Cloudflare CEO Says the Internet Is Changing Exponentially

Cloudflare disclosed that AI-agent (agentic) internet traffic surpassed human activity on its network in May and could reach 1,000x human traffic within five years, supporting the AI-infrastructure demand narrative despite debate over the magnitude. Separately, Intel plans to sell equity to raise $15B to fund capex/working capital, citing progress in areas like physical AI and advanced packaging; panelists broadly endorsed the timing given Intel’s strong valuation and need to avoid taking on additional debt. The mining segment highlighted a $100M workforce/education commitment under the Trump administration (and broader mineral-security push), but speakers were skeptical that funding alone will fix cyclicality, remote labor constraints, and execution risk in mining start-ups.

Analysis

The real economic winner from agentic traffic is not the application layer; it is the tollbooth layer around identity, security, routing, and metering. Names like NET can monetize every incremental machine-to-machine request even if humans disappear from the page, while the ad stack is exposed to a structural pricing reset: if agents aggregate decisions, impression-based CPMs and SEO arbitrage lose scarcity value. That creates a second-order headwind for TTD and any publisher-adjacent model, while search/commerce platforms like GOOGL may defend share only if they become the default agent rail rather than a destination that gets bypassed.

Near term, the cleanest beneficiaries are picks-and-shovels tied to compute density and power delivery: NVDA on accelerator demand, plus GEV and PWR on grid, cooling, and datacenter buildouts. The market may already discount the first derivative of AI capex, so the better trade is relative value versus exposed monetizers rather than outright index beta. On the financing side, INTC’s equity raise reads as a signal that the foundry/packaging thesis is capital-hungry but still alive; that helps suppliers if execution holds, but it also tells you returns are being pulled forward by balance-sheet capacity, not yet by cash flow.

Contrarian view: the consensus is taking the 1,000x traffic narrative too literally. Agent usage can scale fast, but ROI gates, power constraints, and model costs will force a lot of the activity into closed systems where the monetization pool is smaller than the traffic headline suggests. That argues for caution on ad-tech shorts if the market overreacts, and for treating the infrastructure trade as a staggered 6-18 month theme rather than a one-week momentum chase.

Key Decisions for Investors

• Short TTD vs long NET for 1-3 months: thesis is agentic traffic is more valuable to routing/security infrastructure than to impression-based ad monetization; cover if TTD re-rates on evidence of agent-targeting product traction.
• Buy GEV or PWR on a 10-15% pullback, or via 3-6 month call spreads, to express datacenter power/cooling demand; thesis breaks if AI capex growth decelerates or utility interconnect delays widen materially.
• Treat INTC as a financing-execution trade rather than a fundamental rerate; only add on weakness if the raise is absorbed without further dilution and foundry margin trajectory improves over the next 1-2 quarters.
• Underweight TTD and other ad-tech proxies until there is proof they can price or target AI-agent traffic; falsifier is a sustained improvement in revenue per agentic request or better-than-expected ad load conversion.
• Maintain a tactical long GOOGL only if it remains the default interface for both humans and agents; otherwise use it as a hedge against ad-stack disruption risk over the next 6-12 months.