Bloomberg reports this week the New York Times is suing OpenAI and Microsoft for use of its copyrighted content in training AI models. I don’t know enough about the particulars of the case to comment on its merits, but it raises an interesting question for AI investors.

The profits of any enterprise are the revenues it receives from sales of its products and services, less its cost of producing and bringing them to market. When you buy oil from Exxon, for instance, what you pay aren’t Exxon’s earnings; most of what Exxon receives represents its cost of producing the oil and getting it to you. It’s earnings are only the difference between the two, its compensation for its addition of value.

I’m not an AI doubter. I’ve said before I think it’s real and spectacular. It’s not new, but it will change the world. My skepticism rather has been that investors, in their emotionally addled excitement, aren’t crunching the numbers and are waaay overpaying for things based on their association with AI, much as they did over twenty years ago merely because of their association with the Internet, when a company could multiply its stock price just by putting dotcom in its name.

The question here though actually cuts to the heart of the profitability story itself. Suppose Exxon simply failed to pay the costs associated with getting the oil it sells. It could report profits many times higher as a result. But it wouldn’t be sustainable. In fact it would be fraudulent.

AI firms are in the business of selling information and entertainment. They too earn profits by producing a product from raw materials and adding value. But what if they were doing this without paying for the raw materials? They would appear to be far more profitable than they really are; they would be taking credit not only for the value added but the value of the materials they used to produce it. Such apparent profits would not be sustainable; eventually the producers of the products AI businesses use to produce theirs would want to be paid. As productive as the AI firms may be, the appearance would be much greater than the reality. Much of the value in their products was actually produced by those that did the considerable legwork in gathering the masses of original data used to train AI models.

Again, I don’t pretend to know the facts of this or any other particular case, but it certainly raises questions any prudent investor should carefully consider before committing funds.