MAD / MAC assumes everyone has the same leverage on each other- that they are all as heavily invested in capital.
Where it falls apart is when you get companies that take open models, rent compute and sell tokens at a profit, without heavy investment into research or training. They spend way less, and if they can keep distilling the frontier models, they can produce reasonable products without the massive losses of subscription plans. In your example, DeepSeek et al aren't nearly so vulnerable. There's nothing mutually assured.
It is much more of a classic race to the bottom. Every new frontier model shifts the goal posts a bit, but eventually the house of cards will crumble - or at least, that's what the market seems to be thinking will happen.
Subscription plans are not massive losses. What are you on about?
They are simply less profitable than API pricing. They are on par with what sellers of open model tokens sell for.
If you price out Subscription limits against API pricing, subscriptions are 20x-100x less depending on model and so forth.
Yes, that's accurate.
Are you trying to say "the Anthropic API surely cannot be highly marked up! Therefore subscriptions surely are selling at a loss!"? If not, I'm not sure how your comment is particularly relevant to the discussion?
I mean, it's worth pointing out because either API pricing is crazy marked up and thus subscriptions are still profitable if they are used at full amount or subscriptions are subsidized and API pricing is more true cost. Sure, it could be second one but due to average usage, it's profitable overall.
Right. Either way, there's nothing at all to support that subscription plans are "massive losses" except the original commenter's imagination. A subscription/API price delta does not a massive loss make.
Agreed, but there is also no information to think subscriptions are profitable either. S-1 doesn't really explain it.