Web scraping proxies — start here before you spend a dollar

1. The two proxy types you actually need

95% of scraping jobs need either datacenter or rotating residential. ISP is a niche middle tier. Mobile proxies are only worth the cost for a handful of mobile-only sites.

• Datacenter: cheap, fast, easy to detect. Good for uncached public APIs, non-blocking sites, internal crawls. Expect $0.30–$1.50 per GB or $0.40–$2 per IP / month.
• Residential (rotating gateway): real home/office ISPs, expensive, much harder to block. Default pick for Amazon, Google, airlines, travel, heavily protected e-commerce. Expect $3–$12 per GB, bandwidth metered.

2. How to size your pool (the quick rule of thumb)

Most solo scrapers over-buy by 5–10× on their first project. Start with a 5 GB / 1 week plan, not a 1 TB / year plan.

• Estimate your sustained requests per minute (RPM) at your target concurrency.
• Assume each residential IP is good for 5–15 successful requests before it burns for that site.
• Multiply by 2 if the site uses aggressive bot management (Cloudflare Enterprise, Kasada, DataDome).

3. 3 questions every provider must answer honestly

• Is bandwidth billed per request byte or per egress byte? Per-request inflates cost 2–3× on compressed responses.
• Are failed / 4xx / 5xx requests billed? Good providers don't bill for upstream errors or proxy connect failures.
• What's the actual residential share? Many sell "residential" pools that are 40% ISP + datacenter. Always sample a random 50 IPs through IP Geo Lookup and Reputation.

4. Captcha "solver" is not a proxy feature

The single most misleading claim on provider landing pages. A good IP pool reduces how often you see captchas, it does not solve them. Budget separately for a solver if your target has aggressive protection.

5. Testing a new provider (2-hour checklist)

• Buy the smallest paid plan — never a "free trial" that whitelists only 3 IPs, those pools are usually hand-picked.
• Pull 20 random exits. Run each through Proxy Tester. Reject the whole provider if >20% come back as Transparent.
• Run a 30-minute small crawl at your target RPM. If success rate drops below 70% in the first 30 minutes, move on — don't "give it time".