How many GPUs to train GPT 4?

Inquiring minds want to know: Just how many graphical processing units (GPUs) are required to train the highly anticipated GPT 4, the next generation of OpenAI's groundbreaking language model? With each iteration bringing increased complexity and capabilities, the computational demands for such an endeavor are surely immense. Are we looking at a few hundred, or perhaps thousands, of GPUs to reach the level of performance expected from GPT 4? The answer may reveal the true scale of the technological feat that lies ahead.