Cloudflare Cut 1,100 Jobs for AI Agents — and Raised Its Growth Target to 50%

The numbers Cloudflare is actually trading on

According to the Yahoo Finance report and the investor coverage behind it, Cloudflare's most recent reported quarter brought in $639.8 million in revenue, up 34% year over year, with recent growth running at roughly 36% YoY. Guidance for the current quarter sits at a $736 million midpoint. Those are solid SaaS numbers — the kind a company defends. Instead, Cloudflare is repositioning.

The strategic signal is the "North Star" target: management lifted the long-term annual growth goal from 40% to 50% during September investor conferences, as reported by Investing.com. The mechanism behind that raise is not a new product line in the usual sense. It is a shift from predictable subscriptions toward consumption-driven AI inference revenue routed through Cloudflare's edge and its Workers developer platform. Management has openly acknowledged the trade-off: consumption models start out more variable than fixed seats.

1,100 jobs, and an internal AI usage spike of 600%

The restructuring is the part that should make European managers pay attention. Roughly 1,100 positions — about 20% of headcount — were eliminated as the company adopted what it calls an "agentic AI-first operating model," according to reporting by TNW. The stated trigger was internal AI tool usage rising 600%.

Read that sequence carefully: the automation came first, the reorganisation followed. This is not a company trimming costs while waiting for AI to arrive. It is a company that already changed how its own work gets done, then rebuilt the org chart to match. Whether that is visionary or brutal is a separate question — both can be true — but it is a template other infrastructure vendors will copy.

Markets liked the story. NET stock gained nearly 12% in early September, trading near $300 per share, after recognition of an OpenAI "Daybreak" security partnership. Security, routing and identity for machine traffic is the commercial pitch: if agents are the new clients, somebody has to check who they are.

Doing the math on "1,000 times more agents"

Cloudflare says it already processes hundreds of billions of agentic requests per month. Take 300 billion as a working figure — roughly 10 billion agent requests per day. Now apply the CFO's projection, reported by TechRadar: roughly 5.5 billion internet users × 1,000 gives about 5.5 trillion agents. If each of those made a single request per day, you would be looking at 5.5 trillion requests daily — around 550 times today's agentic volume. That is the scale being priced in.

Now flip it. If every dollar of the $736 million quarterly guidance came from agentic traffic, each request would be worth about 0.08 US cents — and that is a generous upper bound, because agent requests are only a slice of total revenue. The honest conclusion: traffic volume and revenue are not the same thing. The edge network earns fractions of a cent per request; the model behind the agent earns a tenth of a cent per turn and often much more. In our own tooling for AI Arena, trimming token counts in an agent loop moves the cost line far more than any network configuration ever has.

The European angle: Article 50 is no longer voluntary

Here is where the agentic traffic story stops being an investor narrative and becomes an operational one. Since 2 August 2026, the European Commission has full enforcement powers over general-purpose AI providers — binding oversight, mandatory technical audits, model recalls and direct financial penalties. On the same date, Article 50 of the EU AI Act became enforceable: labelling and watermarking of AI-generated text, audio, images and deepfakes is mandatory, not a nice-to-have. The era of voluntary disclosure frameworks is over.

For a European company deploying agents that browse, book, purchase and post, that creates a practical requirement the network layer does not solve for you: the identity question now has a legal dimension. Knowing whether a request came from a human, a disclosed agent, or an undisclosed one is no longer only a bot-management problem — it is a compliance problem with a paper trail. Cloudflare's agent-management tooling sits directly in that gap, which is precisely why the pitch resonates with regulated European buyers.

Two caveats worth keeping in mind. First, routing inference through a global edge does not answer GDPR questions by itself: where the model actually runs, what data leaves the region, and who is the controller for agent-generated output are still your problem. Second, Cloudflare's developer entry point — the Workers paid plan, listed at $5 per month, about €4.60 at a $1.08-per-euro rate, with a free tier covering 100,000 requests per day — is cheap precisely because the margin lives elsewhere. Check the pricing page before you build a budget on it; consumption tiers move.

Where the thesis could break

The skeptical reading is straightforward. Consumption revenue is lumpy, and a 20% headcount cut executed ahead of the revenue curve is a bet, not a proof. The 1,000x figure is a projection with no measurable basis yet — nobody has 5.5 trillion agents to count. And "agentic traffic" includes a great deal of polling, retrying and crawling that produces no economic value at all.

What is verifiable today: Cloudflare's growth target, its guidance, its headcount decision, and its pivot to consumption-based inference revenue. What is not verifiable: whether the web's next billion users will pay for anything. European operators should plan for the identification and labelling obligations now, and treat the traffic projections as a directional signal rather than a forecast.