Why Groq’s IP Matters to Nvidia; AI Experts’ Favorite Books of the Year

Before we get to today’s column, we need to explain something about last week’s big news: Nvidia’s agreement to pay an astounding $20 billion to license technology from AI chip startup Groq and hire its leaders. 

The reasons for Nvidia’s interest in Groq are getting a little clearer. Groq’s chips have the potential to help AI developers when they’re training so-called reasoning models, or models that can spend more computational resources on more difficult problems. These models may not be popular among consumers, but they’re super handy for tasks such as AI coding, a key growth area in 2026 and beyond that Nvidia CEO Jensen Huang has also expressed excitement about.

There’s a step in the training process for reasoning models that looks a lot like inference (the process of running AI models), called a Monte Carlo Tree Search. Since Groq’s chips are theoretically more efficient at handling inference, Nvidia could harness the Groq team’s know-how to improve how its chips handle this step during the training process for reasoning models.