Cloudflare released its Q2 FY2026 earnings after market close on August 6. Revenue was $696.1 million, up 36% YoY. Gross margin hit 73.1% — the first sequential improvement in eight quarters. Non-GAAP operating income was $96.1 million (+33% YoY), and free cash flow was $56.4 million (+69% YoY). RPO (remaining performance obligations) reached $2.732 billion, up 38% YoY. Large customers (annual spend >$100k) grew 27% to 4,698 — a net addition of 986 over the past 12 months, with records set at every tier ($100k, $500k, $1M, $5M+). Dollar-based net retention was 120%, up 2pp QoQ and 6pp YoY. New paying customers exceeded 80,000, up 74% YoY.
But the real highlight of the earnings call wasn't the financials — it was a remark from CEO Matthew Prince during the Q&A. When asked in November 2025 when non-human traffic would exceed human traffic, his team ran the data and confirmed: second half of 2027. When asked again in March 2026, the team re-ran the numbers and moved the forecast up to the first half of 2027. Then in May 2026, the team came to Prince's desk and said: "You won't believe this — non-human traffic has already surpassed human traffic."
Prince's exact words on the call: "In other words, humans will be a rounding error on the Internet, not because human traffic goes down, but that's just how fast we're seeing non-human traffic grow."
He then offered a more aggressive prediction: if the trend holds, non-human traffic will be 1,000x human traffic within five years.
What makes that number plausible is a fundamental difference in agent behavior: a human shopping for a camera checks five retailers; an AI agent might check 5,000 websites. With millions of users simultaneously having assistants compare prices, research topics, and summarize articles, traffic scales exponentially. Cloudflare offered a supporting data point: nearly 2 million net new developers added in Q2 alone — more than all of 2025 combined.
That number signals something bigger: Cloudflare has repositioned itself as the internet's foundation for the agent era. Three pieces of evidence:
First, a full product suite launched during Agents Week. Cloudflare released Cloudflare.pay, ADLC (Agent Development Life Cycle), and Agents in one week. Cloudflare.pay is a payment layer designed for agent traffic — each request through Cloudflare's network is billed at "fractions of a cent," replacing the ad-defined internet business model of the past 27 years with pay-per-request. Prince said it plainly on the call: "I think that the business model of the Internet for the last 27 years has been largely defined by advertising and really defined by Google. I think the business model of the next 27 years of the Internet is going to be very different."
Second, 80%+ of major AI companies are already Cloudflare customers. Prince repeatedly cited this figure in the Q&A. The subtext: all major model companies and agent platforms route their external access through Cloudflare's network, making Cloudflare the "gateway + billing layer" of the agent era — a position comparable to Google's in the search advertising ecosystem circa 2003.
Third, CapEx at 14–15% of revenue (~$430 million) — low for a cloud provider. Prince countered the hyperscaler comparison: "We know what the typical kind of inference loads are at the hyperscalers, and their GPU utilization. It's super low." "If we can, and in some cases, get 10 times as much utilization out of every CapEx dollar, then we can play these games and continue to grow revenue." Cloudflare's model isn't selling GPU boxes for customers to optimize themselves; it's scheduling workloads onto existing hardware to extract 10x efficiency per CapEx dollar.
Several sub-points worth unpacking:
First, AI traffic isn't neutral. Prince split AI traffic in two: "For a portion of it, we block it proactively. We don't charge customers extra, because we think that's what a security company should do." The typical malicious traffic is AI companies scraping ad-dependent media content. This boundary clarifies Cloudflare's agent-era role — both agent gateway and copyright moat.
Second, losses are widening. Q2 net loss was $205.7 million, more than triple YoY — a direct result of large-scale infrastructure investment for the agent era. But with free cash flow of $56.4 million (+69% YoY), management raised full-year revenue guidance to $2.864–2.870 billion (32% YoY growth). This is "CapEx investment loss," not "operating loss."
Third, geographic balance. US: 51% of revenue (+41%); EMEA: 27% (+30%); APAC: 14% (+32%). Growth is well-balanced with no single-region dependency.
Placed in the broader early-August AI coding × embodied intelligence context:
- Agent gateway layer — Cloudflare's ADLC + Agents + Cloudflare.pay trio makes agent-era SDLC and payments platform defaults. This earnings call frames them within a revenue-to-traffic-ratio financial lens.
- Training data layer — non-human traffic is the primary vehicle for future training data collection. Controlling distribution and billing at this scale means bargaining power over AI companies' next-stage data costs.
- China comparison — Alibaba Qwen's integration into Apple Intelligence, Ant's Ling-3.0-flash end-to-end open-source pipeline, and ByteDance SeedRealtime full-duplex are all domestic optimizations of the "model + inference + cache" stack; Cloudflare is doing the same thing at the network layer — making the agent-era entry point, gateway, and billing a one-stop platform.
- Cloudflare official Q2 2026 earnings call transcript (Yahoo Finance / Quartr): https://finance.yahoo.com/quote/NET/earnings/NET-Q2-2026-earnings_call-662709.html
- Zhitong Caijing earnings analysis (Chinese): https://www.zhitongcaijing.com/content/detail/1477439.html
- IT Home August 8 report (revenue/loss/Prince quotes): https://www.ithome.com/0/987/438.htm
- Biggo Finance earnings data roundup: https://finance.biggo.com/news/US_NET_2026-08-06
Three takeaways:
First, the agent internet's business model is being forcibly defined. The past 27 years ran on advertising; for the next 27, Cloudflare is betting on pay-per-request. The real risk: agent traffic is far more complex than human traffic (5,000 sites vs. 5), and Cloudflare must prove it can handle that request volume without quality loss. Prince's 10x CapEx utilization story is his answer to "why not build massive GPU datacenters" — the moat lies in scheduling, caching, and protocol handling, not bare-metal GPUs.
Second, "humans are a rounding error" isn't rhetoric — it's a roadmap. A 1000:1 non-human-to-human ratio in five years means Cloudflare's product strategy will keep tilting toward the agent side: CAPTCHA replaced by agent credentials, Bot Management evolving from "blocking malicious traffic" to "managing agents," CDN routing evolving from "page acceleration" to "tool-call acceleration." The pivot of recent years turned from "narrative" into "numbers" this quarter for the first time.
Third, practical implications for To B / agent startups in China. If you build agent products, "your agent's cost of calling third-party websites/APIs" will become a core operational metric within three years. If Cloudflare.pay's pay-per-request model works, a domestic equivalent will likely follow (Alibaba Cloud and Tencent Cloud edge networks already have reusable components). Calculating your product's "request bill" is more urgent than optimizing per-request latency.
Sources: