Groq
AI Inference Chips
- Valuation
- $3.5B(Aug 2026), down from a $6.9B peak in Sep 2025
- Total Funding
- $640M
- Profit Status
- Pre-profit
- Employees
- ~300
About Groq
Groq is a chip startup that has developed the Language Processing Unit (LPU), custom silicon designed specifically for LLM inference. Their hardware delivers inference speeds 10-25x faster than GPU-based alternatives.
Company History
Founded in 2016 by Jonathan Ross, a former Google TPU architect. Burst into public awareness in early 2024 with API demonstrating 500+ tokens/second for Llama and Mixtral. In December 2025 Groq announced a non-exclusive licensing agreement with Nvidia; Jonathan Ross and other team members joined Nvidia, while Groq continues as an independent company and GroqCloud continues. Press reports put the arrangement near $20B, a figure Nvidia has not confirmed.
AI Models & Benchmarks
1 modelsLPU (Language Processing Unit)
Custom ASIC for deterministic, low-latency LLM inference. 750GB/s memory bandwidth.
Benchmarks
- Llama 3.1 70B speed
- 250 tok/s (vs ~40 tok/s GPU)
- Mixtral 8x7B speed
- 575 tok/s
- Latency
- <0.5s first token
Funding Rounds
2 rounds| Date | Round | Amount | Investors |
|---|---|---|---|
| 2021 | Series C | $300M | Tiger Global, D1 Capital |
| Aug 2024 | Series D | $640M | BlackRock, Neuberger Berman |
Products & Services
3- LPU Inference Engine
- GroqCloud
- Groq API
People & ecosystem
Key People
- Jonathan Ross (Founder; departed to Nvidia Dec 2025)
Key Investors
- BlackRock
- Tiger Global
- BatCapital
- Neuberger Berman
Competitive Landscape
- NVIDIA
- Cerebras
- SambaNova
- Tenstorrent