The semiconductor giant claims its new Vera CPUs and Rubin GPUs, plus other linked hardware, can run AI agents and large-scale AI models at a tenth of the cost of its last generation of systems. Nvidia also released new AI models and software for autonomous vehicles, which will be launched in Mercedes-Benz’s new CLA model later this year. (The Logic)

Talking point: As more businesses and consumers adopt AI, firms selling those tools and services should spend more on inference—running AI models to generate responses and do tasks—than on training them. Nvidia’s GPUs have dominated AI’s initial training phase, and now the firm is positioning itself for what comes next. That plan includes both its new homegrown chips, and its reported US$20-billion deal for Groq, which bought Nvidia most of the startup’s key talent and a non-exclusive license to its inference technology. Nvidia’s new autonomous vehicle systems, meanwhile, let automakers add more self-driving capabilities to their cars without having to develop them in-house.

Correction: This briefing has been updated to correct the reported value of Nvidia’s deal with Groq.