Traning a LLM takes a lot of computing power. NVIDIA provides the chips. The US has put a limit on these chips that are being sold to China. So now China does not have the NVIDIA chip resources the US has. Here is the thing: they just created a better large language model training that requires less computing power and bites OpenAI in the ass, in several benchmarks.
Open source
But the real middle finger is the decision to just open source everything. This puts a lot of pressure on the value proposition of the OpenAI models. Bad for OpenAI maybe. Splendid for us! We can freely use a better LLM than OpenAI and even build our own.