I don’t get why some otherwise smart people treat API prices like the true cost, and consumer plans like loss leaders.
It’s simply differentiated pricing. ie. finding ways to to make those who can pay more actually pay more. Averaged out, even consumer plans will be covering their operational costs and likely turning a profit too.
Enterprises can pay API pricing and get all the privacy, auditing, security controls, proof of certifications, and all the special contract terms they want. Small businesses and individuals can get a standardized subset of these.
Custom agents and other potentially competing products usually have to pay API prices too.
This is just business…
Averaged out, even consumer plans will be covering their operational costs and likely turning a profit too.
How do you square this expectation that consumer plans turn a profit with the reality that Anthropic lost $42 billion on $4.6 billion of revenue?
Ignore capital investment for a moment… pure operational costs… Probably 95% profit margin on inference in that limited context - hard to know for sure, but it seems likely.
Looking at capital investment, Musk recently claimed that a new AI datacenter pays for itself in one year… with capital investment… That’s quite telling.
So, people and companies in the know are investing in more AI datacenters…
Now… some companies are not on a great/independent financial footing, relying on external investors and having made purchasing commitments, have creditors to pay, etc. if they overestimated demand for their product, or they fail to refinance, they could be in real trouble, and at the mercy of others…
As for comparing “lost” to “revenue”, it is useful to separate paper-only figures from circular capital deals from purchasing obligations with penalties from operational expenses from direct investments, expenses and revenues (ie. it’s complicated and everyone has their own take on it). Many companies run at a loss for years until they make a profit (or are restructured / sold, ultimately still being seen as good investments) … what makes Anthropic & OpenAI interesting are the vast amounts involved, and the vast market potential.
My own bet is that both OpenAI and Anthropic survive and thrive in an expanding industry because they will be seen as more valuable as going concerns than being sold for spare parts. Ownership structures may change, of course. The next year or two will be telling.
-
Elon is a reliable, shameless liar
-
Model training is not capex, it’s an operational cost
-
The “datacenter investment” is pure smoke and mirrors.
-
there is no demand for the insane amount of projected compute they’re building aside from anthropic and OAI
-
datacenter buildouts are effectively stalled, and this is being ignored by investors
-
-
The market for chatbot services is effectively saturated. All reported gains come from one of the big 2 cutting token costs to cannibalize the other one
It’s normal and right to have a healthy loathing of Elon. He does have the annoying habit of making good business bets though.
Model training is an investment with its own opex. The rest is accounting.
I think overall demand is growing … whether it will grow as fast as they hope/want, is a big Q. I think the average person/employee will only need a limited amount, and the demand will grow elsewhere. The world isn’t static, and your average household income is already peanuts compared to the capital sloshing about. I don’t like it, but I don’t see why the trend will stop. Hopefully there will be some serious societal change along the way, but I wouldn’t count on it. If there’s means to fuck over someone else to marginally improve one’s own position, there’s someone ready and waiting to do it. Invest in pitchforks, I say.
DC growth will continue worldwide (and eventually beyond) … but they are also getting increasingly efficient, along with the algorithms. We may reach an equilibrium of sorts for a while.
Chatbots are cheap… tokens burn with agents. If you have a problem that can be drastically sped up with agents, it is perhaps stupid to rely on lots of humans and time (from a business perspective). There will be a lot of fuckups along the way, which will be both amusing and sad. A few individuals and organizations will both nail it and be able to capitalize on it.
-
This seems to succinctly sum up the situation.
Or, change my mind.
