AI Stack
Capabilities and model layers highlighted by the company.
- LPU
- GroqCloud
Groq
PrivateAI inference chip company founded in 2016 by former Google TPU engineers, pioneering the Language Processing Unit (LPU) architecture for ultra fast AI inference. In December 2025, Nvidia agreed to acquire Groq's assets for approximately $20 billion.
Founded 2016$750M raised
Introduction
Groq designs Language Processing Units (LPUs), custom AI chips optimized for ultra-fast, low-latency inference. Founded in 2016 by Jonathan Ross, inventor of Google's TPU. In December 2025, NVIDIA announced a deal reportedly worth ~$20 billion to license Groq's inference technology.
Technology
Platform maturity, autonomy stack, and flagship-system specifications in one view.
Investors
December 2025Company MilestoneCorporate
NVIDIA licenses Groq inference technologyIn December 2025, NVIDIA announced a ~$20 billion deal to license Groq's inference technology, validating that inference-optimized hardware represents an essential category in AI computing.
No news articles mentioning Groq yet.
Valuation History
Follow Groq's valuation growth through its funding rounds.
Management Team
Company Story
Groq designs Language Processing Units (LPUs), custom AI chips optimized for ultra-fast, low-latency inference. Founded in 2016 by Jonathan Ross, inventor of Google's TPU. In December 2025, NVIDIA announced a deal reportedly worth ~$20 billion to license Groq's inference technology.
- 2016
Jonathan Ross founds Groq around inference-first silicon
Groq was founded in 2016 by Jonathan Ross, who had created Google's TPU as a side project. The company designed a novel statically-scheduled architecture delivering predictable, ultra-low latency performance.
LPUTPU lineage - 2021 - September 2025
Three late-stage rounds lift Groq to $6.9B
Funding included $300 million Series C in 2021, $640 million Series D in August 2024 at $2.8 billion, and $750 million Series E in September 2025 at $6.9 billion. Total raised: ~$1.75 billion.
Series CSeries DSeries E - 2024
Groq breaks out on inference speed
Groq gained attention in early 2024 with cloud API speeds of 500+ tokens/second, far outpacing GPU alternatives. The speed difference went viral among AI developers.
Inference500+ tokens/second - December 2025
NVIDIA licenses Groq inference technology
In December 2025, NVIDIA announced a ~$20 billion deal to license Groq's inference technology, validating that inference-optimized hardware represents an essential category in AI computing.
NVIDIALicensing