Everyone knows TSMC can't make the newest AI chips that have to multiply a 1,000,000,000x1,000,000,000 matrix every clock cycle. What if you're trying to compete with Nvidia?