In a world where AI can build any software, replicate any product, and automate any process, what actually makes a business defensible? Most of the answers I hear are wrong. They're all bets on the same thing: that intelligence doesn't get 100x faster, better, and cheaper. It will. Soon.
Defensibility has always come from two things. Things that are hard to do and things that are hard to get. AI is making the first one worthless. Building software, maintaining integrations, embedding your product so deep it takes a year to rip out? Hard to do. And increasingly easy. But having ten million users, a government license, a chip fab, a billion dollars in deployable capital? Hard to get. AI compresses the time it takes to do things. It does not compress the time it takes for things to happen. That distinction is now the single most important filter I apply when investing. Five things pass it.
Compounding proprietary data. Not all data qualifies. A static dataset that was merely expensive to collect will eventually be synthesized or worked around. That's hard to do, not hard to get. The moat is living data: proprietary information continuously generated through operations that are themselves defensible. Orchard AI mounts cameras to farm equipment and tracks billions of fruit across millions of trees across multiple growing seasons in multiple geographies. The data changes every day and every pass through an orchard makes the dataset richer and the models smarter. You can’t replicate that by training a model on public data. You’d have to drive the same cameras through the same orchards for years.
Network effects. Every user that joins makes the product more valuable for every other user. DoorDash is the clearest example: every driver makes delivery faster, every restaurant gives customers more choice, every customer makes the economics work better for everyone else. Clone the app overnight. The drivers, restaurants, customers, and delivery density in ten thousand cities don't come with it. An exchange fund like Cache shows how this works beyond marketplaces: a larger pool means better diversification, which attracts more participants. The cold start problem might actually get harder as AI makes it trivial to build competitors, because now a hundred well-built alternatives are fighting to bootstrap the same network. Whoever already has the liquidity compounds. Everyone else fights over scraps.
Regulatory permission. Governments move at the speed of politics, not technology. Anduril needs procurement clearances and classified contracts to sell to the Department of War. No amount of AI compresses that timeline. A bank charter takes years. FDA approval takes years. And the surface area of regulation is expanding, not shrinking, because as AI capability increases the stakes get higher. Will the specific forms change? Almost certainly. Will the need for human permission go away? I can't see how.
Capital at scale. The one almost everyone is underweighting. The endgame is physical. Chip fabs cost $20B. Nuclear plants cost $10B. Satellite constellations cost billions more. There's a reason Elon is raising $75B and IPOing SpaceX even while saying money might not matter in 15 years. When the bottleneck shifts from software to atoms, the ability to finance and deploy capital at massive scale becomes one of the defining advantages of the era. And capital access isn't just money. It's institutional trust, track record, and relationships that take decades to build.
Physical infrastructure. Factories, power plants, battery networks, data centers. Base Power is deploying thousands of battery units across Texas homes while building its own manufacturing facility. Every unit installed is a physical asset in somebody's yard generating revenue. Soon you can design the system in a week with AI. You cannot manufacture, install, and interconnect thousands of units in a week. Physics sets a floor on timelines that intelligence can't break through. Whoever started building first has a lead that grows every month a competitor hasn't broken ground.
I wrote about endgame positions recently, the assets that get stronger as AI gets better. Every one of them shares this property. They required years of real-world elapsed time to accumulate, and no amount of intelligence compresses that clock. Network density takes years of human adoption. Regulatory approval takes years of political process. Infrastructure takes years to build. Data takes years to compound. Capital relationships take decades to earn. Time that can't be parallelized is the meta-moat underneath all five. And the companies that already hold these positions aren't just defensible. They're pulling further ahead every day, because the head start is the moat.
The things that didn't make this list are the things where time can be compressed. Workflow embeddedness sounds durable until you realize the switching cost is really just engineering time in disguise. Ecosystem lock-in sounds like a fortress until AI can rebuild integrations as fast as you can describe them. Software scale, spreading engineering costs across millions of users, stops mattering when engineering costs approach zero. These were moats against the scarcity of intelligence. That's the one form of scarcity we know is ending.
There are things I'm still genuinely uncertain about. Does trust become its own moat when AI is doing more of the work? Someone has to be accountable when things go wrong, and the institution that bears that liability might become more valuable, not less. Is human attention a moat? When the cost of creating content drops to zero, the brands that already hold attention might have the most compounding advantage of all. I think both are real. I haven't figured out whether they're their own categories or byproducts of the five I've named.
But the core filter holds. Hard to do, or hard to get? If your moat is bottlenecked by intelligence, you're on borrowed time. If it's bottlenecked by years, by human behavior, by physics, by political will, by capital, you're probably building something that lasts.