Cloudflare Is Betting That AI Agents Will Rewrite The Internet. The Problem Is The Price Tag.

Over 50% of the traffic on Cloudflare's network is now non-human. AI agents and automated systems have surpassed human visitors for the first time. That is the single most important signal from Cloudflare's Q2 earnings — not the 36% revenue growth, not the raised guidance, not the new AI platform suite. It is the signal that the internet's traffic composition has already flipped, and Cloudflare is the infrastructure company sitting directly on that transition.

CEO Matthew Prince called it a "fundamental rewrite of the Internet for machine-to-machine traffic" during the August 6th earnings call. Our strategic position leading the paradigm shift of the agentic Internet has never been stronger. That quote tells you how Prince sees the company's architecture advantage. The rest of the call was essentially evidence stacking to prove that this transition is real, measurable, and already driving the top line.

The architecture play: Workers, not GPUs

Most of the AI infrastructure narrative has centered on GPUs — Nvidia, AMD, custom silicon, hyperscaler data centers. Cloudflare is explicitly avoiding that race. The company declined to participate in the capital-intensive GPU infrastructure arms race. Instead, Cloudflare is building the orchestration and compute layer that sits above commodity compute. Workers — its serverless platform running JavaScript and WebAssembly across 330+ edge locations — is optimized for lightweight AI agent workloads. No cold starts, near-zero latency, usage-based pricing.

Put plainly, Cloudflare's architecture bet is that inference at the edge for millions of small AI agents is a bigger structural opportunity than training clusters for a handful of hyperscalers. Workers AI lets developers run ML models serverlessly without provisioning GPUs. The company is selling work done on its network, not commodity compute hours.

This matters because it means Cloudflare's margin profile should look different from the GPU rental businesses. Gross margins were 73.1% in Q2 — down 320 basis points year-over-year, but management attributes that compression to a mix shift toward higher-compute developer products, not to structural economics. Operating expenses fell to 59% of revenue, down from 62% a year ago. The operating leverage is showing up even while the company adds compute capacity.

The counterpoint is that gross margin compression is real and structural while AI traffic grows. Management says long-term unit economics for developer products remain identical once scaled. That's a claim that will need to hold over the next two quarters to validate the margin thesis.

The demand evidence

The Q2 numbers accelerate across every leading indicator:

• Revenue: $696.1 million, up 36% year-over-year, the third consecutive quarter of acceleration
• Current RPO... up 35% year-over-year
• Large customers... 4,698... up 27% year-over-year, with a record 986 net additions over the past 12 months
• Dollar-based net retention: 120%, up 6 points year-over-year
• Developers: 7.4 million, adding nearly 2 million in Q2 alone — more than the entire year of 2025 combined

Free cash flow was $56.4 million in Q2 (8% of revenue), up 69% year-over-year. That's a better metric than net income here — Cloudflare's GAAP results are distorted by a $150.7 million restructuring charge tied to the 20% workforce reduction announced in Q1. Excluding restructuring, GAAP net loss was approximately $18 million, nearly at breakeven.

The developer adoption number is the one I find most interesting. Nearly 2 million new developers in a single quarter tells you the bottoms-up demand from actual technology builders is accelerating faster than the enterprise sales motion. That's the kind of adoption signal that precedes revenue inflection by several quarters.

The new product layer

Cloudflare unveiled three infrastructure elements for what it calls the Agentic Internet: Monetization Gateway (allowing any resource — web pages, APIs, datasets, MCP tools — to be sold behind Cloudflare), Wallets (enabling autonomous payment by AI buyers), and cloudflare.pay (agent-friendly identity and micropayment rails). Cloudflare OS was open-sourced for organizational AI automation.

This is not a product announcement exercise. It is a positioning move. If AI agents begin transacting with each other — querying thousands of sites, calling APIs, purchasing data in real time — someone needs to provide the network, the compute, and the payment rails. Cloudflare is trying to be all three. The internet's business model has been ad-supported for 27 years. Prince's argument is that AI ends that model and replaces it with machine-to-machine commerce and micropayments.

Whether this materializes into meaningful revenue this year is uncertain. But the architecture is being laid now, and the 50% non-human traffic milestone means the underlying condition — AI agents consuming network resources at scale — is already happening.

The guidance tells the real story

Full-year 2026 revenue guidance was raised to $2.864–$2.870 billion (midpoint $2.867 billion), up from the prior midpoint of $2.809 billion. That implies approximately 32% full-year growth. Q3 guidance of $736–$737 million midpoint implies roughly 31% year-over-year growth.

That's a slight deceleration from Q2's 36% print. The market didn't seem to mind — shares jumped around 16% in premarket trading following the report. But I'd flag the deceleration. Three quarters of accelerating growth followed by a guided deceleration could mean the low-base effect of the pandemic-era growth cycle is finally working itself out. Or it could be prudence given the company's usage-based pricing model for Workers, which introduces revenue variability as more pay-as-you-go contracts replace annual commitments.

Full-year network capex is expected to be 14–15% of revenue. That's not negligible, but it's dramatically lower than the 25%+ capex ratios we're seeing from hyperscalers building GPU clusters. Cloudflare's network-as-computer model is more capital-efficient than the data-center buildout model. That's a deliberate architecture choice with margin implications.

The valuation question

Cloudflare's market cap is $106.9 billion. The stock trades at 42.52 times trailing sales. For comparison, Snowflake — also a high-growth infrastructure play — trades at 22.8 times sales. Even allowing for Cloudflare's faster growth and broader platform, that multiple gap is large.

I believe the market is rewarding Cloudflare for two things: the Agentic Internet narrative, which is genuinely a structural shift, and the evidence that this company is on the right side of the training-to-inference transition. But at 42.6x sales, the stock is priced for flawless execution on a thesis that will play out over multiple years, not quarters.

The company targets GAAP profitability by the end of 2028 and says it's pacing ahead of that goal. That's still two years away. Free cash flow margins are 12.5% on a trailing twelve-month basis — good for a growth company, but not at the 25%+ levels that justify the highest-end software multiples. The balance sheet is solid ($4.16 billion in cash and securities, $4.85 billion in total debt, current ratio of 182%), but the debt-to-equity ratio of 202% reflects aggressive leverage for a company that isn't yet profitable on a GAAP basis.

Jefferies maintains a Hold rating citing 38x EV/next-twelve-month revenue — still the highest multiple among major software infrastructure names. That rating captures what I see as the core tension: Cloudflare is a clear AI winner with multiple sustainable revenue drivers, but the valuation already reflects a very optimistic reading of the Agentic Internet timeline.

Where the capital goes

I still believe Cloudflare is positioned on the right side of the AI infrastructure transition. The shift from human-driven to agent-driven network traffic is real, the Workers platform is the right architecture for lightweight inference at the edge, and the developer adoption metrics suggest this is a bottoms-up motion that won't reverse. The company is avoiding the GPU arms race where margins compress and capex balloons — a deliberate and, in my opinion, correct strategic choice.

However, the question isn't whether Cloudflare will be important in three years. The question is whether a $106.9 billion market cap with 32% growth and negative GAAP earnings is the best place for capital today.

The 31% Q3 growth implied by guidance, versus the 36% we just got in Q2, signals that the acceleration curve may be flattening. Gross margin compression, while management says it's temporary, could persist as AI traffic share grows. And the 42.6x sales multiple demands that every quarter validates the Agentic Internet thesis — because if it doesn't, the multiple contraction risk is severe.

In my opinion, Cloudflare deserves a position in a portfolio focused on the AI infrastructure trade, but the size of that position should reflect the valuation. The thesis is long-term — 2028 profitability, Agentic Internet monetization, continued developer adoption. But much of the return from here is likely to be back-half weighted. A smaller allocation, held with patience, is the right approach. The market is paying premium pricing for a genuinely important structural transition, and that demands capital discipline.

What would change my view? If gross margins hold steady or expand in Q3 and Q4 despite growing AI traffic share, the margin thesis validates and the premium is justified. If Workers developer growth continues at the Q2 pace, the bottoms-up demand story strengthens. But if the guidance deceleration repeats in Q3 and margins compress further, the multiple will face pressure — and at this valuation, pressure means correction.

The debate is not whether Cloudflare remains important as AI agents rewrite the internet's traffic patterns. It is whether the return profile at $107 billion is still as compelling as what can be found elsewhere in the AI infrastructure trade.