordinary sites | Very low | Very low |
| Supply | Effectively unlimited | Finite |
| Best for | APIs, documentation, most company sites, high volume of undefended pages | Marketplaces, search engines, social platforms, travel and retail |
The rule that saves the budget
Start every target on datacenter addresses. Measure the success rate over a real sample, not over three requests. If it holds above ninety percent, stay there and keep the speed and the price. If it collapses, move that target to residential and leave the rest of your crawl where it was.
Most crawls are a mix. A price monitoring pipeline reading forty retailers usually runs thirty of them on datacenter addresses without any trouble and needs residential for the handful with serious protection. Paying residential prices for all forty is a common and expensive habit.
Where address type stops being the problem
Once a target rejects datacenter addresses and residential ones get through, teams assume address type is the whole game. It is not. A site with real protection reads the TLS fingerprint, the header order, whether JavaScript ran, how evenly your requests are spaced and whether the session has any history. A residential address attached to a client that fails every one of those checks still gets blocked, and now you are paying more for the privilege.
The signals and what each one gives away are listed on the web scraping proxy page, and the mechanics of rotation are in rotating proxy explained.
Two mistakes worth naming
The first is rotating on every single request against a site that tracks sessions. A visitor who changes country between page one and page two of a checkout flow is not a visitor. Sticky sessions exist for that, and holding one address for the length of a flow is often what fixes a stubborn target.
The second is buying pools by the gigabyte and then discovering the bill is driven by retries. Every blocked attempt, every challenge page, every redirect chain is bandwidth you paid for and data you did not get. That is the argument for billing on successful responses instead: the retries stop being a line item you fund.
What this looks like as one parameter
On a crawl you run yourself, this is two contracts, two gateway strings, a router that picks between them and a spreadsheet tracking which target uses which. On an API it is one field in the request body, chosen per call, alongside the country. The pool is included in the plan rather than metered separately, so the choice above is about success rate rather than about your budget.
The short version
Datacenter first, because it is fast and cheap. Residential where the numbers say you need it, target by target. And when residential does not fix it either, the problem was never the address, it was everything else your client was announcing about itself.