the floor

The price of artificial intelligence, measured as the cost to the person using it, is collapsing. This is the good news. The cost of building it, measured as the money required to run the servers and buy the chips and keep the lights on in buildings that cannot be allowed to get warm, is not collapsing. This is the other news.

Bloomberg reported this week that equity investors have spent the summer trying to work out whether this is good news. This seems like a reasonable way to spend a summer, given that "the most powerful technology in human history is becoming free to use, and the bill is going somewhere" does have some features that would normally require clarification.

The way it works, in simplified form: AI models improve through training, training requires computation, computation requires hardware, hardware is expensive, hardware is getting more expensive, and the companies making hardware are doing extremely well. Meanwhile, the companies selling access to the AI are engaged in a competition to charge less for it, and in several cases have moved to charging nothing for it at all, on the premise that if enough people use the free version they will eventually subscribe to the paid version, at which point the math is expected to work out.

The math is still being worked on.

In a normal industry, the dynamic runs the other direction. You manufacture something, you sell it, as you manufacture more of it the per-unit cost drops due to economies of scale, eventually the thing becomes cheap, then very cheap, then a commodity, and somewhere along the way it becomes profitable for someone. This is the normal arc. Steel went through this. Electricity went through this. Personal computers went through this.

What appears to be happening with AI is the reverse. The more people use it, the better it needs to be, and making it better requires more computation, not less. The user's experience is that the thing is free and very good. The industry's experience is that the thing is increasingly expensive to provide and is expected to somehow remain good while becoming cheaper. These two experiences are related. One of them is yours.

The equity investor, who is the entity Bloomberg identified as having spent the summer on this, is in the position of having given money to people who are building very expensive things to give away for free, in the hope that this is the part of the story that comes before the part where it becomes profitable. This is a reasonable thing to believe. It has historically been true of certain industries. It is not yet clear whether it is true of this one.

The investors are still working on it. They have until next quarter's earnings call.