Member-only story
AI is profitable. Its problem is another.
Don’t let margins fool you
Contrary to popular belief, AI, as an industry, is now poised to make money. I know this sounds hard to believe, as everyone assumes AI bleeds money.
But that’s very likely wrong now, and I think I can prove it to you.
And yet, this doesn’t necessarily speak well of AI. The real problem is another.
This post aims to clarify where you should look for answers and how incumbents are telling half-truths to keep you from seeing the real problem, which is potentially just as bad.
The Lemonade Stand Fallacy
In an industry filled with unsubstantiated promises, incumbents, companies, their investors, and compromised analysts are all pumping the idea that AI is “now profitable.”
And it’s funny that I say this, but they are right. AI inference, serving AI models to customers, excluding training and equity-compensation costs, now has very good margins.
Inference margins can be closer to 85%, as DeepSeek’s CEO Liang Wenfeng acknowledged, and US frontier Labs are known to have higher than 80% margins (on the API business, not on subscriptions, though).
