Forget the Moat

“A startup is a company designed to grow fast.” -PG

Why Moats Matter for Software Startups

Almost every startup burns cash while chasing growth. That is intentional. Investors fund these loss-making machines because they expect them to eventually become large, high-margin businesses whose future profits will dwarf their current losses.

Software companies are especially attractive to venture investors. They can scale distribution at the speed of the internet, and once the product works, each new user costs virtually nothing to serve. That’s how Microsoft sustains ~45% operating margins, while Walmart operates at around 5%.

But the same factors that enable fast growth also make software companies vulnerable to competition once they hit steady-state profitability. A small team can build a Microsoft Excel competitor, but no one can build a Walmart rival without billions of dollars in capital.

For a software business to deliver durable, outsized profits, it must develop intrinsic barriers to entry. In other words, it needs a moat.

Moats vs. Competitive Advantages

Some competitive advantages evolve into moats, but not all do.

For example, OpenAI had a massive first-mover advantage when it launched ChatGPT. But that edge didn’t translate into a true moat. Competitors like Anthropic caught up quickly.

Sergey Brin and Larry Page’s PageRank algorithm initially gave Google a superior search product compared to existing search engines. They were able to transform the high usage of their superior product into a real moat as they layered user behavior data (clicks and engagement) into the algorithm, creating a data flywheel that improved the product with usage. No matter how much money competitors like Microsoft poured into their search efforts, they couldn't make much of a dent, and the result was a cash cow that has financed Google’s empire.

Types of Moats in AI

There are many types of moats: Brand, economies of scale, data flywheels, high switching costs, regulatory barriers and patents, talent, etc. It is still too early to say which moats will prove to be durable for AI-native businesses. People don't even agree whether the model is the product, or a commodity. However, it seems there are differing moats emerging for AI infrastructure (foundational lab) vs. applied AI companies. On the AI infra side, the major labs are trying to build moats around economies of scale and talent. And on the applied AI side, it seems that moats based on brand and switching costs related to long-term memory and owning customer context may be emerging.

OpenAI is a great case study. It is increasingly two separate entities: A leading foundational lab building the world's best closed-source models and a consumer AI application company with products like ChatGPT and Sora. The lab’s moat comes from compute scale and concentration of top talent. The application moat, on the other hand, consists of things like the "ChatGPT" brand and the increasing memory and context it has about its users. If the Sora app takes off, it will be a classic social media network effect and data moat.

One area that seems surprisingly underdeveloped is the data flywheel for reinforcement learning and fine-tuning. Very few companies are meaningfully building continuous learning loops to modify their models' weights and biases based on user interaction. This may be because today’s LLMs built on the transformer architectures don’t support true continual learning well.

Moats in Early-Stage Funding Rounds

With money pouring into AI startups, moats often come up in funding conversations. That’s understandable. If investors write a check for an agentic framework startup, they want to know whether the first-mover competitive advantage will translate into a durable moat by the time OpenAI launches their own.

But at the pre-seed (and even seed) stage, founders shouldn’t waste time over theoretical moats around businesses that don't exist. The only job is to get to product–market fit.

Good investors at these stages will be looking for signals of founder-market fit (pre-seed) or early traction (seed), not for long-term defensibility. It’s fine if they ask about moats simply as a litmus test to gauge your strategic thinking. But if they pass because the moat isn’t “strong enough,” that usually means they don’t believe in the founder or the market, and they’re sugarcoating the rejection.

Parting Thoughts

If you’re an early-stage founder: focus on traction — and forget the moat.