The era of human dominance over the internet is coming to an end far faster than anticipated.
Global content delivery network giant Cloudflare (NET) disclosed during its Q2 2026 earnings call that over 50% of traffic on its global network no longer originates from human users, but from various bots and automated programs. This marks the first time in the company’s recorded history that machine traffic has eclipsed human traffic. Chief Financial Officer Thomas Seifert, while explaining this trend to investors, stated bluntly that if current growth rates are extrapolated linearly, human activity on the internet could shrink to a “statistical rounding error” within five years.
This crossover moment arrived a full year earlier than Cloudflare CEO Matthew Prince himself had expected. Prince initially predicted that human traffic would be overtaken by late 2027, later revising his forecast to early 2027. However, in May 2026, data from the technical team confirmed the milestone had already been reached.
The core driver behind the explosive growth in machine traffic is not billions of new internet users, but the exponential expansion of AI agents. Prince used an analogy to explain the fundamental nature of this traffic transformation: when a person wants to buy a digital camera, they might manually visit five websites to compare prices. But if they delegate the task to AI, the agent might crawl 5,000 websites to gather information. Every single human intent is amplified into thousands of machine requests. This signals a wholesale shift in internet usage patterns from “going yourself” to “sending a machine.”
According to Cloudflare’s public data monitoring platform Radar, this traffic comparison only counts HTTP requests returning HTML content, excluding images, scripts, and other ancillary resources as well as API calls. The classification logic is based on a “Bot Score” system, where scores of 1 to 29 are deemed “likely automated,” and scores of 30 and above are classified as “likely human.” Therefore, a more precise formulation is: on Cloudflare’s network, among web page requests defined by specific criteria, machines now account for half.
Within this vast machine traffic landscape, the fastest-growing and most disruptive segment is undoubtedly AI traffic. According to the “2026 AI Traffic and Cyber Threat Benchmark Report” released by cybersecurity firm HUMAN Security, while training crawlers still account for roughly 67.5% of AI-driven traffic, their share is declining. Real-time scraping crawler traffic surged 597% year-over-year, while “AI agent” traffic—capable of autonomously clicking links, filling out forms, and even completing checkout—skyrocketed by 7,851% over the past year.
The sources of this machine traffic are highly concentrated. Data shows OpenAI alone accounts for roughly 69% of the total, Meta approximately 16%, and Anthropic around 11%, with the combined share of dozens of other companies falling below 5%. This means a handful of AI giants are reading the internet on behalf of all humanity. Prince views this as a platform shift on par with the transition from desktop to mobile, with the key difference being that the previous transformation took a decade, while this one unfolded in mere months.
When the consumer of content shifts from humans to machines, the foundational business logic of the internet begins to crumble. For nearly three decades, the internet’s free ecosystem has rested on an implicit contract of “advertising in exchange for content”—users watch ads, websites earn revenue, and that revenue subsidizes free content. Machines, however, do not watch ads, nor do they make impulse purchases. On the earnings call, Prince declared that the advertising-supported model defined and dominated by Google is approaching its end. The new paradigm he proposed is “machine-to-machine, pay-per-request”—charging a fraction of a cent for each AI-initiated request. Cloudflare has already begun laying the payment infrastructure for this, with Prince even vowing to build a payment network three orders of magnitude larger than Visa, specifically designed to serve the machine economy.
Data from payments company Stripe (STRIPE) provides corroborating evidence: 70% of command-line requests on its platform are now initiated by agents. Investment bank Pitchbook has defined this trend as the “machine economy,” and startup Forsy estimates that global “agent GDP” currently stands at roughly $36 billion annually. However, considering the total volume of tasks AI could potentially handle, this represents a penetration rate of only about 1%.
Yet, while this structural shift in traffic unleashes immense productivity, it is also boomeranging on tech giants with extraordinarily high costs.
According to the Financial Times, citing internal Amazon (AMZN) employee sources, the company recently encountered severe cost overruns when attempting to use Anthropic’s Claude Sonnet model to populate detailed author information on its website. This seemingly simple task ultimately cost $1.8 million, exceeding the budget by 860%, and the problem was only discovered after five months. The project ultimately failed to deploy successfully. Based on Claude Sonnet’s public pricing of $3 per million input tokens and $15 per million output tokens, $1.8 million implies consumption of up to 600 billion tokens—equivalent to twice the entire training corpus of GPT-3.
This is not an isolated incident. Inside Amazon, similar episodes of AI agent-related budget blowouts have occurred multiple times. Reports indicate that some employees spontaneously created an informal usage leaderboard called “KiroRank,” which inadvertently encouraged competition to inflate personal data consumption; the leaderboard was later shut down. In April of this year, a Meta (META) employee also built a leaderboard called “Claudeonomics,” aggregating AI usage data from over 85,000 employees. Within 30 days, Meta employees’ total token consumption climbed to 73.7 trillion, translating to a monthly bill of approximately $221 million based on Anthropic’s public pricing. By June, Meta formally issued a memo to roughly 6,000 employees, announcing token usage limits and the establishment of a central platform to monitor AI spending across teams in real time.
Ride-hailing giant Uber (UBER) burned through its entire annual AI programming budget in the first four months of 2026, subsequently imposing a monthly spending cap of $1,500 per employee per tool. Its chief operating officer acknowledged that the link between token expenditure and measurable output has yet to be established. An industry survey shows that only 26% of enterprises have comprehensive visibility into their AI costs. OpenAI CEO Sam Altman also admitted in June that AI cost issues, which no one was asking about at the beginning of the year, have now evolved into a major challenge. He revealed that the highest-usage individual user within OpenAI consumes roughly 100 billion tokens per month.
Despite the growing pains of cost overruns, tech giants show no signs of slowing their march toward automation. In its latest Q2 2026 earnings report, Amazon disclosed that its cloud computing division, AWS, generated net sales of $42.2 billion, up 37% year-over-year, and contributed roughly 60% of the company’s total operating profit of $27.5 billion—even though AWS revenue accounts for only 21% of total company revenue. Amazon CEO Andy Jassy announced that the company’s 2026 capital expenditure is projected to reach approximately $220 billion, a nearly 60% year-over-year increase, with the vast majority directed toward AWS, proprietary AI chips, and power infrastructure. This represents the highest single-year capital expenditure among the world’s mega-cap companies.
Meanwhile, Amazon has conducted two rounds of large-scale layoffs since October 2025, cumulatively cutting roughly 30,000 positions. Its robotics division aims to achieve approximately 75% warehouse operations automation around 2033, which implies that by 2027, Amazon will employ roughly 160,000 fewer people in the United States. Nobel laureate in economics and MIT professor Daron Acemoglu warned that if Amazon’s automation blueprint materializes, America’s largest employer would transform from a net job creator into a “net job destroyer.”
From the upheaval in traffic composition, to the shaking of business models, to the struggle for corporate cost control, the internet is undergoing a profound restructuring. The last internet was sustained by humans watching ads. The foundation of the next internet may well be built on an economic model where machines pay tolls. As Prince put it, the visitors have changed, and the internet must be redecorated to suit the tastes of its new guests. Humans will not disappear from the internet, but from now on, we are no longer the only visitors—nor the only paying customers.