By The DCo Podcast
Date: TBD
This summary provides a framework for evaluating the economic viability of decentralized AI projects. It identifies the rare patterns that actually generate revenue in a sea of speculative tokens.
Ejaaz Ahamadeen argues that crypto has only mastered two economic designs: hard money and payment processing. As AI integrates with blockchain, builders must decide if they are selling raw compute or gated intelligence.
"The crypto business model has only been figured out for a few things."
"What the stack has that's most valuable is the model itself."
"Revenue positive crypto apps are ones that either don't have a token or just had a token."
Podcast Link: Click here to listen
I think the crypto business model has only been figured out for a few things, really few things. You could argue that it's a monetary asset, which is Bitcoin, and you get emissions from that. Eventually, once that emission schedule declines exponentially, you're hoping that fees on the Bitcoin network will be worthy enough for miners to do their thing, or for people to transact and run nodes and validators. So that's one thing. The other thing is stable coins. Stable coins are like, okay, cool, we'll take just a fee. It's kind of like the Visa model in a way. Visa actually makes a decent chunk of money off of fees from that.
When it comes to the AI stack, you can look at it in similar ways. Let's approach this. Let's look at the bottom layer of the stack and let's look at the top layer of the stack because there are two arguments for each one. On the bottom layer of the stack, where it comes at the infra layer, you're betting that the layers above it, so the coordination network, the middleware network, and the app layer, are all paying some cost or fee back to that network. So it's like, okay, we're going to distill these components down to the most valuable things: compute, data, and models. Everyone's going to need it, and maybe I make all my money from inference costs or training costs that are fed down to me.
The reason why I say inference and training is at the end of the day, what the stack has that's most valuable is the model itself. So if you're training a model, you need to pay the compute providers and the data providers. Maybe that's just through their own token, and maybe it's just like a payments mechanism. That's what I'm seeing right now today. I don't know how useful that'll be. Another kind of model that I'm seeing is they pay each other or they pay for each of their resources in whatever currency and then they buy back their token and burn it.
My question then becomes what happens when these tokens reach a certain point where it becomes inordinately expensive to use or whatever that might be. That's an open question for a lot of crypto protocols. It's not specific to AI at all. This is a crypto problem that needs to be resolved.
On the application layer, if I'm being honest, you see a similar thing. All the AI apps that I've seen exist today in the form of agents use some kind of staking or payment model. The common example is pay to access this terminal where this agent will now give you alpha, or stake a certain amount of this token and now you can get different tiered access. That's cool and everything, and I'm seeing there's potential buybacks and other agents and stuff. It's all cool, whether it's sustainable in the future, I have no idea because most of these things aren't revenue positive yet.
If you want to talk about revenue positive crypto apps, they're ones that either don't have a token or just had a token. I look at Kaio as the most obvious example. They're technically profitable, but they have an AI powered app versus a cryptonative AI app, if that makes sense.
