All production scrapers use proxies. Not to hide identity — to not look suspicious to anti-bot systems. The two main types have different prices and different use cases. The choice is not obvious.
Datacenter proxy: cheap, mass, easily detectable. Residential proxy: more expensive, real IPs, harder to detect. Most projects need a hybrid.
Datacenter proxy — what it is
IPs from datacenters AWS / Google Cloud / DigitalOcean / Hetzner. Thousands of addresses, available instantly, cheap. Price: $1-5 / GB at major providers (Bright Data, Smartproxy, IPRoyal).
Problem: anti-bot systems know Bob from Iowa does not surf the internet from AWS Frankfurt. Detection rate for protected sites: typically 60-95% blocked.
Good for:
- Unprotected sites (gov portals, plain HTML, no Cloudflare)
- API endpoints without aggressive rate limiting
- Sites where the target lacks resources for anti-bot (small shops, blogs)
- High-volume low-value scraping (e.g. monitoring 1M products from generic catalogs)
Residential proxy — what it is
IPs from real ISPs (Comcast, Verizon, Vodafone, T-Mobile). Routed through devices of users who installed "free VPN" or "browser extension" in exchange for lending their IP. Controversial but legal.
Price: $5-15 / GB at premium providers (Bright Data, Oxylabs, Smartproxy residential). Sometimes up to $25/GB for top quality (sticky sessions, fresh IP rotation).
Looks like a regular user. Detection rate for Cloudflare protected: typically 5-25% blocked. For PerimeterX / Akamai: 15-50% blocked.
Good for:
- Premium e-commerce (Nike, Adidas, most fashion)
- Social media (LinkedIn, Instagram, X)
- Banking / fintech / SaaS with aggressive anti-bot
- Sneaker drops, ticketing, hype products
- Sites with geo-restrictions (need IP of specific country)
Mobile proxy — premium tier
Bonus category: IPs from mobile carriers (4G/5G LTE). Hardest to detect because carriers use carrier-grade NAT (thousands of real users share the same IP).
Price: $20-60 / GB. Slower than residential, but nearly undetectable. Used when even residential gets rate limits — typically gaming, sneakers, top retail.
Decision matrix
Choice per target:
- Datacenter — gov, public procurement, small e-commerce, most RSS/sitemap workflows
- Residential — premium e-commerce, social, most B2C, geo-targeting
- Mobile — when residential blocked / rate-limited, sneakers, ticketing, gaming
Most projects: hybrid. Cheaper targets via datacenter, more expensive via residential. Routing logic in code picks automatically.
Hidden cost: bandwidth
Proxy cost = price/GB × actual bandwidth. Heavy sites (Nike with 5 MB JS) use more than light ones (small shop with 200 KB HTML).
Practical calculation for monitoring 100 SKUs daily:
- Light site (~200 KB / request): 100 × 200 KB = 20 MB/day × 30 days = 600 MB/month = $3-9/month (residential)
- Medium (~1 MB / request): 100 × 1 MB = 100 MB/day × 30 = 3 GB/month = $15-45/month (residential)
- Heavy (~5 MB / request with JS): 500 MB/day × 30 = 15 GB/month = $75-225/month (residential)
Plus retry overhead — if 20% of requests fail, multiply by 1.2-1.5×.
Practical tips for the buyer
- Ask about proxy strategy before signing — which type, which provider, estimated bandwidth/month
- Cap monthly bandwidth in the contract — to not get a surprise $2000 invoice
- Audit cost after the first month — bandwidth is often higher than estimate
- Geo requirements put in the brief — if you must scrape from a specific country, mention explicitly
The point
Choosing the proxy type is not your problem as a client — it is the vendor's problem. But it is worth understanding what is behind the price difference. Unprotected target = $1-5/GB, works. Protected target = $5-15/GB, works 95%. Mobile-only target = $20-60/GB, works 80%. Each category has sensible use cases and sensible prices — watch out for vendors promising "everything for $50/month, for every site, with 100% accuracy".