Four Dollars a GB to Rent a Human Identity
Jiang Wu · 2026-08-11
For four dollars a gigabyte you can rent a stranger's home broadband address. Traffic leaves from that household's router, runs over a real ISP line, and looks like an ordinary person browsing to any website's detection system. The service is called a residential proxy. At the end of 2025, Oxylabs, a Lithuanian company in this business, took $130 million from Warburg Pincus at a post-money valuation of $3.6 billion, its first outside money in ten years. Around the same time Israel's Bright Data disclosed annualized revenue above $300 million, up more than 50% year over year. Together the two earn under $700 million a year, with an implied valuation near $7 billion. The line about Bright Data's 50% growth carries a qualifier: thanks to AI. NetNut, a listed proxy provider, now has AI companies as its largest customer group, ahead of e-commerce. Rayobyte's data-center proxy revenue fell from 85% in 2023 to 42.4% in 2025, while the more expensive residential proxies grew faster on AI demand. After a new set of buyers arrived, the valuation of a business that has existed for over a decade rose by an order of magnitude. The new buyers don't want bandwidth. They want what comes attached to it, a human identity. Trust on the internet rests on IP addresses The most basic way a website decides whether a visitor is a person or a machine is the IP address. Data-center IP ranges have long been flagged, and requests from those addresses are probably crawlers. This worked in the past because most legitimate traffic really did come from homes and offices. Over the past two years this trust check has tightened sharply. Cloudflare's data shows that by May 2026, automated requests made up 57.5% of global HTML web traffic and human traffic 42.5%. For the first time machine traffic exceeded human traffic. AI crawlers account for about 20% of verified bot traffic, with GPTBot second (11.48%) and ClaudeBot third (9.73%). More than 2.5 million sites block AI training crawlers outright. Cloudflare made AI crawler blocking the default for new domains in July 2025, and within five months its customers blocked 416 billion requests. The more doors close, the more a human identity is worth. Data-center IPs are mostly blocked, and getting around anti-crawling means using the broadband addresses of real home users. Residential proxy traffic runs over real ISP lines and looks no different from ordinary users in detection systems. Training a frontier model needs text, code, and multilingual content across billions of web pages. The data sits on the public internet, and to get it you first have to look like a person. Behind every IP stands a real person Oxylabs publicly claims an IP pool of 175 million, Bright Data 72 million, and SOAX 155 million. A mid-sized provider like BestProxy has 80 million. A few leading providers turn over several hundred million residential IP addresses.
This is a paid article. Subscribe to read the full text.
Home · All articles · Topics · 简体中文