Not to mention they will need to research how to make their models faster and cheaper to run in order to fit some margin within what people are actually willing to pay.
It's more like once you figure out how to make a really good lamp then producing lots of lamps will be profitable. But the lamps are currently suboptimal so we'll be in the red until that time.
It's more like you have a business making engines, each generation of engine has eventually turned out to be profitable over its lifespan, but each generation has an exponentially increasing R&D cost and your customers will switch from the old engines to a competitor if they don't like the newest generation.
You're stuck racing against your competitors with the distinct possibility that your R&D costs will outgrow the market demand, and you can't stop because otherwise your customers will stop investing in your dead end tech and switch.
OpenAI won't be able to cut R&D spend and collect rent on their existing models as long as the Chinese models keep up the pace of being ~6 months behind them for a fraction of the price.
And then someone will come up with lamp pro max and you’ll be out of business. You realize why R&D exists in tech companies even though it’s a cost center right?
Watch them flare out like a star… but there is lots of questions re the the return on RnD. Is it worth spending another order of magnitude for only marginal frontier gains?
People keep overlooking the fact that costs for these providers scale along with customer acquisition. Most startups don't have that linear expense. Also, training costs are accelerating to get new models out faster. One doesn't simply "get rid of R&D" costs as a comment upstream mentioned. I can't actually imagine R&D goes down anytime soon unless you're willing to play third fiddle.
Unless these frontier providers feel some type of squeeze or constraint the Chinese are well positioned to leave the US bag holders of an NVidia bound system. And if anyone has to wonder how one provider for a critical piece of infrastructure will go, well...
Even if they keep the R&D costs, more efficient inference and 0 Marketing spend also gets you there. Inference is honestly super inefficient at this point, we can do far better than GPUs, push utilisation up, build more efficient datacentres.
R&D costs are hurting profit side and while you can cut that one just becomes irrelevant overnight in this space if you do, hence the problem.