Jonathan Ross

Jonathan Ross is the founder of Groq, the inference-chip company behind the Language Processing Unit architecture. He previously helped design core pieces of Google's first Tensor Processing Unit, then spent years as Groq's CEO building silicon and a cloud around deterministic, low-latency inference before joining Nvidia in late 2025 while Groq continued as an independent company.

From Google's TPU to a new inference chip

At Google, Ross worked on what became the first-generation TPU as a side project that turned into production silicon for neural-network serving. He later spent time on Google X's rapid-evaluation team, then left to found Groq around a different bet: a compiler-first, deterministic architecture aimed at inference rather than general GPU training.

Building Groq around speed and determinism

As CEO he pushed Groq's LPU story in public: high tokens-per-second inference, predictable latency, and a cloud (GroqCloud) that let developers feel that speed without owning racks. The company raised large growth rounds on the claim that language models would be gated by inference cost and latency as much as by training clusters.

Handing off the CEO seat

In December 2025 Groq announced a licensing and talent arrangement with Nvidia; Ross and others joined Nvidia to advance the licensed technology while GroqCloud continued operating under new CEO leadership. He remains the founder most closely identified with Groq's chip thesis.