Humans becoming 'rounding error' as machine traffic set to dwarf human internet use 1000x

Thomas Seifert, the chief financial officer of Cloudflare, stood on an earnings call in early August 2026 and made a prediction that should have sounded absurd if the numbers weren't already bearing it out. Within five years, he said, machine-generated traffic flowing across the internet would be roughly one thousand times larger than the traffic produced by human beings. Not because humans would stop using the internet—they'd keep clicking, streaming, scrolling at their steady pace. But the machines would simply overwhelm them, the way a river overwhelms a cup.

Cloudflare had been wrong before. The company had predicted that machine traffic would surpass human traffic sometime in 2027. But when the engineers actually measured the network, they found the crossover had already happened, back in May 2026. The machines had taken over without announcement or fanfare. So Seifert's new forecast came with a caveat: he acknowledged his track record of miscalculation. Still, if current trends held, humans would become what he called a rounding error on the internet—not a meaningful category, just a small remainder after the real numbers were tallied.

The driver of this shift was artificial intelligence. Machine-to-machine traffic, the constant chatter of AI systems training, inferencing, and communicating with one another, had begun to dwarf everything else. Seifert noted that if his prediction proved accurate, the internet infrastructure industry would need to become dramatically more efficient. He also warned that more machine traffic would likely mean more security threats—a problem Cloudflare positioned itself to solve.

The company's financial performance suggested investors were listening. Cloudflare reported $696 million in quarterly revenue, a 36 percent jump year-over-year. The company had also signed record numbers of large enterprise customers. Losses had more than tripled to $205.7 million, but the revenue beat and customer wins seemed to outweigh that concern. The stock rose 16 percent in after-hours trading, reaching a new high and marking 68 percent growth for the year so far.

What made Cloudflare's position distinctive was how little it was spending to achieve that growth. The company planned capital expenditure of roughly $430 million—about 14 or 15 percent of its projected revenue. That was a stark contrast to the hyperscalers like Amazon Web Services and Google Cloud, which were pouring trillions into AI infrastructure, buying servers in bulk and renting them back to customers. CEO Matthew Prince made the distinction explicit: Cloudflare wasn't in the business of selling commodity compute. "If you're basically letting an AI company use your balance sheet and your credit rating in order to buy servers that are the same as everybody else's servers, then that's just not attractive business for us," he said.

Instead, Cloudflare focused on efficiency—wringing maximum performance from every piece of hardware it owned. Prince observed that the hyperscalers had notoriously low GPU utilization rates because they sold the box and left it to customers to figure out how to use it. Cloudflare's model was different: the company did the work to optimize the equipment itself, building scheduling systems and efficiency layers that meant customers paid for actual work completed, not just rented capacity. The traditional cloud providers, Prince suggested, were trapped in a commodity business—buy a server, lease it back, try to get five turns of revenue from it. Cloudflare wanted no part of that race to the bottom. The company would stick to serverless offerings, where the margins were better and the business model more defensible.

What emerged from the earnings call was a picture of an industry in transition. The internet was becoming a machine-dominated space, and the companies that would thrive were those that could deliver efficiency at scale, not those that simply accumulated hardware. Cloudflare was betting it had found the right position in that new landscape—not a server vendor, but a platform that made the machines themselves work smarter. Whether that bet would pay off depended on whether the company could maintain its efficiency advantage as the traffic continued its exponential climb.