PART 01
The five moats that built the Fortune 500
Warren Buffett made "economic moat" the default language for competitive advantage: a structural barrier that protects profits and market share over decades, not quarters. For most of the 20th century, five moat types were enough.
Network effects
Every new user makes the product more valuable for everyone else. Visa, Windows, and WhatsApp scaled this way.
Intangible assets
Brands, patents, regulatory licences. Coca-Cola commands premium pricing on a commodity. Pfizer holds drug exclusivity through patent portfolios.
Cost advantages
Lower unit economics through scale, logistics, or proprietary processes. Walmart and Amazon turned supply chain mastery into pricing weapons.
Switching costs
Financial, procedural, or psychological friction that locks customers in. Salesforce embeds so deeply in workflows that migration costs years.
The fifth, efficient scale, applies where markets are too small for multiple large players. Utilities and rail networks sit here. Together, these moats powered Procter & Gamble, GE, and IBM through decades of competition.
PART 02
Why AI is draining the moat
Foundation models like GPT-4, Claude, and Llama accelerate one thing above all: commoditisation. Coding copilots democratise development. Open-source models via APIs let anyone ship a working product in a weekend. Rapid iteration erodes IP barriers faster than legal teams can file patents.
The traditional five are weakening in specific, measurable ways.
- 1
Technology and IP erode as open-source LLMs replicate proprietary algorithms within months of release.
- 2
Scale advantages compress when cloud GPUs let a three-person startup train models that previously required a data centre.
- 3
Brand loyalty thins when AI-generated alternatives flood every category, unless the brand carries deep emotional meaning beyond the product.
- 4
Switching costs drop as AI-driven feature parity means the next tool works just as well on day one.
The key shift
When any competent team can build a comparable product in weeks, the defensible advantage moves from what you build to what you control around it: the data that feeds it, the distribution that reaches users, the workflow depth that makes removal painful.
PART 03
Four moats that matter now
These are not entirely new concepts. What's new is their position. They've moved from tactical advantages to the primary source of defensibility. Each one exploits something AI cannot easily replicate: access, context, speed, and entrenchment.
1. Distribution ownership
Models commoditise. Reaching users does not. The organisations that own the customer pipe - through direct relationships, embedded partnerships, or viral acquisition mechanics - control the one asset that cannot be open-sourced. OpenAI's integration into Microsoft Office. NVIDIA's CUDA lock-in. ServiceNow's workflow dominance. In each case, the moat is not the model. It is the channel.
Ask yourself: if a competitor cloned your product tomorrow, would your customers still be unreachable to them?
2. Proprietary data loops
Static datasets commoditise quickly. What holds value is dynamic, high-dimensional, closed-loop data: behavioural signals, real-time interactions, outcome feedback that continuously improves your model. Tesla accumulates driving data with every mile. Meta's social graph deepens with every interaction. The data is not just an asset. It is a compounding engine.
The critical distinction: data you collect once is a snapshot. Data that feeds back into your system and makes it measurably better is a moat.
3. Momentum and iteration speed
When AI landscapes shift weekly, the organisations that ship first and iterate fastest capture mindshare before competitors finish their planning cycles. This is not "move fast and break things." It is disciplined speed - building in public, launching demonstrations that force the market to react to you rather than the other way around. Perplexity seized search momentum this way. ElevenLabs used hackathons and viral demos to define a category before incumbents could respond.
4. Workflow depth
A standalone AI tool is easy to swap. An AI system embedded into daily operational workflows - touching decisions, approvals, evidence trails, and reporting - becomes load-bearing infrastructure. Removing it means rebuilding processes, not just switching vendors. This is the "system of intelligence" layer: not a feature, but a substrate the organisation runs on.
For regulated industries, this matters double
In financial services, healthcare, and legal, workflow depth intersects with compliance obligations. An AI system woven into your regulatory evidence chain, your accountability mapping, your supervisory submissions, does not get swapped out on a quarterly vendor review. It becomes part of your audit trail. That is a moat most SaaS vendors never reach.
PART 04
The moat transition at a glance
The table below maps how each traditional moat is being pressured by AI and what is emerging to replace it.
| Traditional moat | AI vulnerability | Emerging replacement |
|---|---|---|
| Patents and IP | Open-source commoditisation | Proprietary data loops |
| Economies of scale | Cloud access levels the playing field | Distribution embedding |
| Brand loyalty | AI-generated alternatives in every category | Momentum and community |
| Switching costs | Feature parity across tools | Workflow depth |
PART 05
What this means for your organisation
The winners in the AI era will not be the teams with the best model. They will be the ones who control what feeds the model, who uses it, and how deeply it is embedded in the work that matters.
Three questions worth asking today:
- 1
What data do we generate that no competitor can access? If the answer is "nothing unique," your AI advantage has an expiry date.
- 2
How deep is our product in customer workflows? If it sits on top of processes rather than inside them, it is replaceable.
- 3
Do we own the distribution, or rent it? If your access to customers depends entirely on a platform or channel you do not control, your moat belongs to someone else.
The hard truth: most organisations are still optimising traditional moats while the ground shifts beneath them. The ones that adapt - by building proprietary loops, embedding into operational workflows, and owning their distribution - will define the next decade. The rest will compete on price.