‘Humans will be a rounding error on the internet,’ says Cloudflare exec

Cloudflare Chief Financial Officer Thomas Seifert predicts machine-generated traffic could reach 1,000 times human traffic within five years, after the company measured automated systems overtaking people in May.

Seifert made the forecast during Cloudflare’s second-quarter earnings call Thursday. The company had expected non-human traffic to exceed human traffic in 2027, but its measurements moved that milestone to May 2026.

“To give you a sense of how this trend is playing out, and with the big caveat that I have called it wrong at every point along the way, if the current trends continue, we think in five years, non-human traffic will be as much as 1,000 times as much as human traffic,” Seifert said.

He said human traffic would remain present. Automated systems, including AI services, would generate data at a much faster pace.

AI changes the traffic mix

AI agents, model training systems, crawlers and automated services now send a growing share of internet requests. Those systems can open pages, query APIs, download files and exchange data at a scale that human users cannot match.

Cloudflare operates a global network that handles web requests, domain traffic and security filtering. The company’s network services give it a view across a large portion of internet activity, although its measurements represent traffic that reaches Cloudflare rather than every packet on the internet.

More machine traffic will force network operators to increase capacity and improve efficiency. Seifert said Cloudflare would need to process far more activity if current growth continues, and he promoted the company’s services as a way to manage that demand.

Automated traffic also creates security problems. Attackers can use bots to scan sites, test stolen credentials, scrape data and probe APIs. AI systems can increase the volume and speed of those operations, giving security teams more events to inspect and more infrastructure to protect.

Cloudflare has built products for bot management, application security and distributed denial-of-service protection. Its Bot Management platform classifies automated requests and helps site operators decide which traffic to allow.

Cloudflare reports revenue growth

Cloudflare reported $696 million in second-quarter revenue, a 36% increase from the same quarter a year earlier. The company recorded a $205.7 million loss, more than three times the loss from the prior-year period.

Investors focused on revenue growth and new large-customer wins. Cloudflare shares rose 16% in after-hours trading, reaching a new high and bringing their year-to-date gain to 68%.

The company expects capital spending to reach 14% to 15% of its projected annual revenue range of $2.865 billion to $2.87 billion. That puts planned capital expenditure near $430 million.

Cloudflare’s capital budget remains small beside the spending plans of major cloud providers. Amazon Web Services and Google Cloud operate at far larger revenue scales, while the leading AI companies and infrastructure providers have committed trillions of dollars to data centers, GPUs and related equipment.

Prince rejects commodity compute

Cloudflare CEO Matthew Prince used the spending comparison to criticize cloud providers that buy large volumes of standard servers and rent access to them.

“If you’re selling what is just 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,” Prince said.

Prince said Cloudflare wants to extract more work from each dollar of capital spending. The company focuses on memory, storage, scheduling and software efficiency across its distributed server network.

He also criticized hyperscalers for GPU utilization rates that he described as “super-low.” Cloud customers must manage much of their own hardware use, while Cloudflare sells completed work through services such as its serverless platform.

Traditional cloud providers buy servers, lease capacity and seek several revenue cycles from the same equipment. Cloudflare takes responsibility for running workloads across its network, Prince said, which gives the company a stronger incentive to keep each machine busy.

That model supports Cloudflare Workers and other edge services. Developers can run code near users without managing a full virtual machine or owning a data center. Cloudflare supplies the network, deployment system and execution environment, while customers pay for requests and compute use.

The hardware trade-off

Cloudflare’s model can reduce infrastructure waste, but it depends on workload patterns and software design. A distributed platform can place short requests near users and share equipment across many customers. Long-running AI training jobs require different hardware, memory capacity and network links.

Cloudflare’s strategy also leaves it exposed to rising demand for servers, storage and network equipment. Better utilization can reduce the number of machines the company needs, but it cannot remove the cost of expanding capacity as automated traffic grows.

The traffic forecast adds another pressure point. If machine-generated requests multiply at Seifert’s projected rate, operators must distinguish useful work from repetitive scraping, abusive automation and attack traffic. That classification requires network telemetry, identity signals and security rules that can keep pace with new AI services.

Human users will continue to create the requests that drive much of the web’s economic value. Machines will generate more of the background traffic that fetches, transforms and redistributes information. Cloudflare’s executives expect that gap to widen as AI systems connect to websites, applications and other software.

Cloudflare’s investor relations page provides the company’s earnings materials and financial filings.