Datacenter proxies versus residential: what to honestly expect from hosting addresses
Comparisons on this topic are almost always written by someone selling residential addresses, and they reach a predictable conclusion. We sell datacenter and ISP subnets; we have no residential or mobile addresses at all — and that is the first line, not a footnote. Now to the substance: how the type is determined, what it changes, and which jobs our product loses.
How an address's type is actually determined
The common simplification is that the type comes from whois. It does not, and the difference matters. MaxMind describes its method as analysing registration information, traffic and usage patterns: registration is one of three inputs. The same vendor also exposes fields that are measurements rather than lookups — for example a score for how static an address is, and an estimate of how many users sit behind it.
The result lands in dedicated fields, which is the most practical thing to know here. MaxMind has a user_type enum of fifteen values in which hosting and residential are distinct values, and a datacenter value does not exist at all. IP2Location gives the same role to usage_type: a data centre is the code DCH, a fixed-line provider is ISP, mobile is MOB; the field ships in databases DB23–DB26 and in the proxy-detection line PX6–PX11. You can also check which range your address belongs to, and who holds it, without any paid database.
# Who the range is assigned to, and which autonomous system is behind it
whois 203.0.113.24 | grep -iE 'netname|descr|origin|country'Why rotation does not change the network
This is the most common misunderstanding, and it is why people buy the wrong product. Changing the address inside one subnet changes the address — and only that. Internet routing does not work in addresses but in prefixes: a network is announced, and every address inside it inherits the same autonomous system. Rotating within a /64 changes neither the announced prefix nor the ASN.
There is a flip side too: reputation databases aggregate records by subnet rather than by individual address. So “many addresses” and “many networks” are different goods, and the first does not substitute for the second. On size, incidentally: the global routing table generally does not accept prefixes longer than /48, but that is an operational convention decided per peering, not a protocol prohibition.
Where hosting addresses do the job
- Your own and partner APIs, where you are admitted by key rather than by address type.
- Multi-threaded software and scrapers on tolerant targets: the ceiling there is concurrent connections, and thread-based billing matches the job.
- Price and assortment monitoring on platforms that score behaviour rather than the address's origin.
- Checks from different countries and load tests, where what matters is simply a different address in a different region.
Where they lose outright
If a target filters hosting autonomous systems by policy rather than by behaviour, rotation does not help — however many addresses you take, they all remain in the same networks with the same ASNs. This is not a question of our pool's quality: it is precisely what such a policy is built to exclude. For jobs like that our product is the wrong fit, and the right answer is not to try more addresses but to look for a different type.
What behaviour decides, rather than the address
For modern anti-bot systems the address is only part of the input. Cloudflare, for instance, describes its ML engine as using various request features: headers, session characteristics and browser signals. Reading that as “the address does not matter” would be a stretch, for two reasons: it describes one engine among several, and the same company publicly noted that it increased the weight of network-level traffic characteristics. It publishes no proportions, and numbers must not be invented here.
The practical point is different: if behaviour is what excludes you, extra addresses change nothing. How to tell the two apart from response codes and from measurement layers is covered in the separate articles on 403 and on how sites detect proxies.
The billing difference is part of the answer too
Residential addresses are almost always sold by the gigabyte; ours by concurrent connections, with no traffic cap. That is not “cheaper”, it is a different unit, and the crossover is computable. Estimate your average response size and request count: a kilobyte across a million requests is roughly a gigabyte. If responses are small and requests many, thread billing wins by a wide margin; if responses are few but heavy, the difference disappears.
Three phrases you will not see here
- “Like residential” — they are not like residential. The databases keep hosting and residential as distinct values of a dedicated field.
- “Indistinguishable from a home address” — they are distinguishable, and it takes one query against any of those databases.
- “Residential quality” — the phrase has no technical meaning; it exists to sell a hosting address at a residential price.
We write this plainly for a practical reason: we offer a one-hour no-questions refund, and buying the wrong product is a refund rather than revenue. It is easier to say so up front.
Hosting addresses for the jobs where they work
Rotating IPv6 from 650 ₽ for 50 threads, IPv4 from 1375 ₽ for 100, static IPv4 from 300 ₽. Traffic on rotating plans is not metered.
View pricingProxies for this job
Check it with our tools
Read next
- How sites detect proxies: three measurement layers, and which one a proxy changesA target measures three independent layers: the address, the TLS handshake and HTTP behaviour. Through a tunnel, TLS and HTTP keep describing your client while the address and TCP describe the proxy machine. That split is itself a signal.
- IPv6 or IPv4 for scraping: what is cheaper and when it failsIPv6 costs less at the same thread count but hits one limitation on the target's side. How to check in a minute whether it will work.
- Threads or gigabytes: working out which billing unit is cheaper for your jobWe bill by concurrent connections; most residential providers bill by traffic. Those are different units, and the crossover is computable. How to measure your own profile before ordering, and how not to pay for idle threads.