AI Artificial Intelligence Security Sentinel Password Cyber Security Ransomware Email Phishing Encrypted Technology, Digital Information Protected Secured Lock
getty
If you are Warren Buffett and you are thinking of buying a business, you worry about whether that enterprise has a moat. A moat is a deep defensive barrier around a business that keeps competitors from getting into your business and therefore stops them from having a chance to scale your corporate walls and pillage your treasury.
For some, it’s the brand. Let’s have a for-instance. Rolex is a watch, and a watch tells the time. An excellent time-telling watch costs $20, not the $10,000 Rolex charges at the get-go. The Rolex brand and its marketing are its moat, and it is wide, deep and impregnable.
Businesses where the moat is shallow and narrow are not only vulnerable, but also not especially highly valued. I’m struggling to think of a big company with a little moat, and that might just be because you can’t get big if you don’t have at least a reasonable one.
Mobile carriers are thought of as having little in the way of moats, but their moats are certainly too big for me to get to their walls and try to compete. However, for other mobile carriers and their prospective investors, their collective moats don’t seem anywhere near defensive enough to justify a fat valuation metric.
So let’s look at AI.
AI is a truly big deal, driven by AI models that are simply incredible. But it turns out that any given model barely has a moat at all. Anyone with a bit of money, a lot of cunning and some AI gurus can copy and paste the best of your model and ingest it into their own for a fraction of the cost you ponied up to train the beast.
AI frontier models therefore seem practically moat-less, at least as the industry and the law currently stand.
“Oops,” so many are saying.
Yet here is the thing: old AI already does most of the things users want, and though the frontier does amazing stuff, most people don’t need it and won’t pay for it. Ninety-eight per cent of the user base don’t want to vibe code and never will. That doesn’t mean frontier AI can’t be massively monetized or have a deep and wide moat.
Let’s go back to Rolex.
Its marketing and brand spin up a $20 watch’s functionality from nothing to a $10,000 price tag. It also gets certain models to $20,000 and beyond, yet the alleged primary functionality is still available for just $20 – or for free on the front of your mobile.
As the dealer told me when I asked him how long it takes to sell a $50,000 Gold President, “It takes 10 minutes when the right customer comes through the door.”
It’s about what the customer wants and needs, not about what you think it’s worth.
There is no doubting the incredible value in LLMs, all the while the sort of people who said the internet would never catch on now pour scorn on AI. Productivity can skyrocket, and it will in the generation of companies that can restructure their processes to leverage it. The tail will perish.
People who think AI has no moat and therefore has a broken business model are missing something.
Access will soon not be universal.
Increasingly, people will lose their access. People are used to free access online. While we all notice this being increasingly taken away, this trend will be extremely accelerated for AI.
Initially, it will be created by premium services and pricing that gets eye-wateringly expensive quickly. Soon enough, you will need to KYC to get access to certain models. Ultimately, you will need a licence and accreditation to use AI models of a certain strength at all.
You can’t fly a plane or drive a car without a licence. Why should you be able to use an AI that can hack a government without your face on a government-issued plastic card?
Leading-edge models will always get bigger, better, more powerful, more valuable and more dangerous. The simplest way to protect them from being ripped off by a cunning adversary is to get the customer in a queue and serve them your special magic on your terms.
Ah, we are once again back at Rolex.
Functionality will be segmented, access will be controlled, and IP will be protected – lo and behold, there is the moat.
Of course, AI is a weapon, and like any weapon at some level of strength, it is quite likely that only governments will have the privilege to use the most powerful models.
We get to drive a car, not a tank.
Will this cramp or limit the AI business model?
No.
Will Anthropic and OpenAI be worth a trillion?
Yes.
Is there a second place in AI for you, your company, your country?
No.
Is there a limit to demand?
No.
Is there a limit to supply?
Yes.
So rather than roll those binary axioms up into a conclusion, there is an easier way to judge the scale of the future of AI, even if not its shape.
If the world’s most successful billionaire entrepreneurs are going all-in on AI, that is more than a hint about how it is going to pan out.
If Rolex can engineer an incredibly deep and wide moat, AI most certainly can.