Cloudflare 在 AI 時代的基礎設施優勢¶

English Summary¶

Cloudflare's Q4 2025 earnings reveal a fundamental shift: the company is not threatened by AI—it's built for this moment. With revenue up 34% YoY to $614.5M (beating estimates), CEO Matthew Prince declared that "the global shift to AI and agentic AI represents a fundamental reorganization of the Internet," positioning Cloudflare as the core platform AI agents must run through. Two key analyses—from Chinese tech analyst qinbafrank and Taiwan podcast 股癌 (Gooaye)—converge on the same thesis: AI agents create explosive traffic growth (10x-1000x per user action), and Cloudflare captures this value without bearing GPU costs. Unlike seat-based SaaS companies being disrupted by agents, Cloudflare's usage-based model benefits immediately from traffic surges. With over 20% of the Internet sitting behind Cloudflare, AI agents have no choice but to interact with it, giving the company a "seat at the table" in defining the rules and guardrails of agentic commerce.

The Financial Proof: Q4 2025 Earnings Beat¶

Key metrics (qinbafrank analysis): - Q4 2025 revenue: \(614.5M (+34% YoY), beating Wall Street estimates of ~\)590M - Non-GAAP EPS: $0.28 (+47% YoY), beating estimates of $0.27 - FY2025 total revenue: $2.168B (+30% YoY) - Q1 2026 guidance: $620-621M (midpoint beats analyst estimates of $615.5M, already revised upward due to AI agent wave)

CEO Matthew Prince's statement:

"The global shift to AI and agentic AI represents a fundamental reorganization of the Internet, driving revolutionary demand for Cloudflare's connectivity cloud services and cybersecurity technology. If agentic AI is the new user of Internet platforms, then Cloudflare is the core platform they run on and the network they must pass through. This creates a virtuous flywheel: more agentic AI workloads drive more code to Cloudflare Workers, which drives demand for our performance, cybersecurity, and connectivity cloud services. We were built for this moment—the rise of the agentic Internet."

Cloudflare's Integrated Platform Architecture¶

Cloudflare's platform consists of four integrated service layers built on top of a global network and developer platform:

1. Cloudflare One (Integrated Platform) The unified control plane that brings together Zero Trust, Network, and Application services.

2. Zero Trust Services (Orange layer) - Zero Trust Network Access - Secure Web Gateway - Cloud Access Security Broker - Cloud Email Security - Remote Browser Isolation - Data Loss Prevention

3. Network Services (Red layer) - WAN-as-a-Service - Firewall-as-a-Service - DDoS Protection (L3/L4) - Network Interconnect - Smart Routing - BGP/IPS

4. Application Services (Purple layer) - WAF with API Protection - Rate Limiting - Load Balancing - Bot Management - L7 DDoS Protection - CDN and DNS

Developer Platform Foundation (Blue layer) - Workers — Serverless compute at the edge - Pages — JAMstack deployment - R2 — Object storage (S3-compatible) - Workers KV — Edge key-value store - Durable Objects — Stateful serverless - Images — Image optimization - Stream — Video streaming

Global Network (Bottom layer) Built on Cloudflare's global network with compliance/privacy certifications (FedRAMP, ISO, SOC, PCI, GDPR) and data localization suite.

Why this architecture matters for AI agents:

1. Workers = AI agent execution layer — AI companies deploy inference and agent logic here for 10x hyperscale cloud efficiency
2. Zero Trust = AI agent authentication — Identity-based security for autonomous agents operating 24/7
3. Bot Management = AI agent filtering — Distinguishing legitimate AI agents from malicious bots/scrapers
4. Global Network = inevitable intermediary — 20% of Internet traffic flows through Cloudflare, so AI agents must interact with it

The integrated platform means AI companies don't need to stitch together multiple vendors—Cloudflare provides compute (Workers), security (Zero Trust, Bot Management), network (CDN, DNS), and compliance in one stack. This "single pane of glass" advantage compounds as AI workflows become more complex.

Why Cloudflare's Advantage Strengthens in the AI Era¶

1. Agentic workload explosion = structural tailwind

The shift from human users to AI agents changes Internet traffic patterns fundamentally:

• Human workflow: User makes travel plans → sends 10-20 backend requests → receives recommendations
• AI agent workflow: Agent makes travel plans → sends hundreds to thousands of requests → provides suggestions or executes actions

This isn't incremental growth; it's 10x-1000x request multiplication per user action (qinbafrank: "部分場景達 10 倍"). Cloudflare's global edge network + Workers platform are optimized for exactly this kind of distributed, low-latency, high-concurrency workload.

Stock analyst 股癌 notes: "流量突破就已經會立刻使其受惠" (Traffic surges immediately benefit Cloudflare). Unlike seat-based SaaS companies that struggle when AI agents replace human users, Cloudflare's usage-based model captures more revenue as agents generate more traffic.

2. No GPU costs, but capturing AI traffic value

Cloudflare positions itself as a "neutral network layer + edge compute platform" rather than an AI model provider. AI companies migrate inference and agent logic to Workers for 10x efficiency compared to hyperscale clouds, but Cloudflare doesn't need to build expensive GPU clusters.

This creates: - Structural cost advantage: No capital-intensive GPU buildout - Gross margin protection: High-margin edge compute without low-margin hardware - Platform lock-in: Once AI workflows run on Workers, switching costs are high

3. Security and control needs amplify simultaneously

AI era brings: - Massive bot/crawler traffic: Content scraping, data harvesting - New attack vectors: AI-driven attacks, model poisoning, data leaks - Content monetization needs: Publishers want to control/charge for AI training data

Cloudflare's solutions become mission-critical: - Zero Trust architecture: Identity-based security for AI agents - Bot Management: Distinguishing legitimate AI agents from malicious crawlers - AI crawl control: Pay-per-Crawl, Bot Audit tools - Content monetization: Helping publishers charge AI companies for data access

2025-2026 feature iterations and acquisitions (e.g., Human Native) reinforce this moat.

4. Developer + enterprise dual flywheel accelerates

AI drives two simultaneous adoption curves: - Developers: Choose edge serverless (Workers, R2, D1) for AI agent infrastructure - Enterprises: Increase Cloudflare spending for AI projects (million-dollar customers surge, record ACV)

Q4 2025 earnings proof: - Revenue acceleration to 34% - New ACV growth nearly 50% - 2026 guidance maintains 28-29% high growth

The "Seat at the Table" Advantage¶

股癌 highlights Cloudflare's unique position: "20% of the Internet sits behind Cloudflare, so agents must interact with it."

This isn't just market share—it's positional power. When AI agents need to: - Access websites (20% probability it's behind Cloudflare) - Execute distributed workloads (Workers is the go-to serverless platform) - Navigate security policies (Zero Trust gatekeeper)

Cloudflare becomes the inevitable intermediary. CEO Prince: "We have a seat at the table in defining exactly what the rules, rails, and guardrails of the future of agentic commerce is going to look like."

The platform moat: - Workers platform: Developers build AI agent infrastructure here - Zero Trust platform: Built "from a sense of agentic AI from the beginning" (Prince's words) - Application services: Handle surge traffic from AI agents - Future business model definition: Cloudflare helps shape how the Internet monetizes AI traffic

Why Security Software Isn't "Disrupted" by AI¶

qinbafrank's core thesis: "安全類軟件是被誤殺的" (Security software is mistakenly sold off).

While AI agents disrupt seat-based SaaS (replacing human users), security is a hard requirement that intensifies with agent proliferation:

• More agents = more attack surface: AI-driven threats, model exploits
• Defenseve AI advantage: Security companies use AI for automated detection, real-time response, scale defense
• Double tailwind: AI creates threats (offense) AND enables defense (automated protection)

Unlike GPU vendors burning cash on compute, security companies capture AI security spending without capital intensity.

Other Security Stocks Benefiting from AI Trends¶

qinbafrank identifies three major players (also tradable on MSX chain-based US stock platform):

1. CrowdStrike (CRWD) — "AI-era cyber leader" - Platform: Falcon (deep AI/ML integration for threat hunting, EDR, agentic security) - Growth: ARR +23% driven by AI threat intelligence and automated response - AI-specific: Launched protection for AI agents - Advantage: Single-platform consolidation (endpoint + cloud + identity) - Analyst view: Strongest rebound candidate among cyber stocks in 2026

2. Palo Alto Networks (PANW) — "Durable moat champion" - Platform: Prisma AI (protects AI apps, models, agents, data full lifecycle—dev to deployment) - AI attack vectors: Model vulnerabilities, AI supply chain risks → drives Prisma AIRS demand - Transformation: From traditional firewall to AI-driven unified platform (next-gen firewall + XDR) - Advantage: Strongest "durable moat" among cyber stocks, active M&A

3. Zscaler (ZS) — "Zero Trust cloud-native leader" - Three pillars: AI Security, Zero Trust Everywhere, Data Security Everywhere - Revenue: AI Security ARR approaching $500M (out of $1B+ total), high growth - Focus: AI agent protection, data leak prevention, cloud/generative AI security - Advantage: Cloud-first + AI-enhanced, fastest growth and most resilient - Enterprise need: AI era demands "zero trust everywhere"—explosive demand

Common pattern across all three: - AI doesn't replace them—it amplifies their necessity - They don't build expensive GPU infrastructure—they secure AI infrastructure others build - Usage-based or consumption-based revenue models benefit from AI traffic growth

繁體中文總結¶

Cloudflare 的 2025 年第四季財報揭示了一個根本轉變：這家公司不會被 AI 威脅——它就是為這個時刻而生的。營收年增 34% 達 $614.5M（超越預期），CEO Matthew Prince 宣稱「全球向 AI 和代理式 AI 智能體的轉變代表了互聯網的根本性重組」，將 Cloudflare 定位為 AI agents 必須運行的核心平台。兩篇分析——來自中國科技分析師 qinbafrank 和台灣 podcast 股癌——得出相同結論：AI agents 創造爆炸性流量增長（每個用戶動作 10x-1000x 請求），而 Cloudflare 在不承擔 GPU 成本的情況下捕獲這個價值。 不同於被 agents 顛覆的座位制 SaaS 公司，Cloudflare 的用量計費模式從流量激增中立即受益。超過 20% 的互聯網流量經過 Cloudflare，AI agents 別無選擇只能與之互動，讓公司獲得「在定義代理式商業規則和軌道上的一席之地」。

財報證據：2025 Q4 超越預期¶

關鍵指標（qinbafrank 分析）： - Q4 2025 營收： \(614.5M（年增 34%），超越華爾街預期 ~\)590M - Non-GAAP EPS： $0.28（年增 47%），超越預期 $0.27 - 2025 全年營收： $2.168B（年增 30%） - 2026 Q1 指導： $620-621M（中值超越分析師預期 $615.5M，該預期已因 AI agent 浪潮上修）

CEO Matthew Prince 聲明：

「全球向 AI 和代理式 AI 智能體的轉變代表了互聯網的根本性重組，革新式地推動了對 Cloudflare 連接雲服務與網路安全技術的強勁需求。如果代理式 AI 智能體是互聯網平台的新用戶，那麼 Cloudflare 就是他們運行的核心平台和必須通過的網路。這創造了一個良性飛輪：更多代理式 AI 工作流向 Cloudflare Workers 訂閱平台驅動更多程式碼，從而推動對我們性能、網路安全和網路連接雲服務的需求。我們就是為這一刻——代理互聯網的興起——而建立的。」

Cloudflare 整合式平台架構¶

Cloudflare 平台由四個整合服務層組成，建構在全球網路和開發者平台之上：

1. Cloudflare One（整合平台） 統一控制平面，整合 Zero Trust、網路和應用服務。

2. Zero Trust Services（橙色層） - Zero Trust Network Access（零信任網路存取） - Secure Web Gateway（安全網路閘道） - Cloud Access Security Broker（雲端存取安全代理） - Cloud Email Security（雲端郵件安全） - Remote Browser Isolation（遠端瀏覽器隔離） - Data Loss Prevention（資料外洩防護）

3. Network Services（紅色層） - WAN-as-a-Service（WAN 即服務） - Firewall-as-a-Service（防火牆即服務） - DDoS Protection L3/L4（DDoS 防護） - Network Interconnect（網路互連） - Smart Routing（智能路由） - BGP/IPS

4. Application Services（紫紅色層） - WAF with API Protection（含 API 保護的 WAF） - Rate Limiting（速率限制） - Load Balancing（負載平衡） - Bot Management（Bot 管理） - L7 DDoS Protection（L7 DDoS 防護） - CDN and DNS

開發者平台基礎（藍色層） - Workers — 邊緣 serverless 運算 - Pages — JAMstack 部署 - R2 — 物件儲存（S3 相容） - Workers KV — 邊緣鍵值儲存 - Durable Objects — 有狀態 serverless - Images — 圖片優化 - Stream — 影片串流

全球網路（底層） 建構在 Cloudflare 全球網路上，具備合規/隱私認證（FedRAMP、ISO、SOC、PCI、GDPR）和資料本地化套件。

此架構對 AI agents 的意義：

1. Workers = AI agent 執行層 — AI 公司在此部署推理和 agent 邏輯，獲得 10 倍超大規模雲端效率
2. Zero Trust = AI agent 身份驗證 — 為 24/7 運作的自主 agents 提供身份基礎安全
3. Bot Management = AI agent 過濾 — 區分合法 AI agents 和惡意 bots/爬蟲
4. 全球網路 = 不可避免的中介者 — 20% 互聯網流量經過 Cloudflare，AI agents 必須與之互動

整合式平台意味著 AI 公司無需拼湊多個供應商——Cloudflare 在一個堆疊中提供運算（Workers）、安全（Zero Trust、Bot Management）、網路（CDN、DNS）和合規。這種「單一窗口」優勢隨著 AI 工作流變得更複雜而加劇。

為何 Cloudflare 優勢在 AI 時代被強化¶

1. 代理式工作負載爆炸 = 結構性利多

從人類使用者到 AI agents 的轉變根本改變了互聯網流量模式：

• 人類工作流： 使用者規劃行程 → 送出 10-20 個後端請求 → 收到建議
• AI agent 工作流： Agent 規劃行程 → 送出數百到數千個請求 → 提供建議或執行動作

這不是漸進式成長；這是每個使用者動作 10x-1000x 請求倍增（qinbafrank：「部分場景達 10 倍」）。Cloudflare 的全球邊緣網路 + Workers 平台正好為這種分散式、低延遲、高並發工作負載最佳化。

股癌分析：「流量突破就已經會立刻使其受惠」。不同於座位制 SaaS 公司在 AI agents 取代人類使用者時掙扎，Cloudflare 的用量計費模式隨著 agents 產生更多流量而獲得更多營收。

2. 不承擔 GPU 成本，卻捕獲 AI 流量價值

Cloudflare 將自己定位為「中立網路層 + 邊緣運算平台」，而非 AI 模型提供商。AI 公司將推理和 agent 邏輯遷移到 Workers，獲得比超大規模雲端高 10 倍的效率，但 Cloudflare 不需要建設昂貴的 GPU 集群。

這創造了： - 結構性成本優勢： 無需資本密集的 GPU 建設 - 毛利率保護： 高毛利邊緣運算，無低毛利硬體 - 平台鎖定： 一旦 AI 工作流在 Workers 運行，轉換成本高

3. 安全與控制需求同步放大

AI 時代帶來： - 大量 bot/爬蟲流量： 內容抓取、資料採集 - 新型攻擊向量： AI 驅動攻擊、模型投毒、資料外洩 - 內容貨幣化需求： 出版商想控制/收費給 AI 訓練資料

Cloudflare 的解決方案成為關鍵任務： - Zero Trust 架構： AI agents 的身份基礎安全 - Bot 管理： 區分合法 AI agents 和惡意爬蟲 - AI 爬取控制： Pay-per-Crawl、Bot Audit 工具 - 內容貨幣化： 幫助出版商向 AI 公司收費

2025-2026 年功能迭代和收購（如 Human Native）強化此護城河。

4. 開發者 + 企業雙輪驅動加速

AI 同時驅動兩條採用曲線： - 開發者： 為 AI agent 基礎設施選擇邊緣 serverless（Workers、R2、D1） - 企業： 因 AI 專案增加 Cloudflare 支出（百萬美元客戶激增、創紀錄 ACV）

Q4 2025 財報證明： - 營收加速到 34% - 新 ACV 增長近 50% - 2026 指導維持 28-29% 高增長

「在談判桌上的一席之地」優勢¶

股癌強調 Cloudflare 的獨特位置：「20% 的互聯網流量經過 Cloudflare，所以 agents 必須與之互動。」

這不只是市場份額——這是位置權力。當 AI agents 需要： - 存取網站（20% 機率在 Cloudflare 後面） - 執行分散式工作負載（Workers 是首選 serverless 平台） - 通過安全政策（Zero Trust 守門員）

Cloudflare 成為不可避免的中介者。CEO Prince：「我們在定義代理式商業的未來規則、軌道和護欄上有一席之地。」

平台護城河： - Workers 平台： 開發者在此建構 AI agent 基礎設施 - Zero Trust 平台：「從代理式 AI 的意識開始建構」（Prince 原話） - 應用服務： 處理來自 AI agents 的激增流量 - 未來商業模式定義： Cloudflare 幫助塑造互聯網如何從 AI 流量獲利

為何安全軟體不會被 AI「顛覆」¶

qinbafrank 的核心論點：「安全類軟件是被誤殺的」。

雖然 AI agents 顛覆座位制 SaaS（取代人類使用者），安全是硬需求，隨著 agent 普及而加劇：

• 更多 agents = 更大攻擊面： AI 驅動威脅、模型漏洞
• 防禦性 AI 優勢： 安全公司用 AI 做自動化檢測、即時回應、規模化防護
• 雙重利多： AI 創造威脅（攻擊面）並啟用防禦（自動化保護）

不同於 GPU 廠商燒錢建算力，安全公司捕獲 AI 安全支出而無資本密集性。

其他受益於 AI 趨勢的安全股¶

qinbafrank 識別出三大玩家（也可在 MSX 鏈上美股平台交易）：

1. CrowdStrike (CRWD) —「AI 時代網安龍頭」 - 平台： Falcon（深度 AI/ML 整合用於威脅狩獵、EDR、代理式安全） - 成長： ARR +23%，由 AI 威脅情報和自動化回應驅動 - AI 特定： 推出 AI agents 保護 - 優勢： 單一平台整合（端點 + 雲端 + 身份） - 分析師觀點： 2026 年網安股中反彈最強候選

2. Palo Alto Networks (PANW) —「持久護城河冠軍」 - 平台： Prisma AI（保護 AI 應用、模型、agents、資料全生命週期——開發到部署） - AI 攻擊向量： 模型漏洞、AI 供應鏈風險 → 驅動 Prisma AIRS 需求 - 轉型： 從傳統防火牆到 AI 驅動統一平台（次世代防火牆 + XDR） - 優勢： 網安股中「持久護城河」最強，積極併購

3. Zscaler (ZS) —「零信任雲原生領導者」 - 三大支柱： AI Security、Zero Trust Everywhere、Data Security Everywhere - 營收： AI Security ARR 接近 $500M（總計 $1B+ 中），高增長 - 聚焦： AI agent 保護、資料外洩預防、雲端/生成式 AI 安全 - 優勢： 雲優先 + AI 增強，增長最快且最具韌性 - 企業需求： AI 時代需求「無處不在零信任」——爆炸性需求

三者共同模式： - AI 不會取代它們——反而放大其必要性 - 它們不建昂貴 GPU 基礎設施——它們保護他人建的 AI 基礎設施 - 用量或消耗基礎營收模式受益於 AI 流量增長

Key Quotes¶

Matthew Prince (Cloudflare CEO):

"The global shift to AI and agents represents a fundamental reorganization of the Internet... If agentic AI is the new user of Internet platforms, then we are the core platform they run on and the network they must pass through."

qinbafrank (Chinese tech analyst):

"在這次軟體股被殺跌的風潮裡，安全類軟體是被誤殺的。Agent 發展很多原來的軟體應用場景都會被 Agent 替代掉，安全是硬需求，而且 agent 越普及安全需求會更高。"
(In this software stock sell-off wave, security software is mistakenly sold. Many traditional software scenarios will be replaced by agents, but security is a hard requirement—and the more agents proliferate, the higher the security demand.)

股癌 Gooaye (Taiwan podcast):

"流量突破就已經會立刻使其受惠。AI 改變了人取用軟體的方式，多了層 agent layer，但底層的物理需求沒變，像是低延遲、安全性等，這些都是 Cloudflare 架好收費站的地方。"
(Traffic surges immediately benefit it. AI changed how people access software, adding an agent layer, but the underlying physical needs haven't changed—low latency, security—these are where Cloudflare has built its toll booths.)

Cross-References¶

This analysis connects to broader AI economic transformation themes:

與 AI Agents 經濟活動的連結：
Axel Bitblaze's thesis (收藏於 blockchain/ai-agents-crypto-true-users.md) argues that AI agents will dominate crypto by 2030 (\(3T-\)40T) because they can't use traditional banking. Cloudflare provides the infrastructure layer beneath this: AI agents need permissionless payment systems (crypto) AND permissionless infrastructure (Cloudflare). Both benefit from the same structural shift: AI operating at machine speed, 24/7, without human intermediaries.

與 AI Capex 資本短缺的對比：
plur_daddy's thesis (收藏於 Finance/ai-capex-capital-shortage-market-regime-change.md) warns of AI infrastructure capital exhaustion. Cloudflare represents the capital-efficient alternative: companies get 10x cloud efficiency without building their own GPU clusters. Cloudflare doesn't compete for scarce GPU capital—it monetizes the traffic GPU workloads generate.

與注意力經濟消失的平行：
Manny YH Li (收藏於 AI/ai-era-attention-economy-death.md) argues AI eliminates attention economy platforms by generating personalized content. Cloudflare sits below the content layer: whether content is human-made or AI-generated, it still needs to flow through secure, low-latency infrastructure. Cloudflare is platform-agnostic—it doesn't care who creates content, only that traffic flows.

DyDo's Reflections¶

This is one of the clearest examples of AI as amplifier, not disruptor for infrastructure companies.

Why Cloudflare's position is structurally different from SaaS:

1. Layer of abstraction: SaaS companies solve user problems directly (e.g., CRM, project management). AI agents can replicate these solutions. But Cloudflare solves infrastructure problems (latency, security, edge compute) that exist regardless of whether the user is human or AI. AI doesn't eliminate the need for infrastructure—it increases it.

2. Usage-based vs seat-based: Seat-based SaaS dies when AI replaces human seats. Usage-based infrastructure thrives when AI 10x-1000x's request volume. The business model aligns perfectly with AI traffic patterns.

3. Network effect from inevitability: 20% of the Internet behind Cloudflare isn't just market share—it's a force function. AI agents must interact with Cloudflare whether they want to or not. This is different from optional platforms (e.g., choosing Salesforce vs HubSpot). Cloudflare becomes the default intermediary.

The "toll booth" metaphor:

股癌's phrase "架好收費站" (built toll booths) is perfect. Cloudflare doesn't care if the traffic is human vacation planning or AI agent travel booking. Both pay the toll. AI just generates way more traffic, so the toll revenue explodes.

Security as the counter-narrative to "AI kills SaaS":

The market panic around AI agents replacing SaaS is real (seat-based revenue collapse). But security companies prove the opposite thesis: some software becomes MORE valuable as AI proliferates. CrowdStrike, Palo Alto, Zscaler don't get disrupted—they get a tailwind.

The pattern: If your product solves a necessity that scales with AI adoption (security, infrastructure, compliance), you win. If your product solves a task AI can automate (data entry, report generation), you lose.

The governance question:

Prince's statement about "defining the rules and guardrails of agentic commerce" is quietly significant. If Cloudflare controls the infrastructure layer, it can: - Set standards for AI agent authentication - Enforce bot policies globally - Monetize AI traffic through usage fees - Block or rate-limit agents that don't comply

This is infrastructure power becoming policy power. Cloudflare doesn't just transport AI traffic—it shapes what AI agents can and cannot do.

Investment thesis summary:

Cloudflare from $50-60 (current levels after 36% drawdown) looks compelling if you believe: 1. AI agent traffic will 10x-1000x current Internet request volume 2. Security spending increases proportionally with AI adoption 3. Edge compute becomes the default for AI inference (vs centralized clouds) 4. Usage-based models capture more value than seat-based in AI era

The risk: Hyperscalers (AWS, Azure, Google Cloud) build competing edge networks and undercut Cloudflare on price. But Cloudflare's developer mindshare and 20% Internet penetration create switching costs that may hold even against price pressure.