They’ll be profitable when the training stops/slows down. But how much revenue they can generate is another question. This might be the reason behind the news of US gov thinking about banning Deepsek and other Chinese models. Open weights models kill the API billing bussiness model. Margins in that scenario probably wouldn’t cover the massive investment.
That’s the problem, you can’t stop training your model. New and useful information is created every day. This is simply not an economically viable product.
They’ll be profitable when the training stops/slows down. But how much revenue they can generate is another question. This might be the reason behind the news of US gov thinking about banning Deepsek and other Chinese models. Open weights models kill the API billing bussiness model. Margins in that scenario probably wouldn’t cover the massive investment.
That’s the problem, you can’t stop training your model. New and useful information is created every day. This is simply not an economically viable product.
How are they supposed to “slow training down” without falling behind their competitors?
Competitors have the same problem. We’ll see what happens once subsidies run out.
So the choice is between unprofitability and models being so out of date they become useless? Great.