Cloudflare, Inc.

Global edge density plus one control plane for security, networking, and developer workloads. The more products a customer consolidates, the harder it is to unwind the platform.

Global edge density plus one policy-and-routing control plane. The moat is the compound switching surface created when a customer uses Access, Gateway, WARP, WAN, Workers, and security controls together; the more of the stack Cloudflare owns, the harder the unwind.

Cloudflare's one bottleneck is the default gate on machine-to-machine Internet traffic. Once a domain sits on Cloudflare's anycast fabric, three things become expensive to unwind together: (1) L3–L7 policy (WAF/DDoS/Zero Trust), (2) the developer/runtime plane (Workers) that now hosts agents, and (3) the crawler/identity/payment control plane that decides which bots may fetch, train, transact, or pay. The quantitative constraint is not "can a hyperscaler ship a competing WAF" — it is how much of the public web's bot policy is set by Cloudflare defaults, because a default that applies at onboarding is a network-level switching cost, not a seat-license switching cost.

1. FedRAMP High / GovRAMP Moderate (announced Aug 10, press dated Aug 11). Cloudflare for Government achieved FedRAMP High (marketplace product FR2000863987A) with NIST as sponsoring agency, plus GovRAMP Moderate, and stated intent to pursue DoD Impact Level 4. Architecture claim: same global software stack with software-defined regionality; High-Impact data processed inside an authorized U.S. boundary across 15 metro areas. Installed base named: >100 U.S. government agencies, including Commerce, Energy, HHS, DHS, Interior, Justice, and State. Named cloud-on-Cloudflare partners: Workday, New Relic, Armis Federal, Darktrace Federal, GitLab. Network footprint in the same release: >335 cities / >125 countries (was ">330 cities" in the FY2025 10-K cited by the 08-08 note).
2. AI-crawler default flip, effective September 15, 2026 (Cloudflare blog "Your site, your rules"). For all new domains onboarding to Cloudflare, Training and Agent crawlers are blocked by default on pages that display ads; Search remains allowed. Multi-purpose crawlers (Googlebot, Applebot, BingBot) are judged by all of their behaviors, so a Training-block also blocks mixed Search+Training bots unless the customer opts out before Sep 15. Controls are available to all plans including Free. This is the first time Cloudflare has set a network-wide default that treats AI-agent traffic as a gated class rather than an opt-in filter.
3. OpenAI Daybreak — Vulnerability Discovery and Remediation (Sep 3, 2026). Invitation-only early access inside Cloudflare Managed Defense. Uses OpenAI Daybreak models including GPT-5.6 Cyber via AI Gateway; inference runs on OpenAI servers, not at the edge. NVD logged 60,475 vulnerabilities by September 2026, already above the 48,185 recorded for all of 2025 (1.255× the prior full-year total with a quarter still open). Every patch and WAF rule requires human approval. Not a revenue disclosure.

A. Contracted-revenue lock (10-Q, period ended 2026-06-30).

Reading: ~30% of a year's revenue is already sitting as deferred, and 64% of $2.73B RPO converts inside 12 months. That is a weaker contractual lock than F5's ~65% deferred/revenue (FFIV note) because Cloudflare's mix is more usage/consumption (Workers pool-of-funds). The stickiness that is contractual is the RPO duration, not a 5-year COBOL core. The stickiness that is operational is the default-policy + edge-identity stack below.

B. Default-flip leverage (network share × new-domain flow).

Cloudflare's own 10-Q MD&A: network presence in more than 335 cities and over 125 countries; international was 49% of Q2 2026 revenue (51% in Q2 2025). The company has previously described itself as sitting in front of roughly a fifth of the web; that share figure is not restated in the Q2 10-Q and is treated here as a company-narrative prior, not a 10-Q fact. What is measurable: the Sep 15 default applies to every new domain that onboards, including Free. Switching cost for a new publisher is not "migrate a WAF"; it is "accept Cloudflare's Training/Agent block or opt out before the crawlers that also do Search get blocked." The economic capture of this default is option value, not booked revenue — Pay Per Crawl / Pay Per Use is the monetization path, still unquantified in SEC filings.

C. Federal clearance vs. booked federal dollars (cost-of-moat vs moat-defended).

Named federal Magic Transit + Network Firewall win cited on the Q2 call: five-year, $7.7M (~$1.54M/yr). Versus Q2 annualized revenue $2,784M that is 0.28% of one year of revenue — a proof-point, not a run-rate. FedRAMP High is the qualification barrier that lets Cloudflare compete for High-Impact civilian workloads the Moderate authorization (2022) could not touch; IL4 is still "intent," not achieved. Cost of the clearance is not disclosed as a line item. Conservative bound: the May 7 Plan's cash severance ($125M of the $165M ceiling) plus Q2 network capex (H1 PP&E purchases $115.2M − Q1 $65.2M = $50.0M in Q2; FY2026 network capex guided 14–15% of revenue) is the cash Cloudflare is spending to keep the fabric scarce. Q2 FCF $56.4M (8% of revenue) on $696.1M sales is the residual after that spend.

• product_service: Cloudflare connectivity cloud — security, performance, reliability, and developer services delivered globally.
• technical function: Anycast edge network/reverse proxy/CDN with DDoS mitigation, WAF, bot management, Zero Trust/SASE, edge compute/Workers, storage, AI inference. (10-K Business)
• why it is hard: Global anycast network scale, edge capacity, proprietary threat intelligence, software efficiency, developer ecosystem, and free-tier funnel. (10-K Business/Risk Factors)
• substitutes: Akamai, Fastly, AWS CloudFront/Shield, Google Cloud CDN, Azure Front Door, Zscaler, Palo Alto, F5, Imperva. (10-K competition/Risk Factors)
• pricing power: MED — high gross margin/retention, but competitive market and negative GAAP operating margin. (10-K MD&A/Risk Factors)

Cloudflare’s pricing power is mixed:

• Strong in security/availability products where switching is operationally risky and the service is embedded in production traffic.
• Weaker in basic CDN/delivery where price per bandwidth is competitive and hyperscalers can bundle.
• Usage-based Workers/R2 can benefit from AI/edge compute demand, but unit economics depend on underlying GPU/compute cost and inference pricing.

If Cloudflare can raise prices without material logo churn, the evidence would show up in dollar-based net retention, gross margin, and large-customer count. Those signals require primary-source validation.

# Who they sell to

Cloudflare sells to:

• Individual developers and small self-serve customers — acquired through the free tier and paid self-serve plans.
• Small and mid-market businesses — often through online sales and automation.
• Large enterprises — direct sales, customer success, and channel partners.
• Governments and public-sector entities — increasingly relevant for security, DDoS protection, and Zero Trust; exposure may vary by country.
• Partners and service providers — including managed security service providers and resellers.

• What share of revenue comes from Zero Trust/security versus developer services?
• What is the real migration effort for a large Cloudflare One deployment with endpoint, policy, and routing dependencies?
• Is the Q1 2026 gross-margin dip mostly AI-infrastructure mix, or an enduring structural shift?

• What share of Q2 2026 growth was driven by Zero Trust/security versus developer services and the new agentic-internet workload mix?
• Is the 35% current-RPO growth durable, or is it mostly an AI/agentic-commerce demand burst that will normalize?
• How much of the modeled $1M-$8M+ switching cost is real economic friction versus a public proxy for policy and traffic-state rebuild effort?

STRENGTHENED

STRENGTHENED

STRENGTHENED. The 0% convertible is a hard capital-markets confirmation of the control-plane thesis, not a distress raise — 0% coupon, 60% premium, and a fully exercised greenshoe are what a scarce AI-infrastructure asset commands. Open question: whether the capex it funds converts to FCF before the notes mature.

INTACT (not STRENGTHENED). The three catalysts thicken the control-plane side of the thesis (who may crawl, who may run High-Impact gov workloads, who may auto-patch at the edge) without yet thickening economic capture. Booked economics are still a 36%-growth, 72%-GM, 8%-FCF connectivity cloud with a 6/9 STRONG PROXY toll-road. Option value in Wallets / Pay-Per-Use / VDR / IL4 is real and dated, but unbooked. Research-only; no live capital.

INTACT (not STRENGTHENED). The thesis's load-bearing assumption — that machine traffic becomes the dominant flow Cloudflare gates — is now empirically true (>50%, ahead of schedule). But that validation thickens the control-plane side without thickening economic capture: no new booked revenue, no 8-K, and the flagship metric is volume, not dollars. Cloudflare remains a 36%-growth, ~72%-GM, 8%-FCF connectivity cloud with a 6/9 STRONG PROXY toll-road, whose durable upside (agentic monetization) is real, dated, and unbooked. Research-only; no live capital.