Every technology wave redraws the competitive map, and the AI wave is redrawing it faster than any before it. Some economic moats are getting deeper — companies with unique data and distribution are turning AI into a force multiplier. Others are quietly draining, as capabilities that took decades to build become available to anyone with an API key. Telling the two apart may be the most important analytical skill of this market cycle.

The core question: what does AI commoditize here?

AI's economic signature is that it collapses the cost of things that used to be expensive to produce: software code, written content, analysis, design, customer support, expertise itself. Any moat that ultimately rests on "producing that expensive thing" is exposed. Any moat that rests on something AI can't produce — proprietary data, physical infrastructure, regulatory permission, an installed network — is comparatively safe, and may even strengthen as AI makes the surrounding layer cheaper.

That gives you the first-pass sort for any company: is their advantage the thing AI is commoditizing, or the thing AI needs?

How each moat pillar fares

Network effects — mostly resilient, with one big caveat. AI cannot conjure a network of users; the cold-start problem survives the technology shift. The caveat is disintermediation: if AI agents become the interface through which users touch platforms — booking the trip, choosing the product, summarizing the feed — the platform's direct user relationship weakens, and with it the network's pricing power.

Switching costs — the most exposed pillar. Much enterprise lock-in rests on migration being painful: rebuilding integrations, retraining staff, porting data. AI is getting good at exactly those tasks. When code migration and data transformation get 10x cheaper, switching costs shrink across the whole software industry at once. Incumbents whose stickiness rests on genuine workflow value will re-lock customers; those whose stickiness rested on migration pain face a repricing.

Intangible assets — split decision. Patents and licenses hold (AI doesn't dissolve legal rights — and AI-driven industries are accumulating their own regulatory barriers). Brands split: trust-based brands may gain value in a world flooded with AI-generated content and lookalike products, while brands that mainly solved discovery ("the name you think of first") lose ground to recommendation engines that don't care about mindshare.

Cost advantages — cuts both ways. Scale players who can afford frontier-model compute and proprietary-data pipelines widen their gap. But AI also hands small competitors automation that erases labor-cost advantages — the offshore call center, the big back office — and levels parts of the playing field.

Efficient scale — largely untouched. AI does not build a second pipeline or a competing airport. Physical and regulatory natural monopolies are the quiet safe harbor of this transition.

Scoring it: opportunity and threat are separate questions

A frequent mistake is treating "AI exposure" as one number. It's two:

  • AI opportunity — can this company use AI to widen its moat? The bar should be high: genuine opportunity means structurally defensible leverage, like proprietary data no rival can assemble, or distribution that makes the company the default AI interface for its customers. Merely adopting AI to keep up — the copilots and chatbots every competitor also ships — is defense, not advantage.
  • AI threat — is AI actively eroding the barriers that protect this company? The tell-tale signs: the core product being commoditized by models, AI-native competitors entering with structurally lower costs, or customers using AI to do in-house what they used to pay for.

The uncomfortable truth is that both can be high at once: a company can be integrating AI impressively and watching its industry's barriers fall. Integration doesn't cancel erosion.

This is why MoatScan scores every company on both dimensions independently — an AI Opportunity score and an AI Threat score, each 1–10 with evidence bullets — rather than a single verdict. The gap between them (the net AI impact) shows up on every analysis page and in the database.

The honest caveat

AI moat analysis in 2026 involves genuine uncertainty — model capabilities, agent adoption, and regulation are all moving targets, and anyone claiming precision is selling something. The framework above tells you where to look and which claims to distrust (especially "we have lots of data" — most data is neither unique nor useful — and "we're integrating AI" — so is everyone). The company-by-company verdicts require doing the work: check the AI Impact section of any MoatScan analysis to see both scores argued with evidence, starting with an obvious case study like NVIDIA or Microsoft.