Openai had 2025 6billion in revenue and 20 billion costs on compute. So just to run the models to get 6billion they need to pay 20billion r&d and marketing etc get on top of that

No, my understanding is that they're bringing in revenue on token generation, but it's exceeded by the costs of token generation (running data centers, so, electricity and cooling). They definitely want to make a profit on token generation, but they're afraid that raising costs that high too quickly would drive customers to switch to other providers. So they've reduced the amount they're subsidizing token costs, but not switched over to making a profit.

I can't find a good citation for this, though, so it's possible I'm mistaken. They also have huge costs associated with buying new GPUs and building new datacenters, so they're operating at a massive loss either way, and it's a little hard to find articles which tease apart the two aspects.

In any case, operating at a massive loss for the first few years is practically standard operating procedure in silicon valley at this point, and sometimes it eventually leads to a profitable, even wildly profitable, business (e.g. Amazon). But it does require a steady stream of investors and a steadily increasing market valuation. That's...we'll have to see what happens on that front.