The amount of computation necessary for AI is skyrocketing. The demand for Nvidia GPUs is skyrocketing because models are increasing by a factor of 10, an order of magnitude every single year.

Obi-wan's introduction was an inflection point for AI - instead of one-shot answers, inference is now a thinking process requiring reinforcement learning and significant computation in post-training.

Test-time scaling causes the number of tokens generated to increase by 5x every year as longer thinking produces better answers.

The AI race is intense with everyone trying to reach the next frontier, causing last-generation AI token costs to decline by 10x annually.

Nvidia must advance state-of-the-art computation every year without falling behind, and Vera Rubin GPU is now in full production.