Post

I was very bearish on AI to begin with, but recently I started experimenting more with APIs and building my own agents. And wow, I did not realize how heavily subsidized all the subscriptions were: 🧵 1/6

The $200/month Claude subscription I have is okish. Current frontier model capabilities are mediocre at best, but I can find $200/month worth of value here and there, so it's worth it for me. But, via the API though, I can burn $200 worth of tokens with a single prompt. 2/6

Most of the die-hard AI fans I talk to are using stuff like Cursor, Claude Code, ChatGPT, etc. All of those are heavily-subsidized to the point where those $20/month and $200/month subscriptions they're paying for are letting them burn thousands of dollars worth of tokens. 3/6

I'd guess some of the heavier subscription users are probably burning like $15-30k/month worth of resources. Then you have to consider the fact that the actual cost of API tokens may also be subsidized. These company are getting discount compute from cloud providers looking to expand. 4/6

The API token cost is *absolutely* subsidized by investors doing loss lead bets. But the more you squint, the dumber it gets. In talking with one of the largest players earlier this year, they noted that it's nearly impossible to calculate the true cost of serving tokens once subsidies stop.

Yeah, it's crazy. Every time I bring it up people are just like "well technological advancements will make it cheap". How? Like what is your basis for this? And then they just hand wave it away

This argument (I refuse to call it "line of thinking" because it is anything but) drives me bananas. Nobody who understands the first thing about how this house of investment cards is built would make such a statement.

IMO, we're 1-2 Chinese open-source quiet releases from everything collapsing. Just assuming some lab is going to come out with a Mythos-level model that can run on a Snapdragon out of nowhere, or we move on from transformer-based models into something completely different w/ diff compute needs.