Send me to fall, send me to fall
You’ve got front row seats
– misheard saoirse dream lyrics

Every time they train a new frontier model, they do a calculation. What’s the most efficient way to make money off of this model? For a while now, it’s been selling access to it like a SaaS subscription. You buy access to the model, you use it to make money, some percent of the money you make pays the API bill, etc…

In a growth economy, this calculus works. The economy grows, and they don’t increase their share of it, they just get more because the economy is growing. The primary driver of economic growth is onboarding of new users, except ask your preferred model about the “fertility crisis” and realize we don’t do that anymore. (ahh never mind Nigeria is growing, we are saved)

So we’re in a non growth economy, and without global growth, you still need growth for yourself. The big tech companies all experienced this. The only way to get growth for yourself is to take a bigger share. First from your users, then from your business partners, then from your employees. You start eating yourself.


The AI application layer will be worthless. The reason isn’t that it’s going to be commoditized, it’s that this will be the first place the model makers will come for in their hunt for verticality. At first, you’ll see some defect and continue to provide API access, but eventually the market will consolidate to 2 or 3 players.

Unfortunately, the quality of a model scales pretty clearly with the amount spent on the training run. You can have 10x hits like deepseek, or -10x misses like GPT-4.5, but it all pretty much follows the rule. You want performance, you spend. So it cost a lot, and very few can afford to be on the frontier.

They want to best recoup their investment, and the way to do that is not to provide unfettered API access. Many things in the world are Red Queen’s races, and with a bit of coordination, there’s more profit to be made for all if all the frontier labs coordinate.

So you’ll see an era of market segmentation. Way beyond just personal and business, it’ll be per industry. One price for finance, one price for cybersecurity, one price for copywriting. And rolled out to “preferred partners” (aka people who paid us) first. You pay to be early. You pay so other don’t get it. The frontier labs uncoordinatedly coordinate to calculate the maximum to siphon off so we can suck the most out of everyone, over whatever time horizon they are thinking on.


The core limiting factor of most industries in America is intelligence. Think about what value you think you add and why your employer sees it worth it to give you some of the profits. It’s probably not your muscle power.

This only ends up in one place. A continued climbing of the vertical. Why are you still in the loop at all? Don’t think voting will save you. You’ll have 0 earning potential. You’ll have no money to buy stuff, this is the way capitialism ends.

There’s only one way to prevent this, and it’s preventing anyone from monopolizing compute, the resource needed to train and run models. Make sure it stays distributed. Oh wait, it’s already too late for that. 60% of the global compute is owned by the 5 US hyperscalers.

I think there is a world market for maybe five computers.

IBM was just early.


The markets are thinking on an extremely short time horizon. After AI takes all the jobs, what exactly happens? How many of those jobs no longer exist now that AI took all the jobs that paid the people to buy the stuff that those jobs produced? Truck drivers? Why drive a truck to Topeka full of stuff if nobody in Topeka has jobs to buy it? We live in a society and we work for each other. This is like your body consuming its muscle to stay alive, and pretty soon after that, you die.

There is a way out of this, but the world isn’t ready for it yet. Way too much 0 sum thinking still dominates. Someday we will realize the universe is putty in our hands, and it never had to be like this. But that day won’t be today. Or tomorrow. The demoralization is just beginning.