NVIDIA's Secret Chip Fuses GPU and Groq for OpenAINVIDIA will unveil a new inference processor built on Groq's LPU architecture at GTC 2026, with OpenAI as its first major customer allocating 3 GW of dedicated capacity.