Stats, analysis, notes from the lab.

Google Cloud publishes 19,385,984 IPv4 addresses. Query only AS15169 and you get 4.2M. Query all five Google cloud ASNs and you get 55.9M - more than 3x the published file. AS396982 is the number most tools completely miss. Updated weekly from live data.

Azure publically disclosed 108,967,860 IPv4 Ips across 46,992 CIDR. That more than the sum of AWS + GCP CIDR blocks . BGP via ASN - AS8075 shows 81.8M.

AWS publicly documents 188,682,587 IPv4 addresses. BGP routing tables show 244,140,288. The 55 million gap is the story no other page tells. Updated weekly from live data.

Technically, there are 4.29 billion IPv4 addresses. If they remove the private, reserved, and special-purpose blocks from that, we are approximately left with 3.72 billion Ips that are publicly routable.

Three runs of the same Wikipedia scrape across two days: httpx swung 5× in latency, primp and nodriver flipped pass rates, browser engines held flat at ~16s. Single-run benchmarks are snapshots — only reruns show the real signal. Data from 7 engines: httpx, curl_cffi, primp, camoufox, patchright, nodriver, cookie_replay.

The honest floor my Google scraper actually ships at. 4/9 pass on Decodo rotating, ~1.95 MB per query, 15-30× more expensive than Serper. Every optimization I tried on top made it worse.

How I'm scraping Google SERPs with Camoufox, minted identity, and Decodo residential proxies. 20/20 pass at ~$0.0007 per SERP, with the full journey from curl to the mechanism that worked.

Interleaved A/B of scroll + dwell + fake mouse moves vs plain baseline on Decodo rotating. Zero pass-rate improvement, 68% more bandwidth, 3-6× latency. Why the mechanism can't work.

My full ladder of what worked and what didn't scraping Google: plain curl, curl_cffi × 17 TLS profiles (30/30 blocked), Playwright, and the stealth recipe that finally pulled a real SERP on my home IP.

4/9 pass on Monday, 0/9 on the next check: same code, same keywords, same pool. Pool reputation is a shared resource you don't control, and it moves faster than any browser-side fix.

The mechanism that beat everything else: Camoufox + one minted Google session + parked tab + in-page fetch(). 20/20 pass on Decodo rotating, ~$0.0007/SERP steady state. At parity with Serper.

What happened when I pointed a working home-IP stealth-Chromium scraper at Decodo residential. 0/3 silent-hang at TLS, then 3/3 /sorry/ on Firefox. The block is at least two walls deep.

Playwright adoption numbers from npm, GitHub, TestGuild, 6sense, Theirstack, and PyPI — all in context. 141M weekly npm downloads, 99,500 GitHub stars, 4,484 to 47,694 companies depending on who you ask.

Web scraping stats for 2026. 2,471 numbers cross-checked and cross-tabulated by hand across Cloudflare Radar, DataDome, F5 Labs, Akamai, and independent site owners.

The world creates roughly 402.74M TB a day, 181 ZB in 2025. Over 842 stats cross-checked and cross-tabulated by hand: per-day activity, the recursion problem, and what the top-line still misses.

Ericsson says 21 GB per smartphone. OpenVault says 767 GB per household. Consumer sources say 5 to 10 GB. Over 663 stats cross-checked and cross-tabulated by hand: how the same question yields five right answers for five different populations.

Goldman puts 2026 AI investment at $1T; McKinsey $5.2T through 2030; Reuters $1.09T in uncommenced leases. 868 stats cross-tabulated by hand across the seven companies driving the buildout.

US data-center counts range from about 2,100 to 8,049 depending on the provider. Over 598 stats cross-checked and cross-tabulated by hand: the 2026 map, the 6.2× planned pipeline, the seven-month opinion swing, and the grid constraint.

The viral 500 ml per query figure is off by roughly two orders of magnitude. The 2026 answer sits between 0.3 and 6 ml — here's how the wrong number spread, and where AI water use actually goes.

AI-related bots hit 34.30% of all bot traffic on Cloudflare's global CDN in the 28 days to August 1, 2026: for the first time larger than search-engine crawlers (26.13%). Four numbers behind the tilt: scale, growth, top operators, and where AI bots actually land.