Voices
Dwarkesh Patel Interview: Jensen Huang Reveals NVIDIA Brought Groq Into CUDA Stack for Premium Token Market
In a 103-minute Dwarkesh Patel interview (April 15), Jensen Huang disclosed that NVIDIA has integrated Groq into the CUDA stack because the value of AI tokens has risen enough to create a premium market for faster response times, even at lower throughput. Huang also explained why Anthropic runs most compute on TPUs/Trainium: NVIDIA wasn't positioned to write the multi-billion dollar equity checks that Google and AWS made at Anthropic's founding. He predicted energy permitting — not silicon — as the binding constraint on AI buildout.
↳ Follow the thread