Trying to do more with less would be one strategy - coming up with more efficient inference that takes less compute and energy. This seems to be what's driving companies in China like Deepseek. Whether one could be profitable with this strategy is another story, probably the best to hope for would be getting bought by one of the FAANGs (if you were able to show good results).
It's also possible that the first movers are doing all of the expensive and risky research and utterly failing to build a moat for themselves. It doesn't look like there's any particularly lock-in for any category of AI model, and open source is following closely behind.
Compute will get cheaper and open models will proliferate.