Skip to content

andrew-ng-three-moats.md

Latest commit

 

History

History
38 lines (27 loc) · 3.09 KB

andrew-ng-three-moats.md

File metadata and controls

38 lines (27 loc) · 3.09 KB
Raw
title Andrew Ng — Three AI Moat Types
source-id landing-ai-playbook
wiki-page (02_References/enterprise-ai/wiki/skills/strategy/build-defensible-ai-moats.md)
last-synced 2026-06-07

Andrew Ng — Three AI Moat Types

What this reference contains

Andrew Ng's foundational framework (Landing AI Playbook, Step 4): AI strategy should produce defensible competitive advantage — a moat — not just adoption. Three moat types named: asset portfolio, industry-specific advantage, and virtuous data cycle / network effects.

Key claims (with citations)

  • Adoption alone is not a moat. 88% of firms now use AI (Stanford AI Index 2026). Adoption is table-stakes. The strategic question is: what makes your company's AI different from anyone else's? Source: Landing AI Playbook.
  • Three moat types:
    1. Asset portfolio moat — several difficult AI assets aligned with a coherent strategy; a competitor must replicate all simultaneously. Source: Landing AI Playbook, pp.3–4.
    2. Industry-specific advantage — become the AI leader in your industry, not in general. Don't compete against hyperscalers. Source: Landing AI Playbook, pp.3–4.
    3. Virtuous cycle / network effects — product produces data when used; data improves the product; the loop compounds. Source: Landing AI Playbook, pp.3–4.
  • Data strategy is specific, not general. "It is NOT true that having many terabytes of data automatically means an AI team will be able to create value from that data." Bring AI engineers into data-acquisition decisions early. Source: Landing AI Playbook, p.4.
  • Unified data warehouses are prerequisite. 50 siloed databases controlled by 50 different VPs means AI engineers cannot "connect the dots." Unification precedes moat-building. Source: Landing AI Playbook, p.4.
  • Don't write the strategy first. "Most companies will not be able to develop a thoughtful AI strategy until it has had some basic experience with AI." Pilots produce the signal; strategy names the moat. Source: Landing AI Playbook, p.3.

Extended moat types (post-2019)

  • Orchestration moat (Stanford 2026): for GenAI specifically, the moat has shifted from raw data to the orchestration / model abstraction layer. The model commoditizes; the orchestration compounds.
  • Strategic-position portfolio (WEF/BCG 2026): elaborates industry-specific advantage into a formal framework with named positions and stopping rules.

When to consult

  • For skill general-idea-diagnostic: Q3 (value accumulates?) maps directly to moat analysis — does this initiative build toward any of the three moat types?
  • For skill tech-buy-vs-build: the "buy the model, build the orchestration" principle follows from the Ng/Stanford synthesis — the orchestration layer is the moat, not the model.
  • For skill general-maturity-assessment: a firm at Stage 3/4 (MIT CISR) should be able to name its moat type; if it can't, it's still in productivity-tool territory.

Source

Landing AI Enterprise AI Playbook (Andrew Ng). Step 4: Build strategic assets and moats. Core moat taxonomy, data-strategy prerequisites, virtuous cycle framing.