Jonathan Ross

Ross began Google's TPU project as a "20% time" side effort, designing the first-generation Tensor Processing Unit that gave Google its custom-silicon advantage in AI. That experience convinced him that inference — serving models — needed a fundamentally different chip architecture from training-oriented GPUs.

He founded Groq in 2016 to build the Language Processing Unit, a deterministic, low-latency design optimized for the sequential token generation that dominates LLM serving. Groq's cloud posts some of the fastest tokens-per-second figures in the industry for popular open-weight models.

Under Ross, Groq has raised substantial capital at a multi-billion-dollar valuation and pursued large sovereign-AI deployments, positioning itself as a leading challenger to NVIDIA on the argument that inference economics favor specialized silicon.