Scored 0–3 across the eight moats that determine defensibility in the AI era. Framework: 8 Moats for the AI Era · Gokul Rajaram, 20VC.
Darktrace has two real moats — Workflow and Distribution — and both are operational rather than the AI story the company leads with. It clears the bar for genuine defensibility in its niche but sits below the three-moat threshold where defensibility starts to compound.
The notable finding: the dimension Darktrace markets hardest — "AI-native, Self-Learning AI" — maps in this framework to Data and Network Effects, the two moats where it scores lowest, because its per-customer learning architecture deliberately forgoes the cross-customer data asset that would make those moats real.
Darktrace's brand is inseparable from AI. "Self-learning AI" is the first phrase in most of its marketing. The implicit promise: more AI means more defensibility. The diagnostic finds the opposite. Darktrace's per-tenant unsupervised learning — the thing that makes the product genuinely novel — is the same architectural choice that prevents cross-customer data compounding and eliminates cross-customer network effects. Data scores 1. Network Effects scores 0.
The moats actually carrying the position are the ones the marketing underemphasizes: workflow embedding in SOC operations (2) and owned enterprise distribution (2). The AI is real. The AI moat is not.
Two moats score 2 or higher — Workflow and Distribution. Both are operational moats the marketing underemphasizes, while the AI story the company leads with maps to its two lowest-scoring dimensions.
The architecture is the tell. Darktrace's Self-Learning AI builds a model of "normal" inside each customer's own environment rather than training on a pooled data lake across the base — that's the explicit pitch on its own homepage. The result is that detection improves per-customer but doesn't compound across 10,000 logos the way CrowdStrike's Threat Graph or Microsoft Defender's cross-tenant telemetry does. There is real proprietary IP here — threat research, 200+ patent applications, detection models — but no compounding cross-customer data asset, which is what a Data moat actually requires. CrowdStrike scores a 2 on this dimension for exactly the asset Darktrace chose not to build; Vectra sits at a comparable 1.
One of the two real moats. Deployed inline across network, email, cloud and identity with autonomous response (the lineage from Antigena), Darktrace takes action on the wire and embeds into SOC operations in a way that's painful to unwind — re-baselining, re-integration, retraining analysts. The ceiling is that NDR is a layer, not the system of record: practitioners report tuning burden and false positives, and rip-and-replace happens (Vectra's customer marketing trades on exactly this complaint). That keeps it at 2, not 3 — real switching cost, not irremovable. CrowdStrike's endpoint agent earns a stronger workflow lock; ExtraHop and Vectra sit at a comparable 2.
Darktrace clears the enterprise-security compliance bar — ISO 27001, the newer ISO 42001 for AI management systems, Cyber Essentials, and FedRAMP High via Darktrace Federal. Those are real and non-trivial, FedRAMP especially for the public-sector segment. But they're table stakes in this category rather than a barrier over the competitive set: CrowdStrike, Palo Alto and Microsoft hold equivalent or deeper government authorizations. Certifications you share with every serious competitor narrow to a 1.
The second real moat, and an unglamorous one: roughly 10,000 customers across 110 countries, a globally recognized brand, and a network of OEM and channel partnerships — owned reach that newer NDR entrants structurally can't replicate. The enterprise sales motion and the early-mover brand position in AI-native cybersecurity make the cross-sell base the real asset: every new module (email, cloud, identity) sells into an installed base competitors have to win account by account. Real and compounding, but contested by players with larger distribution infrastructure — CrowdStrike's 28,000+ customer base and Microsoft's bundling reach cap this at 2.
Darktrace integrates widely — SIEMs, SOARs, ticketing systems, cloud providers — but it is the integrator, not the platform others build on. No developer economy, no marketplace, no third parties whose revenue depends on Darktrace as a substrate. The integration surface is real enough to earn a 1 (removal touches other tools in the stack), but an Ecosystem moat requires others to have built on top of you, and the per-tenant product architecture forecloses that in the same way it forecloses network effects. Compare to CrowdStrike's Falcon platform and marketplace, which earns a 2 for hosting third-party modules.
Darktrace scores 0 on Network Effects. The per-tenant learning architecture that defines the product also eliminates cross-customer network effects. One customer's network anomaly data provides zero benefit to any other customer. This is the starkest expression of the narrative trap: Darktrace's most prominent marketing asset (AI that learns your network) is structurally incompatible with the moat most associated with AI scale (cross-customer network effects).
Darktrace scores 0 on Physical / Infrastructure. Pure software. The moat is structurally unavailable to cybersecurity SaaS — not a Darktrace-specific weakness.
Darktrace scores 1 on Scale. ~$700M in ARR is a real number — category leadership by revenue in pure-play NDR. But Scale at 2 requires demonstrated cost or pricing leverage that smaller rivals structurally can't match. CrowdStrike at $3.4B ARR and Palo Alto at $8B+ operate at a scale that lowers unit economics in ways Darktrace cannot match. Category leadership without pricing power or structural cost advantage is leadership, not scale.
Not every moat is available to every business. The map below separates what Darktrace can deepen, what it could build, and what its architecture rules out.
Workflow — deeper SOC embedding as autonomous response expands across modules. Distribution — cross-sell into the 10,000-logo base and channel expansion under Thoma Bravo ownership.
Data — would require an architectural pivot toward opt-in cross-customer intelligence. Ecosystem — opening the platform for third parties to build on. Regulatory — deeper public-sector authorizations beyond category parity.
Network Effects — foreclosed by the per-tenant architecture as designed. Physical / Infrastructure — unavailable to cybersecurity SaaS.
Partial is the in-between tier — real moats, but not yet a stack. The Gokul threshold for durable defensibility is 3+ moats at score 2 or 3. The Defensibility Stress Test reveals which of your existing moats are worth deepening vs. which are eroding, and which additional moats are within reach. The Stacking Moats Playbook follows once the stack threshold is crossed.
Run all three steps yourself with the $179 Stacking Moats Playbook — the complete DIY process. Or jumpstart Step 1 right now: submit a few details and we'll audit your moats for you, free. Your scorecard arrives THE NEXT BUSINESS DAY.
The full three-step process this diagnostic is Step 1 of — score your moats, match your stack to one of the 6 durable archetypes, and identify the missing leg to build next. Six guides plus a companion resource:
We run the diagnostic and deliver your 8-moat scorecard by the next business day — tier, narrative trap, and structural availability map included. No call required.
While you wait: the Stacking Moats Playbook covers the complete self-service process.
Framework: 8 Moats for the AI Era according to Gokul Rajaram, 20VC Interview