The big picture: Currently, Nvidia's chips power much of the AI training phase, but inference is a bottleneck that Nvidia doesn't fully control yet.
- Groq's chips are purpose-built for infer… [+2377 chars]


