When you think of chips for artificial intelligence, the name that automatically comes to mind is, naturally, Nvidia. The company has made huge profits selling semiconductors used in popular AI applications like large-scale language models, and stock investors have been handsomely rewarded for it. But, of course, Nvidia’s GPUs can be used for more than AI. They’re also used in video games, graphics, cryptocurrency mining, and more. But a new startup called MatX is aiming to develop the ultimate chip specifically for LLMs. Co-founders Rainer Pope and Mike Gunter spent years working at Alphabet, which had its own in-house semiconductor business, but now they’ve gone independent and started a new chip company from the ground up. We spoke to them about how they approach their work, what it takes to actually design and build a chip, and what it takes to get customers to switch from the industry leader. The transcript has been lightly edited for clarity.