Cloudflare in cybersecurity and the agentic internet
Cloudflare is best understood as an internet control-point business. A customer sends web, application, employee or network traffic through Cloudflare; the company then secures, accelerates, routes and increasingly executes code against that traffic. Cybersecurity is the commercial foundation, but the same network also supports connectivity, developer infrastructure and AI. This is why Cloudflare can plausibly grow beyond security without abandoning the architecture that created its moat.
The investment question is not whether Cloudflare can launch more products. It already has one of the broadest portfolios in infrastructure software. The question is whether one shared network produces superior economics and customer outcomes as the portfolio expands, or whether product breadth creates sales complexity, higher compute costs and direct competition with companies that dominate larger individual markets.
The business model: monetising the request path
A conventional software company sells a seat or a database. Cloudflare sells access to a distributed network and the policies applied when a request crosses it. The same request can be protected by DDoS mitigation and a web application firewall, screened for bots, accelerated through caching, routed across the backbone and handled by a Worker. The customer’s application may still live on Amazon Web Services, Microsoft Azure, Google Cloud or private infrastructure; Cloudflare earns its place by sitting in front of, between and increasingly inside those environments.
| Platform layer | Representative products | Buyer and value | Strategic role |
|---|---|---|---|
| Application services | DDoS protection, WAF, Bot Management, API Shield, CDN and load balancing | Security and application teams protect and accelerate public applications. | The original control point. Reverse-proxy traffic supplies threat intelligence and a low-friction path to attach more services. |
| Cloudflare One | Access, Gateway, DLP, CASB, Browser Isolation, Email Security and Cloudflare WAN | Security and networking teams replace VPN, web gateways and branch appliances. | Extends Cloudflare from protecting external applications to governing users, devices, offices and private resources. |
| Network services | Magic Transit, Network Firewall, interconnection and network analytics | Infrastructure teams connect data centres, branches, clouds and large networks. | Brings more enterprise traffic onto Cloudflare’s backbone and makes security an inline service rather than a separate appliance chain. |
| Developer and AI platform | Workers, R2, D1, Durable Objects, Containers, Workers AI, AI Gateway and Agents SDK | Developers build globally distributed applications and agents without managing servers. | Moves Cloudflare from handling a customer’s traffic to hosting parts of the application that generate it. |
This model produces several forms of revenue. Self-service plans charge a fixed monthly price; developer services can charge for usage; enterprise agreements commonly combine a recurring commitment with capacity or usage allowances. In cap-based contracts, rising traffic may not appear immediately as revenue: the customer pays more when it exceeds its allowance or resets the cap. Pool-of-funds contracts let large customers commit a budget and consume it across multiple products. They reduce procurement friction and encourage platform adoption, but revenue can ramp later as consumption develops.
The implication is important for AI. Exploding agent traffic is a demand indicator, not a revenue metric. It becomes economically meaningful when it drives earlier cap increases, faster committed-fund consumption or new security and developer contracts. Investors should follow monetised traffic and account expansion rather than treating raw request growth as equivalent to sales.
Why one network can compound
Cloudflare deploys a common software stack across its global edge. Services run close to users and can inspect traffic in one pass rather than sending it through separate regional appliances. A customer that already routes traffic through Cloudflare can turn on another policy or service without introducing a new network path. This is the architectural basis for consolidation: product attach should be faster, latency should be lower and incremental services should reuse existing network, telemetry and customer integration.
The second flywheel begins with developers. Workers removes infrastructure management, R2 removes data-egress charges and the surrounding primitives make it easier to build at the edge. A developer can start with a small application and later consume databases, storage, queues, observability, security and enterprise networking. The attraction is openness and ease of entry; the commercial payoff comes when application logic, state and traffic make Cloudflare operationally important.
This is not costless. Compute and AI workloads consume more third-party technology, servers and colocation capacity than traditional security rules. Gross margin has recently moved lower as paid traffic and developer products grew. That does not automatically mean weaker economics: a platform can accept lower product gross margin if it reduces customer acquisition cost, extends retention and creates large lifetime revenue. But management must prove those benefits rather than using “platform mix” as a permanent explanation for margin compression.
The cybersecurity franchise
Application security is Cloudflare’s strongest and most defensible cyber position. Because customer DNS and traffic already point to its edge, Cloudflare can stop volumetric attacks, malicious requests, automated abuse and API attacks before they reach the origin. Security and performance reinforce each other: the same network that absorbs an attack can cache content, route around congestion and reduce latency. The threat data generated across a large and diverse traffic base improves detection for the next customer.
Cloudflare One attacks a different budget. Access replaces legacy VPN for private applications; Gateway is a secure web gateway; DLP, CASB, Browser Isolation and Email Security protect data and users; Cloudflare WAN and Network Firewall connect and enforce policy across branches and data centres. The technical proposition is elegant: use the same nearby edge and policy engine for workforce access, internet egress and branch traffic, avoiding service chaining and backhaul.
The commercial challenge is harder than the technical one. Zscaler has stronger security-category focus, Palo Alto Networks offers a broader security platform, Microsoft controls identity and endpoints, and Fortinet is deeply embedded in branch networking. Cloudflare has no equivalent endpoint-detection franchise, authoritative identity directory or large security-operations platform. Its right to win is the network and application path, not ownership of every security control.
The company also does not disclose revenue by application security, Zero Trust, network services or developer products. Cloudflare reports as one operating segment. That supports the claim that products share one platform, but it limits external visibility into product mix, security growth and the profitability of newer businesses. Product adoption data should be treated as evidence of direction, not a substitute for segment economics.
AI: security layer, execution layer and possible market maker
The realistic AI thesis is broader than inference and narrower than replacing the hyperscalers. Frontier models require concentrated accelerator clusters and vast capital. Cloudflare’s stronger role is model-neutral infrastructure around them: route requests, apply identity and data policy, cache repeated answers, observe cost and latency, protect AI endpoints, run smaller models and execute the code generated by agents close to users.
| AI role | Cloudflare position | Economic potential | Our assessment |
|---|---|---|---|
| Secure employee AI use | Access, Gateway, DLP and browser controls sit between users and model providers. | Extends Cloudflare One as companies standardise policy for sanctioned and unsanctioned AI. | Near-term and credible. It maps directly to existing security budgets and traffic inspection. |
| Protect AI applications | WAF, API security, bot controls and AI-specific protections defend public model endpoints. | AI creates more valuable APIs and more automated abuse, expanding application-security demand. | A natural extension of the core franchise and probably the highest-quality AI revenue. |
| Control model consumption | AI Gateway routes providers, measures spend, caches responses and applies per-user or application controls. | Can become the neutral control point above multiple models and clouds. | Strategically attractive; differentiation must extend beyond observability into policy, routing and integrated security. |
| Build and run agents | Workers, Durable Objects, Containers, R2 and Agents SDK provide distributed execution and state. | Usage expands as agents create many short-lived applications and continuous tasks. | Large opportunity, but the toughest competition and greatest pressure on infrastructure economics. |
| Govern crawler access | AI Crawl Control, Web Bot Auth and pay-per-crawl identify agents and let content owners allow, block or charge. | Cloudflare could intermediate economic exchange between content owners and AI systems. | Genuine option value, not yet a base-case revenue stream. Both sides must adopt common identity, pricing and settlement. |
| Agentic commerce | Request-level verification and payment protocols could authenticate agents before a transaction. | Transforms machine requests into metered commercial events. | Conceptually powerful but early. Payment networks, merchants and agent platforms must cooperate. |
Pay-per-crawl is the most original part of the thesis. Search engines historically crawled content and returned referral traffic. Answer engines and agents can consume the content while keeping the user inside their own interface. Cloudflare is attempting to give publishers scarcity and pricing at the network layer. Its neutrality matters: it does not own a frontier model or compete with publishers, and it already sees both crawler demand and content supply.
Yet market making is much harder than bot blocking. Publishers need sufficient leverage, crawler operators must prefer paid access to circumvention or direct licensing, and payment and identity standards need broad acceptance. The sensible investment treatment is a free option attached to Cloudflare’s existing control point. Assigning large revenue before adoption would convert a differentiated experiment into speculative valuation support.
Current evidence and operating leverage
The mid-2026 evidence is encouraging but mixed. Quarterly revenue growth accelerated to 36%, dollar-based net retention reached 120% and the number of customers generating more than $100,000 of annualised revenue rose by roughly a quarter. Those indicators suggest improving enterprise execution and expansion rather than traffic growth alone. At the same time, gross margin fell to about 72% as network, technology and capacity costs increased. Cloudflare remains loss-making under reported accounting, with substantial stock-based compensation.
Management’s long-term model calls for at least 30% operating margin and 30-35% free-cash-flow margin. The path relies on sales productivity, R&D reuse and lower administrative cost as the platform scales. The company also announced a roughly 20% workforce reduction as part of an “agentic AI-first” operating model. This could demonstrate that AI changes the cost structure; it could also disrupt a company simultaneously expanding products and enterprise sales. Productivity should be measured through faster innovation, stronger customer outcomes and durable growth—not headcount reduction itself.
Competitive landscape
| Market | Principal competitors | Cloudflare advantage | Structural limitation |
|---|---|---|---|
| Application delivery and security | Akamai, F5, Fastly, Palo Alto Networks and hyperscaler-native services | Security and performance on one global reverse-proxy network, supported by broad threat visibility. | Mature CDN pricing is competitive; large customers can use multiple providers for resilience and bargaining power. |
| SASE and Zero Trust | Zscaler, Palo Alto Networks, Netskope, Cisco, Fortinet and Microsoft | One-pass networking and security close to users, with strong application and developer adjacency. | Enterprise sales maturity, endpoint context and security-suite breadth are stronger elsewhere. |
| Enterprise networking | Carriers, Cisco, Fortinet, Cato Networks, Palo Alto Networks and managed-network providers | Internet-native any-to-any connectivity without dependence on proprietary branch hardware. | Network replacement is operationally sensitive and channels favour incumbents with installed equipment and services. |
| Developer edge | Amazon Web Services, Microsoft Azure, Google Cloud, Vercel, Fastly and edge platforms | Instant global deployment, isolate efficiency, integrated security and low-egress storage. | Hyperscalers own deeper compute, database, data and enterprise ecosystems; Cloudflare’s primitives are less complete. |
| AI gateway and governance | Hyperscalers, API-management vendors, observability platforms and specialist AI gateways | Model neutrality plus inline identity, network and application-security policy. | Gateway features can commoditise unless they become deeply embedded in policy and application architecture. |
The investment debate
| Question | Bull case | Bear case | Evidence to watch |
|---|---|---|---|
| Does the platform compound? | Every product reuses the network, threat data and customer traffic path, lowering deployment friction and increasing retention. | Breadth creates a catalogue that is technically unified but commercially fragmented. | Higher product attach, larger committed contracts and customer expansion without disproportionate selling cost. |
| Is Cloudflare One a durable second act? | Security and networking converge on the same global policy fabric. | Focused security vendors and networking incumbents retain buyer trust and deeper control portfolios. | Large branch, workforce and WAN deployments that replace—not merely supplement—incumbent products. |
| Will developer services improve economics? | Workers and data primitives bring application logic onto Cloudflare and create usage-led expansion. | Compute is capital intensive, competitive and structurally lower margin than security software. | Contribution after network cost, improving gross margin by workload and conversion from developers to enterprises. |
| Is AI a new demand curve? | Agents multiply requests, APIs, security events and execution workloads, all of which touch Cloudflare’s strengths. | Traffic can be low-value, cached, shifted to hyperscalers or absorbed inside existing contract caps. | Cap upsizes, pool-of-funds consumption and disclosed growth in paid AI products. |
| Can margins scale? | One network and one code base create substantial operating leverage as revenue compounds. | Capacity, model costs, stock compensation and the expense of selling many products delay reported profitability. | Durable free cash flow, lower sales cost per new contract and progress toward reported—not only adjusted—profitability. |
What could break the thesis
- A major reliability or security failure. Concentration is the source of the platform’s value and its blast radius. A broad outage undermines the promise that Cloudflare is safer than fragmented infrastructure.
- Product sprawl outruns sales execution. The company must translate technical breadth into repeatable enterprise bundles without requiring every account team to sell dozens of specialist products.
- Developer growth lacks attractive contribution. Rapid Workers and AI usage is not sufficient if compute, third-party model and capacity costs absorb the value.
- Hyperscalers close the edge gap. Their control of compute, data, identity and procurement can outweigh Cloudflare’s neutrality and network design.
- Open-web economics deteriorate faster than new monetisation develops. AI may reduce human referrals and weaken media or software customers before crawler controls, gateways and agentic commerce become meaningful revenue.
- Governance and reputation become constraints. Operating a global internet control point exposes Cloudflare to censorship disputes, abusive customers, regulation and difficult decisions about who may use the network.
What matters from here
The best leading indicator is not the number of products or AI requests. It is whether customers commit more money to the platform and consume it across multiple acts. Large-customer growth, retention, product attach and earlier cap resets show commercial compounding. Gross margin, network cost and free cash flow show whether new workloads create economic value. Reported profitability and dilution remain necessary checks on adjusted results.
In cybersecurity, watch whether Cloudflare One wins full network transformations and whether application-security customers adopt workforce or developer controls. In AI, watch paid AI Gateway usage and Workers-based production workloads. Treat crawler payments and agentic commerce as milestones only when content owners and machine buyers transact at scale.
Bottom line
Cloudflare is one of the few cybersecurity companies with a credible route into general-purpose infrastructure. Its security products are not an adjacency to the network; they are the reason enterprises trust the network with traffic. Developer services then monetise the same architecture from the other direction by placing code and state where security and routing already occur.
