alt Hacker NewsTo me the main question is the long term pricing.
It is said that major providers more than break even on what they're charging.
But at the same time that's not the point of capitalism, is it? The point is to charge close to the value you're providing.
My lunch money is approximately $10 and I often blow through as much in Claude tokens generously provided by the company which hired me. But I'm not getting $10 value from those tokens, but much more.
The cost of entry to this market is extremely high. Should Anthropic win and become an almost monopoly, it is bound to keep increasing prices to the point, where the value it's providing matches the cost.
That's the endgame of every AI company out there. It's worth using these tools now, while there's still competition and moats weren't established.
Replies
This is where I landed too. At least in the js frameworks "wars" it was reading the docs and prototyping angular and react for free. To try and keep up with the latest AI tools you're expected to spend ~$1k / year.
I don't think there's a universe where Google stops competing here, though. The second Anthropic gets greedy, there will be alternatives.
> But at the same time that's not the point of capitalism, is it? The point is to charge close to the value you're providing.
The question is how large the value delta will be from open models. And I’m not sure if the cost of entry is really “extremely” high comparatively if as you project the market will be so profitable. Surely investors will see a chance to get a larger piece of that profit. While model training is costly, there is a ceiling imposed by the training material being limited (at least for text). LLMs also don’t have a network-effect moat like social media has, or a web search moat like Google has, or a chip technology moat like TSMC has. It’s unclear if a significant moat will emerge.
NVIDIA Jensen predicted today that soon engineers will spend $20k in tokens every month.
Luckily the models were built on copyrighted materials so hopefully the big players won't have strong legal standing to kill off model destilling. Then, with models like current deepseek or kimi k2.5, you are perhapy 1/2 years behind in their capabilities at the fraction of the cost. For real inference costs, look at prices on openrouter. For hobby, I wasn't able to burn trough more than $5 in a month.