Castles in the Cloud: Can AI Companies Build Moats That Last?

Dr Vassilia Orfanou, PhD, Post Doc, COO, LUDCI.eu
Writes for the Headline Diplomat emagazine, LUDCI.eu

The AI Gold Rush: Growth vs. Survival in the Era of ChatGPT

In late 2022, a quiet revolution burst into the open. Within just two months of its launch, ChatGPT crossed 100 million users — faster than TikTok, Instagram, or any consumer app in history. Artificial intelligence was suddenly no longer abstract theory or niche research. It was in classrooms, offices, and households across the world.

Billions in venture capital soon followed. Startups in every vertical — from biotech to sales to law — declared themselves “AI-first.” For a moment, the trajectory seemed unstoppable.

But history warns us otherwise. Every technology boom leaves behind a graveyard of companies that grew fast but failed to endure. The dot-com bubble had Pets.com. The mobile app rush had Flappy Bird. Velocity and hype may propel you into the spotlight, but they rarely guarantee survival.

The defining question of this era is defensibility: which AI companies can convert early growth into enduring advantage — and which will be overtaken by competitors using the very same tools?

The Fragile Early Moats

Traditional startup moats — scale, distribution, brand — are fragile in the AI era.

• Low barriers to entry: With APIs from GPT, Claude, or Mistral, a “good enough” product can be built in weeks.
• Rapid commoditization: Features don’t last long as differentiators; copycats proliferate overnight.
• Capital intensity: Training or scaling custom models demands billions in compute and deep partnerships with hyperscalers — far beyond the reach of most startups.

Early advantages, then, are shallow by design. They resemble the bailey of a medieval castle: fast to build but easily abandoned under siege. Real defensibility requires constructing the motte — painstakingly fortified, difficult to dislodge.

The Stanford AI Index 2025 shows just how fast moats can erode: inference costs for GPT-3.5–class models fell 280-fold in under two years, while the gap between open and closed models narrowed significantly. Even technical advantages don’t last.

Consider EvenUp, the legal AI firm valued at over $1 billion. It now serves more than 1,000 law firms with AI-drafted demand letters. Yet its tools still require close human oversight — a reminder that market traction doesn’t always equal unassailable defensibility.

What Moats Still Hold?

Despite the fragility, some moats endure — if built deliberately.

• Network effects: The strongest moat in tech, when genuine drives the race. GitHub Copilot becomes more valuable as teams use it together. Character.AI thrives on user-generated agents fuelling emergent engagement. But shallow virality, as Groupon showed, can collapse overnight.
• Data moats: Proprietary, real-time, or domain-specific data still provide an edge, as Tesla’s driving data demonstrates. But static datasets degrade quickly, and open-source models reduce exclusivity.
• Distribution & brand: Perplexity and Cursor prove that modern PLG and community tactics can spark adoption. In a market plagued by hallucinations and privacy risks, trusted brands matter. Yet distribution and brand are easily mimicked if not backed by substance.
• Scale & embedding: Compute scale boosts performance but favors giants. Workflow embedding — like EvenUp integrating into legal processes — creates switching costs, but requires time, trust, and reliability.

The lesson is clear: no single moat is enough. Defensibility must be layered, sequenced, and reinforced.

Lessons From Google… and Groupon

Google wasn’t the first search engine, but it was the first to sequence moats effectively:

1. A superior algorithm and simple UX.
2. A distribution flywheel via AdWords.
3. Network effects from scale and advertisers.
4. Workflow embedding with Gmail, Maps, Android, and Chrome.

The result: dominance that endures to this day.

By contrast, Groupon sprinted to scale but never transitioned. Its viral growth and investor hype weren’t matched by lasting network effects or loyalty. Once the novelty wore off, its moat vanished.

The parallel for AI startups is stark: move too slowly, and competitors will outpace you. Move too fast without building a deeper foundation, and you risk Groupon’s fate.

Three Tests for Builders

Ask yourself:

1. Switching Cost Test: What does a user lose if they leave?
2. Collaborative Value Test: Does value increase as more people adopt it?
3. Hub-and-Spoke Test: Does your system create outsized rewards for key users, anchoring their loyalty?

If the answers aren’t compelling, your moat is little more than a trench in the sand.

The Next Frontier

Three emerging patterns may define the next generation of AI defensibility:

• Personal Utility Networks: AI copilots that accumulate unique context — your workflows, history, and processes. Leaving means losing memory.
• Hub-and-Spoke Ecosystems: Platforms like Character.AI elevate select creators or agents, pulling in entire user bases through asymmetric rewards.
• Agent-to-Agent Networks: Tomorrow’s agents won’t just serve humans but interact with each other, creating compounding network effects akin to the early internet.

Wisdom From Builders

“When people say that an entrant is disruptive, what they really mean is that customers are adopting that new way. … If you can invent a better way, and if customers agree, then they will use it.”
— Jeff Bezos

“If you’re not embarrassed by the first version of your product, you shipped too late.”
— Reid Hoffman

Recommendations for Long-Term Defensibility

1. Set milestones early: Define defensibility goals — network effects, embedding, switching costs — by funding stage.
2. Measure depth, not just speed: Prioritize churn, retention, workflow embedding, and value lost when switching.
3. Invest in unique, evolving data: Proprietary datasets paired with real-time feedback loops are still powerful.
4. Balance offense and defense: Chase growth, but begin layering deeper moats once product-market fit emerges.
5. Protect trust as a moat: Guardrails on privacy, reliability, and ethics are essential. Once broken, trust is almost impossible to rebuild.
6. Build a culture of durability: Encourage experimentation but enforce rigor in quality, reliability, and post-mortems.

Conclusion: Building Castles That Last

AI will mint both giants and failures. Some castles will rise in the cloud, fortified for the long term. Others will shimmer briefly before washing away like sandcastles at the tide line.

The difference won’t be who moved fastest, but who built with discipline. Who layered their defenses, sequenced their moats, and earned trust.

For founders, the call is urgent: build not just for today’s adoption, but for tomorrow’s resilience. For investors and leaders, the challenge is to look past hype and ask — when the flood of competitors arrives, what remains?

Because in the end, the AI race won’t be won by those who sprint first — but by those who build castles that endure.

Featured photo:  Google DeepMind: https://www.pexels.com/photo/an-artist-s-illustration-of-artificial-intelligence-ai-this-illustration-depicts-language-models-which-generate-text-it-was-created-by-wes-cockx-as-part-of-the-visualising-ai-project-l-18069696/