What Counts as a Moat in the Age of AI?
TLDR: Last weekend a friend who runs a nine-figure AI services business told me he has “no conviction in his long-term moats” of anything he’s built, including the parts working best. He isn’t depressed; he’s paying attention. Patents, proprietary code, distribution, even Nvidia’s CUDA, the most valuable software moat on Earth: AI is inventing around all of them. Two things still hold, and neither is what business school taught you.
A decade ago, I was in a conversation with Astro Teller, who runs Google X, and Steve Jurvetson, one of the great venture investors of the last thirty years, at Singularity University. We landed on a conclusion none of us loved. In a world of AGI/ASI, patents will be worth far less than they are today, because an AI that can read every patent ever filed can likely design around every patent ever filed, in hours. If your company’s protection is a patent, you are cooked.
Then, last Sunday, I had lunch with a dear friend who is the founder of a company that provides training data for OpenAI. That business hit $160 million in revenue and has since fallen to $50 million, not because the work got worse but because the market moved. His words: “I had no conviction in the long-term moats in that business.” So he rebuilt around something else, which I’ll come back to.
And on Moonshots podcast last month, Dave Blundin, Ramez Naam, and I spent three separate segments on one question: is Nvidia’s moat real? Five trillion dollars of market cap hangs in the balance.
I’ve spent thirty years telling entrepreneurs to build moats. This newsletter is about why that advice is expiring, and what will replace it.
WHAT A MOAT USED TO BE
Warren Buffett popularized the word in the 1990s: a durable competitive advantage that lets a business earn above-average returns for decades. The classic list has five entries. Patents and trade secrets. Proprietary technology. Network effects. Switching costs. Brand.
Every one of those assumed something that is no longer true: that copying an existing business is expensive and slow. A competitor had to hire the engineers, reverse-engineer the product, build the distribution, and wait years. That moat offered years, perhaps decades of protection.
AI collapses time and effort.
THE MOST VALUABLE MOAT ON EARTH IS LEAKING
Take the strongest example anyone can name: CUDA, Nvidia’s programming layer.
Dave Blundin explained it on our August 28 Moonshots episode better than I can: “Every AI researcher on Earth writes in Python and PyTorch. CUDA is the layer that translates that code into instructions that NVIDIA’s chips run at blazing speed. NVIDIA spent years of human engineering time building it before anyone cared about AI. When AI took off, it was the only door, because porting all that math to AMD or anyone else would have been a nightmare.” That, Dave said, is specifically why Nvidia is the most valuable company in the world.
Now listen to Ramez Naam, who said on August 4: “The CUDA moat is broken this year and next.” Why? Because you can now tell a frontier model, “take my AI code and recompile it to run fast on this AMD chip,” and it does. Dave confirmed he’s been writing custom kernels for Fable 5 himself over the last few weeks. Work that used to require a team of specialists is now done with a prompt.
Dave’s conclusion: CUDA as an inference-time moat is already dead. As a training-time moat, limited lifespan. Ninety to ninety-five percent of the AI workload is moving to inference.
Here is what makes Jensen Huang one of the best CEOs of his generation. He saw it coming. NVIDIA bought Mellanox for $7 billion in 2019 to own high-speed interconnect, so that if you want to run 100,000 or a million GPUs as one coherent machine, you still have to buy NVIDIA. Dave’s phrase: “Your moat is only good until the next moat. Jensen is moving Nvidia’s moat very intelligently.”
“Your moat is only good until the next moat. Jensen is moving Nvidia’s moat very intelligently.”
That is the new playbook. The moat is not a wall you build once. It’s a wall you iterate on and move faster than anyone can attack it.
CHIPS ARE THE EASY CASE
Procter & Gamble and Gillette have owned deodorants and razors for a century. Brand, shelf space, manufacturing scale, the textbook moats. Now imagine someone genetically engineers a bacterium that you keep in a petri dish in your closet to eliminate body odor for a year. Synthetic biology is disrupting a multi-decade-long moat.
Software is worse. Any SaaS product with a visible interface can now be cloned by a frontier model in a weekend. Already, OpenAI is doing this in finance and legal. Anthropic is doing it in design. Switching costs, the moat that protected enterprise software for forty years, evaporate when an AI agent does the work for you in minutes or seconds.
Even distribution is wobbling. The Coasean Singularity paper released from MIT and Harvard two weeks ago argues that when AI agents search, compare, negotiate and transact for us at near-zero cost, the customer relationship stops belonging to whoever owns the storefront and starts belonging to whoever the agent trusts.
THE THREE MOATS THAT SURVIVE
So, what’s left? Across all four of those conversations, the same two answers kept surfacing.
One: speed of reinvention. The question is no longer who has the best moat. It’s those who are reinventing themselves faster than anybody else. Jensen is moving from CUDA to interconnect. Elon is shifting Tesla from manufacturing cars to offering robotaxis services, while SpaceX goes from rockets to Starlink to orbital compute. The durable advantage is the organization's metabolic rate, and AI is the fastest way to raise it.
“The durable advantage is the metabolic rate of the organization, and AI is the fastest way to raise it.”
Two: capital velocity, which shows up as balance sheet. On the August 10 Moonshots episode we dug into what actually protects the hyperscalers. Not access to compute; anyone can rent an H100. Not scaling ability. The moat is that Microsoft, Google, Amazon and Meta can pay for hundreds of billions of dollars of GPUs immediately, out of cash flow, and already have the customers to fill them. The hyperscaler is just a financing system. It’s a real estate game. The oil analogy is exact: a liquid global oil market didn’t hurt OPEC, because the moat was having the oil in the first place.
Three: Brand. There’s a third, smaller moat… brand. Brand still holds up at the high end, because a luxury brand is a signal of the buyer’s wealth, and no AI can clone what a Hermès bag says about the person carrying it. For everything else in the low-end consumer space, brand is a head start, not a wall. The future of AI auto-magically transacting your purchases of toothpaste or paper towels doesn’t care about brand; it cares about cost, the ingredients, and thousands of online reviews that it consumes.
Note what’s missing from the list. Patents. Code. Features. Distribution. Everything that can be described can be copied, and AI can describe anything.
THE MACRO CAVEAT
Back in 1999, the Internet was the biggest thing ever. But it was the second owners, not the first, who made the money. Very few first owners, Bezos (Amazon) and Larry and Sergey (Google) being exceptions, pulled through.
You can hold a fully abundant view of what AI will do and still be blind to what happens to your capital structure when moats evaporate, and margins compress across every industry at once. GDP just printed at 4.7 percent annualized for the third quarter. Anthropic’s own report projects 15 percent GDP growth in its extreme case. That growth is real. The question is who captures it, and the historical answer is not always the pioneer.
WHAT THIS MEANS FOR YOU
If you’re an entrepreneur: Stop pitching your moat. Pitch your rate of reinvention. Show investors three pivots you’ve made over the past 12 months and the cycle time for each. If your defensibility slide says “patent pending,” delete it.
If you’re an executive: Audit your moat list this quarter and ask one question of each entry: could a frontier model plus a motivated team rebuild your product or service in ninety days? Anything that fails is not a moat. It’s a head start, and you should price it that way.
If you’re an investor: Underwrite management speed and capital access, not IP. Francis’s rule from the services market applies everywhere: Google has 60 percent of search revenue and 90 percent of search profits; Accenture has 4 percent of services revenue. Know which kind of market you’re in before you assume a winner takes all.
If you’re a student: The skill with the longest half-life is learning faster than the tools change. Every specific skill you master today is a moat AI erodes. The ability to master the next one is not. Stay curious. Have a purpose.
If you’re a parent: Teach your kids that reinvention is normal. Mine are 15, and the most useful thing I can give them is comfort with the ground moving.
Jensen knows his current moats are temporary, and he’s the most valuable CEO on Earth. He also knows how to keep reconstructing new ones. The founders who get hurt in the next five years will be the ones who still believe they have a wall.
So here’s my question for you. If AI could copy everything you’ve built in ninety days, what would you still have?
Whatever your answer is, that’s your actual moat. Go widen it.
ONE MORE THING
The changes in education we are seeing are mainly about one thing: giving students agency and raising a generation that doesn’t wait for permission to build the future.
This Friday, I’ll be surrounded by 1,500 people who are doing exactly that.
I’m hosting Moonshots LIVE alongside Alex Wissner-Gross, Salim Ismail, and Dave Blundin, with Palmer Luckey, Cathie Wood, Neil deGrasse Tyson, Neal Stephenson, Astro Teller, Ben Lamm, Anousheh Ansari, Emad Mostaque, Logan Kilpatrick, Mark Pincus, Mira Lane, Nikhil Chandoc, Rod Roddenberry, and many others joining us on stage.
And in one afternoon, we’ll award two XPRIZEs live.
The $3.5M Future Vision XPRIZE started with 2,500 films. Five finalists will pitch on stage, and the winning film will actually get made.
The $2M Build with Gemini XPRIZE challenged builders to turn an idea into a real company in 90 days. Five finalists will pitch what they built, with $500,000 going to the first-place winner.
Both will be decided in the room on Friday. VIP is sold out, while we’re only 1 day out.
To a future of abundance,
Peter