NET vs ZS Head-to-Head: In the Age of AI Agents, Should You Back the Highway Builder or the Tollbooth Operator?

Introduction: One Highway, Two Toll-Collection Logics

Investors in 2026 face a question that looks simple but is actually decisive: as the age of Agentic AI moves into full swing, should you buy "the company that builds the highway," or "the company that sets up the tollbooth on that highway"?

Cloudflare is the highway builder. It has deployed nodes in 330 cities worldwide, giving AI agents somewhere to run, somewhere to compute, and somewhere to store data — what it sells is "the right to use the road." Zscaler is the tollbooth operator on that highway. It lets no traffic pass freely — every car, every passenger, every piece of cargo must be authenticated at its "sky checkpoint" — what it sells is "safe passage."

On the surface, both companies sit in "cloud security" and compete in the SASE (Secure Access Service Edge) race, and their share prices have been highly correlated over the past 12 months. But once you strip away the marketing language and dig into their technical architecture, revenue models, and AI strategy, you find a counterintuitive fact: NET and ZS have completely different DNA — they are betting on two entirely different structural theses for the AI era.

NET's bet is: AI agents need a native runtime environment, and Cloudflare's edge network is that environment.

ZS's bet is: the more numerous and autonomous AI agents become, the more rigid the need for governance and security becomes — and Zscaler is the only platform that can authenticate trillion-scale traffic at millisecond latency.

This report will not tell you "which one to buy" — it will use data and logic to let you arrive at your own answer.

I. The Underlying Logic: Two Companies With Completely Different DNA

1.1 Cloudflare: From CDN to "the Operating System of the Agentic Era"

Cloudflare was founded in 2009, originating from an open-source anti-spam project called "Project Honey Pot." Seventeen years later, CEO Matthew Prince redefined the company's evolutionary arc using an "acts of a play" framework: Act 1 was reverse proxy (WAF, DDoS protection), Act 2 was forward proxy (Zero Trust, SASE), Act 3 was the developer platform (Workers serverless computing), and Act 4 — happening right now — is infrastructure for the agentic internet.

In his opening essay for Agents Week in April 2026, Prince offered a precise analogy: Workers' V8 isolate architecture is like "giving every agent a knife, a stove, and just enough counter space," while a traditional container is like "giving every agent an entire commercial kitchen, even if all it wants to do is make a cup of coffee." That explains why Cloudflare's newly launched Dynamic Workers can spin up in milliseconds, execute AI-generated code, and vanish — 100 times faster than containers, at a fraction of the cost.

Cloudflare positions itself as a "Connectivity Cloud" — a neutral, global network fabric that frees customers from being locked into the closed ecosystems of AWS, Azure, or GCP. Its core architectural advantage is that every server can run every function — from security to AI inference. That lets it deliver sub-100ms latency to 95% of the world's connected population.

1.2 Zscaler: From VPN Replacement to "the Identity-Verification Hub of the AI Era"

Zscaler was founded by Jay Chaudhry in 2007. Chaudhry was born in a village without electricity at the foot of the Himalayas in India, made his way to the US on a scholarship, and had already sold four security companies in succession before founding Zscaler. In 2007 he saw something the industry wouldn't admit: the cloud was moving enterprise applications outside the castle walls. Spending money to reinforce the walls (firewalls, VPNs) amounted to guarding an empty city.

His answer was the "Zero Trust Exchange" — a cloud-based "checkpoint in the sky." All traffic — whether from human users, IoT devices, or now AI agents — must be authenticated before it is allowed to communicate. No trust, only verification.

At the Q2 FY2026 earnings call in February 2026, Chaudhry formally declared Zscaler "the security platform for the AI era" and appointed Dr. Swamy Kocherlakota as Executive Vice President of "Agentic AI Security Engineering" — a role designed specifically around securing AI agents. ZS also became an early partner for Microsoft Entra Agent ID, extending Zero Trust principles to "non-human identities."

II. Financial Head-to-Head: Nine Key Metrics

Lay the two companies' latest earnings side by side and the numbers don't lie. Below is a direct comparison as of the most recently completed quarter (NET: Q4 CY2025 / ZS: Q2 FY2026, both through the end of January 2026).

III. The Age of AI Agents: Whose Positioning Is More Irreplaceable?

This is the most important chapter in this report. Agentic AI is not the future — it is already happening. As Anthropic announced Managed Agents in April 2026, OpenAI rolled out an enterprise version of Codex, and Microsoft rolled out its Copilot Agent ecosystem in full, a new question has surfaced: what infrastructure do these autonomously operating AI agents actually need?

3.1 Cloudflare: "I Am the Agent's Home"

Matthew Prince's declaration at Agents Week (April 13, 2026) was blunt: Cloudflare wants to become "the underlying operating system of the agentic internet."

Concretely, Agent Cloud provides an entire value chain from development to deployment:

Dynamic Workers — an ultra-lightweight runtime built on V8 isolates, where AI-generated code spins up, executes, and disappears within milliseconds. 100 times faster than traditional containers, supporting millions of concurrent executions.

Artifacts — a Git-compatible storage primitive designed for the agentic era. It supports the creation of tens of millions of repositories, giving agents a "permanent home" to store code and data.

Sandboxes (GA) — a full, isolated Linux environment where agents can clone repos, install Python packages, and run builds, just like human developers.

Think Framework — a persistent Agents SDK that lets agents support long-running tasks across sessions, rather than only responding to a single prompt.

Unified Model Catalog — after acquiring Replicate, developers can switch between OpenAI GPT-5.4, Anthropic Claude, and open-source models within a single interface by changing just one line of code.

OpenAI's head of Codex product, Rohan Varma, publicly endorsed Cloudflare — and that is no coincidental partnership. When the world's most capable AI lab chooses to run its agents on Cloudflare, it's worth thinking seriously about what that implies.

3.2 Zscaler: "I Decide What Agents Can and Can't Do"

Chaudhry's logic runs in the opposite direction: the more numerous and autonomous agents become, the more they need to be governed. At the Q2 earnings call he said it directly: "Our Zero Trust Exchange architecture is fundamentally different from a firewall — a firewall lets a user or an AI agent roam freely once it's on the network, which dramatically raises the risk of a breach."

ZS's AI security strategy runs along three lines:

AI Protect — real-time monitoring of enterprise AI application traffic, including prompt monitoring, data-loss prevention, and Shadow AI asset inventory. ThreatLabz's 2026 report found that the number of enterprise AI/ML applications surged to over 3,400 within a year, with sensitive data transmission volume up 93%. If you don't know how many people at your company are secretly using ChatGPT, or how much data is flowing to external models, you've already lost.

Agentic AI Security — a governance framework designed specifically for non-human identities. ZS integrates with Microsoft Entra Agent ID, applying Zero Trust verification to every AI agent — as strictly as it does for humans. Kocherlakota's new role leads this product line.

ZDX CoPilot — uses AI agents to automatically diagnose network and application performance problems. Over the past 12 months, ZDX Advanced Plus bookings surpassed $100M, growing over 80%.

The scale at which ZS processes AI traffic is a number that cannot easily be replicated: in all of 2025, it processed nearly one trillion AI transactions. That isn't just data volume — it's a threat-intelligence flywheel. Every intercepted attack, every new infiltration technique, trains ZS's models further, making the next interception more precise.

IV. A Deep Cross-Comparison of Moats

A common mistake investors make is treating "moat" as a binary — you either have one or you don't. The genuinely useful analysis is: where is the moat deepest, and where is it most fragile?

4.1 Network Effects and Scale

NET: 330+ cities, 13,000 interconnects, a 4.5-million-developer ecosystem. The free tier is its most powerful acquisition mechanism — the natural upgrade path from free to paid to enterprise lowers customer-acquisition cost. DBNRR has snapped back to 120%, meaning existing customers are ramping spend faster. Million-dollar customers surged 55% for the full year, with 69 net adds — a single-quarter record.

ZS: 150+ data centers, processing more than 500 billion transactions per day, and nearly one trillion AI transactions across all of 2025. More than 45% of the Fortune 500 and roughly 40% of the Global 2000 are customers. 728 million-dollar ARR customers. More than 450 enterprises have gone Zero Trust Everywhere, three quarters ahead of schedule.

Verdict: NET's network is wider (330 vs. 150), ZS's security traffic runs deeper (500 billion transactions per day). The two moats are deep along different dimensions — NET is deep in developer stickiness, ZS is deep in threat-intelligence data volume.

4.2 Switching Costs

NET: Once an enterprise has deployed serverless applications on Workers, stored data in R2, and managed state with Durable Objects, moving away means rewriting the entire backend. As Agent Cloud deepens, switching costs will stack further. But NET's free tier also means the "trial" bar is extremely low — customers find it easy to come in, and, in theory, easy to leave (if they haven't gone deep).

ZS: The Zero Trust Exchange replaces an enterprise's entire network security architecture — VPN, firewall, SWG, all swapped out. Once deployed, ZS becomes the mandatory path for every packet of traffic. Moving away isn't just a technical cost — it's a compliance risk, and you'd have to explain to the CISO and the board why you're tearing down the security architecture and starting over.

Verdict: ZS has the higher switching cost. Replacing security architecture is more painful than migrating a developer platform — it involves compliance, audits, and organizational politics.

4.3 Data Flywheels

NET: observes more than 20% of global internet traffic every day and is one of the largest sources of DDoS attack intelligence. But NET's data flywheel is mainly used for "performance optimization" (routing, caching, latency minimization) — its depth on the security side doesn't match ZS's.

ZS: nearly one trillion AI transactions and 500 billion routine transactions a day, all of it security intelligence. Every interception trains the model further — attacker techniques, targets, timing patterns. This flywheel has been spinning for nearly 20 years, and no new entrant can replicate it.

Verdict: ZS's security-data flywheel is unmatched in depth. NET's traffic-data flywheel is stronger in breadth, but its security-intelligence density doesn't match ZS.

4.4 Brand and Mindshare

NET: mindshare in SASE is around 6.0-6.4%, but its mindshare within the developer community is extremely high. 100% of PeerSpot users say they'd recommend it. NET's brand is extremely strong on perceptions of being "fast, innovative, and developer-friendly."

ZS: SASE mindshare is around 9.1-9.7% — meaningfully higher than NET. In the minds of CISOs and enterprise security decision-makers, ZS is synonymous with Zero Trust. 98% of users say they'd recommend it. A regular in the Leaders quadrant of the Gartner SSE Magic Quadrant.

Verdict: ZS has stronger brand mindshare in enterprise security procurement. NET has a stronger brand among developers and technical decision-makers. Different audiences, different moats.

4.5 Overall Moat Scorecard

V. Crossfire: Are They Really Fighting Over the Same Market?

5.1 The SASE Overlap: Real, But Overstated

On the surface, NET and ZS have clear product overlap in SASE: Cloudflare One (ZT Access + Gateway + Browser Isolation) goes head-to-head with Zscaler's ZIA + ZPA. In October 2025, Cloudflare even published a 37-feature comparison document claiming it had "surpassed Zscaler" in overall feature completeness within four years.

But the data reveals the truth: Cloudflare's security revenue is about 30-35% of the total (Act 2), while Zscaler's is 100%. NET's SASE business is more of a natural upsell from its CDN customer base rather than a direct raid on ZS's large enterprise accounts. In terms of deep penetration into the Fortune 500, ZS (45%+) far exceeds NET.

Cloudflare's SASE advantage lies in simpler deployment, lower cost, and a unified management console. Zscaler's advantage lies in the depth of its security features, the completeness of its compliance certifications, and its trust foundation with large enterprises. The two attract different buyers — NET attracts "enterprises already using Cloudflare's CDN who add security on top," while ZS attracts "CISOs for whom security is the primary requirement."

5.2 The Real Threat Isn't Each Other

For NET, the biggest threat is the edge services of the hyperscalers — AWS, Azure, GCP (Lambda@Edge, Cloudfront Functions, Azure Edge Zones). If the hyperscalers commit seriously to edge AI inference and cut prices aggressively, NET's differentiation could get squeezed.

For ZS, the biggest threat is Palo Alto Networks' (PANW) "platformization dumping" strategy — giving away SASE tools for free in exchange for customers abandoning other vendors — plus the possibility of CrowdStrike extending into network security.

The competition between NET and ZS looks more like "accidental overlap" than "head-on collision." If you're a CISO, you wouldn't skip buying Zscaler just because Cloudflare has Zero Trust — any more than you'd skip buying Cloudflare because AWS has a WAF.

5.2.1 The Competitive Triangle: NET × ZS × PANW × CRWD

You can't just look at these two companies head-to-head. Bringing Palo Alto Networks (PANW) and CrowdStrike (CRWD) into the picture is necessary to see the full competitive dynamic:

A few key observations:

First, NET has almost no rival in AI inference / edge computing. PANW, CRWD, and ZS none of them touch this space. NET's Agent Cloud is an exclusive position — among security companies, no one else offers a one-stop platform combining "a runtime environment for agents + inference services + storage + security." This is the core logic behind NET's valuation premium.

Second, PANW is ZS's biggest near-term threat. Nikesh Arora's "platformization dumping" — giving away SASE tools for free in exchange for full customer lock-in — is a highly effective sales strategy. Chaudhry has repeatedly emphasized on earnings calls that "we win the vast majority of head-to-head Zero Trust competitions," but PANW's strategy isn't to out-feature ZS — it's to fight a war of attrition on price. ZS's 22% non-GAAP operating margin gives it some room to cut prices, but if PANW's free-tools strategy persists for more than three years, ZS's new-customer acquisition rate could be systematically suppressed.

Third, CRWD is more complementary to ZS than competitive. CrowdStrike does endpoint security (Falcon), ZS does network security (Zero Trust Exchange). Large enterprises typically procure both, forming a dual layer of "endpoint + network" defense. Chaudhry has publicly called CRWD a partner on multiple occasions. But over the long run, if CRWD extends into network security (its Charlotte AI is already moving in that direction), the overlap could grow.

5.2.2 Comparing Management Quality: Prince vs. Chaudhry

Management quality is often the decisive variable in long-term investment returns. Below is a comparison of the two CEOs' track records:

5.3 The Lesson From the Mythos Episode: What Did the Panic Actually Reveal?

On March 27, 2026, Anthropic was reported to be testing a frontier model called Claude Mythos, capable of autonomously discovering network vulnerabilities. On the news, security stocks fell an average of 5%. ZS, already down 38%, took a further hit. On April 7, Anthropic officially released Mythos Preview along with the Project Glasswing safety framework, triggering a broader SaaS sell-off.

The market's panic logic was: if an AI model can find vulnerabilities on its own, do you still need Zscaler?

That logic makes a fundamental error — equating "vulnerability discovery" with "security protection." Mythos is an X-ray machine — it can show you what's wrong inside your body. Zscaler is the emergency room — it saves you in real time when you're under attack. A stronger X-ray machine doesn't make the ER redundant — if anything, it means more people learn they need treatment, and demand for the ER only rises.

Broken down by product line:

Impact on ZS: neutral to slightly positive. Mythos finding more vulnerabilities means enterprises become more aware of the scale of their attack surface, accelerating Zero Trust deployment budgets. ZS's AI Protect — Shadow AI inventory, AI asset management, prompt monitoring — is precisely the answer to the risks Mythos exposes. More importantly, ZS's own ThreatLabz 2026 report had already revealed that enterprise AI/ML application count surged to over 3,400 within a year, sensitive data transmission rose 93%, and AI-driven attacks can go from discovery to lateral movement to data exfiltration in as little as 16 minutes. Mythos merely validated, in a more dramatic way, the story ZS was already telling.

Impact on NET: mildly negative, then quickly reversing to positive. NET's share price also got dragged down, but Cloudflare resolved it perfectly at Agents Week on April 13 — announcing Managed OAuth (letting AI agents securely browse internal applications), MCP security governance, Cloudflare Mesh (private network access for agents), and Shadow MCP detection rules. NET's message was clear: agents need tighter security? I'll provide the secure runtime environment for that — that's infrastructure's job.

The most important lesson from the Mythos episode isn't that some company will get wiped out — it's that: the TAM for AI security is being dramatically expanded. As AI models themselves become double-edged swords — both attack tool and defense tool — enterprise security spending can only go up. This is a structural tailwind for both NET and ZS — they simply benefit through different channels.

5.4 Comparing TAM: Whose Ceiling Is Higher?

TAM (Total Addressable Market) is the anchor for growth-stock valuation. The two companies' TAM narratives are entirely different, and understanding the difference is key to understanding the valuation gap.

NET's TAM is "wider" — it can point to nearly all of cloud infrastructure spend. But that also means NET must compete against different rivals in every sub-market (Akamai in CDN, ZS in security, AWS Lambda@Edge in compute). A big TAM doesn't mean you can actually capture it.

ZS's TAM is more "focused" — concentrated in the single vertical of security, but going deeper globally than anyone else. Penetration of just 3.5% means it could grow several-fold simply by capturing more of the existing market — before even counting the new market for non-human identity governance that AI agents create. Chaudhry noted on the Q2 earnings call that Zero Trust Everywhere customers have ARR 2-3 times that of average customers — that's "density expansion" within the existing TAM.

5.5 Five-Year Scenario Valuation Model

Laying out all the variables, here's a projection of fair market cap five years out under three scenarios:

5.6 Comparing Revenue Quality: Whose Dollars Are "Easier to Earn"?

Earning a dollar isn't the same quality everywhere. Below is a breakdown of revenue quality along four dimensions:

Predictability: ZS's revenue comes almost 100% from multi-year subscription contracts; ARR and RPO offer extremely high visibility (FY2026 ARR guidance of $3.73-3.745B). NET is also predominantly subscription-based, but usage-based models like Workers AI and R2 are growing — this adds growth elasticity but also adds forecasting difficulty.

Gross Margin Structure: ZS's 80% non-GAAP gross margin reflects the economics of pure software — it doesn't need to buy GPUs or build data centers (it rides on public clouds). NET's 74.9% gross margin reflects a "hardware-intensive" business model — servers, GPUs, and interconnect equipment across 330 cities worldwide are all NET's own capital expenditure. As Agent Cloud scales, GPU capex could compress margins further.

Customer Concentration Risk: NET's large customers account for 73% of revenue (Q4 2025), and million-dollar customers surged 55% — good news (enterprises are spending more), but it also means losing a handful of large accounts could have an outsized impact. ZS's customer base is more distributed (3,886 accounts at $100K+, 728 at $1M+); while it likewise leans on large customers proportionally, the absolute count is larger.

Cash Conversion Efficiency: ZS's combined H1 FY2026 Rule of 62 (26% growth + 36% FCF margin) is top-tier across SaaS companies market-wide. NET's FCF margin doubled from 10% in Q4 2024 to 16% in Q4 2025 — a good trend, but the absolute level still trails ZS. NET's capex (13% of revenue) is an ongoing cash drain that ZS barely has to deal with.

VI. Valuation and Risk: Whose Margin of Safety Is Larger?

6.1 Valuation Comparison

6.2 Risk Matrix

VII. Dual Evaluation Through the Four-Layer Defensive Screen

Below, we apply ProfitVision LAB's Four-Layer Defensive Screen to evaluate whether either stock is suitable as a Bull Put Spread candidate.

VIII. Conclusion: It's Not Either/Or — It's Understanding What You're Buying

By this point, the conclusion writes itself.

If you buy NET, you're buying:

A "toll-road" platform company transitioning from a CDN/security company into the operating-system infrastructure of the AI-agent era. Its revenue acceleration (34% growth, DBNRR 120%, RPO +48%) is one of the strongest forward-looking signals in the entire market. Agent Cloud has just received a public endorsement from OpenAI. TAM is expanding from $181B (2025) to $231B (2028).

But you're also buying one of the highest valuations anywhere in the market — P/S of 25x, P/E of 165x. You're paying "the day-after-tomorrow's price" for "tomorrow's story." AI inference revenue is still negligible right now — the CEO admits it himself. Gross margin is under pressure from GPU capex (down from 77.6% to 74.9%).

Suitable for: investors willing to pay a premium for high-quality growth, with a longer holding horizon, who can stomach a 30-40% drawdown.

If you buy ZS, you're buying:

Security infrastructure that is "essential" to the AI era. The more numerous and autonomous AI agents become, the more rigid ZS's demand becomes — that's logical deduction, not marketing copy. A data flywheel built on one trillion AI transactions, 20 years of accumulated threat intelligence, and Zero Trust penetration above 45% of the Fortune 500 — these are walls built over time.

More importantly, you're buying a stock the market has seriously mispriced. A P/S of 5.4x, Rule of 62, non-GAAP operating margin of 22%, and $3.5B in cash — this doesn't look like the valuation of "a company about to be replaced by AI." It looks like a quality asset being sold off in a panic.

But you're also buying some real risks: organic ARR growth slowing to 21% (excluding Red Canary), PANW's platformization offensive, SBC at 26.5% of revenue, and succession risk with Chaudhry at 72.

Suitable for: disciplined investors who prioritize margin of safety, are willing to wait for technical confirmation, and prefer "value found in panic."

Options Strategy Framework: Discipline Comes Before Conviction

Even the best story requires waiting for the right entry timing. Below is a concrete options framework for each stock:

NET Options Strategy (Pending Confirmation)

NET currently cannot execute a Bull Put Spread, because the Four-Layer Screen has not yet cleared on MarketSurge data. But if the following signals appear after the Q1 FY2026 earnings report on 4/30, deployment can be initiated:

Entry Conditions: A/D ≥ C+, RS ≥ 80, share price holding above the 50MA, DBNRR sustained at 120%+. If all four conditions clear, execute a Bull Put Spread: Delta <30, DTE 30-45 days, Short Put set below the nearest support level (estimated $160-165 zone, to be confirmed against post-earnings technical structure). Risk unit: 5% of account (~$600).

Alternative Strategy: If the share price breaks above $200 after earnings but positioning hasn't fully confirmed, consider a Poor Man's Covered Call (PMCC) — buy a long-dated, deep-in-the-money call (Delta 80+, DTE 180+) and sell a short-dated out-of-the-money call (Delta 30, DTE 30-45) — harvesting the time-value spread to capture IV premium while retaining upside participation.

ZS Options Strategy (Watch Cautiously — downgraded from Watch Actively)

ZS's positioning metrics are worse than previously estimated: A/D has deteriorated to E (the lowest grade), RS is only 6, 3M RS = 3, U/D Vol Ratio 0.70. This isn't a "bottoming soon" signal — it's a "institutions are still systematically selling" signal. But the fundamental metrics (EPS 98, EPS Growth 69%, SMR A, Forward P/E 29) diverge sharply from the positioning data — that kind of divergence usually means either the market is right (there's bad news you don't know about) or the market is wrong (excessive panic).

Trigger Conditions (all must be met before executing):

① A/D Rating recovers to B or above (indicating institutional selling pressure has stopped)
② RS Rating recovers to 70 or above (indicating relative strength has been restored)
③ Share price reclaims the 50MA (technical confirmation of a bottom)
④ IV Rank >30% (ensuring premium is rich enough)

Execution Parameters: Bull Put Spread, Short Put Delta <30, DTE 30-45 days, Short Put strike set in the $100-105 zone (below the long-term support band), risk unit 5% of account (~$600).

Early-Positioning Strategy: if you have strong conviction in ZS's long-term thesis but don't want to wait for positioning confirmation, consider scaling into Cash-Secured Puts — sell puts at a $95-100 strike, DTE 60 days, rolling monthly. If assigned, you build a long-term position at a very low cost basis; if not assigned, you keep harvesting time value. The precondition is that you're genuinely willing to hold ZS at $95-100.

Final Framework

If you must choose only one, ask yourself: what do you think is the actual bottleneck of the AI-agent era?

If you believe the bottleneck is "nowhere to run" — buy NET.

If you believe the bottleneck is "running unsupervised" — buy ZS.

If you believe the bottleneck is "both" — you already know the answer.

— Shiba the Disciplined · ProfitVision LAB · 2026/04/15