Examples
An ISP proxy is usually the thing you try when plain datacenter IPs work in testing but fall apart once volume goes up.
- Datacenter proxy: cheap, fast, blocked more often
- Residential proxy: harder to block, slower, usually more expensive
- ISP proxy: server speed with a reputation that often looks closer to residential
# Example request sent through an ISP proxy
curl -x http://user:pass@isp-proxy.example:8000 https://httpbin.org/ip
import requests
proxies = {
"http": "http://user:pass@isp-proxy.example:8000",
"https": "http://user:pass@isp-proxy.example:8000",
}
r = requests.get("https://httpbin.org/ip", proxies=proxies, timeout=30)
print(r.json())
In practice, teams often route targets like this: - low-friction pages: datacenter - medium-friction pages: ISP - high-friction or very sensitive targets: residential or browser-based scraping
Practical tips
- Don't treat ISP proxies as magic. They still get blocked, they still burn, and they still need rotation and retry logic.
- Use them when the target is sensitive to obvious datacenter ranges but doesn't justify full residential cost.
- Watch the economics:
- datacenter is usually cheapest
- ISP is often the middle ground
- residential gets expensive fast at scale
- Test by target, not by marketing label. One site may accept ISP IPs just fine, another may flag them immediately.
- Measure the things that actually matter:
- success rate
- challenge rate
- cost per successful page
- latency
- how often you need to swap providers
- If you're already juggling multiple proxy types across targets, that's usually where a routing layer starts making sense. ScrapeRouter exists for exactly that kind of mess: picking the right path per request instead of hardwiring one provider and hoping it keeps working.
Use cases
- Retail scraping: product pages where datacenter IPs start getting throttled once concurrency increases
- SERP collection: workloads that need better trust than datacenter, without paying residential rates for every request
- Account-light flows: public pages with moderate bot protection, where residential is overkill but raw server IPs die quickly
- Fallback layer: when a target mostly works on datacenter and only certain routes, geos, or volumes need something less obvious