Tech people love to talk about competitive moats—those things that make it harder for new companies to take their existing customers. Maybe because tech people all want to live in castles. But the idea is hardly specific to tech: Pat Dorsey is a well-known finance analyst who classified businesses by their moats while working at the finance research firm Morningstar, and wrote books about it. Morningstar even has moat ratings; they’re big moat guys over there.

The Morningstar Moats are: Network effects, switching costs, cost advantage, efficient scale, and intangibles. I think it’s worth thinking through each moat vis-à-vis the current AI landscape. Let’s run them down, then get to the AI part.

Network effects: The service gets more valuable the more people use it. Think LinkedIn, TikTok, WhatsApp, Facebook, Instagram, or Substack. These are some of the best-known internet moats; we’re all living inside of the network effects. Also Visa and Mastercard, or the New York Stock Exchange.

Switching costs: It’s expensive to change services. A good example would be home insurance—once you’ve talked to an agent and been approved, you tend to keep paying their premiums forever. Or choosing a primary care physician. Tech people often call this “lock-in.” It’s interesting that a lot of tech “switching costs” companies are secretly network effects in a trenchcoat. The Bloomberg Terminal has amazing proprietary data, but it also serves as a rolodex for Wall Street. Airtable isn’t radically different from other low-code platforms, but once everyone starts using it, it’d be a real pain to move them off. Most software-as-a-service companies would love to get some network effects, but they’ll settle for switching costs.

Cost advantage: Think Costco. It’s in the name—it’s their whole thing. It’s not always the most convenient, but you’ll buy so much toilet paper that it won’t fit in your Land Rover. Digitally, think Zoho. Of course, once you wedge in on cost advantage, you keep going on the other moats. Android (or Windows on a Dell) tend to be cheaper, but once you choose that platform they try to get you on their networks and locked in to their platforms, so your switching costs go up.

Efficient scale: They use their enormous size to keep costs down—and it’s hard to compete with them as a result. Amazon Web Services is a good example here. There are cheaper or easier hosting services, and there’s Google Cloud and Microsoft Azure, but…AWS. Mailchimp is another one: The email they send tends to get to people, which is really hard to do at a smaller scale–there’s no such thing as “just send email” in 2026.

Intangibles: This often gets shorthanded as “brand.” Examples: Apple (think blue bubbles versus green in the messages app) but also Supreme (people line up to buy some red letters on a piece of merch). A lot of other stuff goes under here, too: “Product strategy,” or “knowledge,” or “skills,” or “culture.” Many digital people live their entire lives inside the intangible moats. Moatwise, though, these are tricky—because people can leave.

So those are the Morningstar Moats. Now let’s run them down for this new, radically disrupted world of AI. Spoiler: A little less change than you’d expect, with some big exceptions:

Network effects: As far as anyone can tell, and somewhat disappointingly, the big social networks are doing just fine. AI slop is everywhere, of course, but that hasn’t stopped anyone from posting on TikTok. People are still sliding into DMs all the time. OpenAI took a swing with their video network Sora, but they’re shutting it down later this month because…I mean, did you use it? People seem to want to interact with bots about as much as they want to interact with brands, and even less with brand bots, unless they can convince them to give 100% refunds. Most AI seems to be experienced one-on-one, not in a network.

Switching Costs: I think this is one to watch. AI is really, really good at transforming one kind of data into a different kind of data. The cost of switching from one CMS to another, from one ERP to another, from one CRM to another—these are going to go very low. You may just be able to…say them, and they’ll happen. As a veteran software person, migration is one of the true sticking points for any project, and it drives costs way, way up. Data isn’t everything, of course. You have to factor in basic human intransigence, and the cost of retraining and getting new licenses, or of getting out of old contracts. Those will remain high. All I’m saying is that one piece of friction is gone—but it’s a big piece.

Cost Advantage: This one is interesting. Building complex products typically costs millions of dollars. But today, with dozens of dollars via vibe-coding, you can get a credible facsimile. At first, I thought of AI-generated software as “slop products,” but as time goes on I’m not so sure. And I’m very curious to learn what the world wants here. Because there have always been cheaper alternatives, and yet more people seem to want to use Salesforce over Zoho, or an open-source competitor. Will this be the era of $2 Photoshop? We talk about this as the “$15 Volvo Problem.” No one wants to put their children in a $15 Volvo, and yet if a Volvo truly did cost $15 and was about as good, or more purpose-suited to your needs, people may want to buy a $15 Volvo.

Efficient Scale: It’s truly hard to imagine AWS going away in ten years. It’s just relentlessly efficient, and willing to pursue every advantage. That said, the hosting company where you tell Claude or OpenAI, “Just host this securely on SuperSecureAIHost,” and it does everything correctly—that could gain a lot of market share. But I also wonder if there will be room for small players, because Anthropic could pretty easily boot up the “AI-first AWS,” powered by Claude(™). I imagine that, given their need for revenue and general nerdiness, they’re working on something like that now. Which is too bad, because I’d love to see lots of dynamic exploration in the hosting space. 

Intangibles: This is where it gets wild. Humans need signals to know when something is “good,” and I think for a long time, we will be skeptical that AI-generated things are “good.” So I expect marketing to steadily shift from “we jammed AI into everything,” which is what the stock market likes to hear, to “we are incredibly human just like you, friendly human,” which is what actual people like to hear. 

But that’s “just” brand. A lot of us—me too—have lived our full careers inside the “intangibles” space. We focused on vague things like “product quality” and “proven process” and other work that is impossible to explain to your uncle at Thanksgiving, so you just end up talking about the Eagles (of American football, not of “Hotel California”). We intangibles types feel the sting the most when someone says, “Hey, Claude made me a PowerPoint that’s pretty good, so I don’t need you to make it,” or “I built my own website for this actually, so we’re set. Let’s stay in touch.”

Anthropic’s CEO, Dario Amodei, keeps talking about AI being a “country of geniuses in a datacenter,” which is how consulting firms have portrayed themselves for decades. So you’d expect consulting to be in trouble—but right now giant consulting companies are growing, at least in part because they are telling everyone they understand AI. I know it feels like doom might be around the corner, especially when you hear companies are laying off thousands of people, but the world really is bigger than we often think. Change is coming, for sure, but if you take a step back, you can see that the moats are still there, and the castles aren’t quite collapsing, no matter how much some people would like to see them fall.