> $20/month was just too big a loss to sustain.

Is there any marginal cost associated with a new subscriber?

I have always heard inference is cheap and the cost was in training, so I assumed any subscriber was making them money, just not enough to cover their insane fixed costs.

But I am just guessing.

There are two extremes, the "approx 0 marginal cost" camp and the "anthropic losing $5000/mo on every Max subscriber" camp.

I strongly suspect both are wrong.