Comment by runako

Comment by runako a day ago

I am not willing to render my personal verdict here yet.

Yet it is certainly true that at ~700m MAUs it is hard to say the product has not reached scale yet. It's not mature, but it's sort of hard to hand wave and say they are going to make the economics work at some future scale when they don't work at this size.

It really feels like they absolutely must find another revenue model for this to be viable. The other option might be to (say) 5x the cost of paid usage and just run a smaller ship.

apinstein a day ago

It’s not a hand wave…

The cost to serve a particular level of AI drops by like 10x a year. AI has gotten good enough that next year people can continue to use the current gen AI but at that point it will be profitable. Probably 70%+ gross margin.

Right now it’s a race for market share.

But once that backs off, prices will adjust to profitability. Not unlike the Uber/Lyft wars.

Reply View 3 replies

runako a day ago

The "hand wave" comment was more to preempt the common pushback that X has to get to scale for the economics to work. My contention is that 700m MAUs is "scale" so they need another lever to get to profit.
> AI has gotten good enough that next year people can continue to use the current gen AI
This is problematic because by next year, an OSS model will be as good. If they don't keep pushing the frontier, what competitive moat do they have to extract a 70% gross margin?
If ChatGPT slows the pace of improvement, someone will certainly fund a competitor to build a clone that uses an OSS model and sets pricing at 70% less than ChatGPT. The curse of betting on being a tech leader is that your business can implode if you stop leading.
Similarly, this is very similar to the argument that PCs were "good enough" in any given year and that R&D could come down. The one constant seems to be people always want more.
> Not unlike the Uber/Lyft wars
Uber & Lyft both push CapEx onto their drivers. I think a more apt model might be AWS MySQL vs Oracle MySQL, or something similar. If the frontier providers stagnate, I fully expect people to switch to e.g. DeepSeek 6 for 10% the price.

Reply View | 2 replies
- babelfish 19 hours ago
  
  The thing is consumers don't care about OSS models. Any non-technical person just wants to "use AI", and think of ChatGPT for that.
  
  Reply View | 1 reply
  
  runako 19 hours ago
  
  Right, the model is a commodity to most users. So all things equal, a ChatGPT clone that costs (say) 70% less will steal share.
  Flipping it again: if the model is a commodity that lets one "use AI," why would anyone pay 2x or 3x as more to use ChatGPT?
  
  Reply View | 0 replies