Groq

Groq is an AI inference company that designs the LPU (Language Processing Unit) chip and runs Groq (GroqCloud), a fast token-as-a-service platform for running open models. Founded in 2016 by Jonathan Ross, who helped create Google’s TPU, Groq raised $750 million at a $6.9 billion valuation in September 2025 and struck a roughly $20 billion licensing deal with Nvidia in December 2025, after which Ross and other leaders moved to Nvidia and Simon Edwards became CEO.

What They Do

side of AI: running models fast and cheaply rather than training them. Its LPU is purpose-built for low-latency, and GroqCloud exposes that hardware as an API where developers run open models like Llama and DeepSeek at speeds conventional GPUs struggle to match. The pitch is speed and cost-per-token for inference at scale.

The December 2025 Nvidia licensing deal reshaped the company: it validated Groq’s LPU technology, brought in significant value, and moved founder Jonathan Ross and others to Nvidia, while Groq continues under new leadership repositioning as an inference neocloud. Groq is also developing next-generation chips with Samsung Foundry.

Current Flagship Products

• Groq: GroqCloud, the company’s fast inference platform (token-as-a-service) for running open models on its LPU hardware, used by developers who need low latency.

Strategic Position

Groq’s moat is differentiated inference hardware and a real speed advantage, now partly validated by the Nvidia licensing deal. Its challenges are formidable: competing on inference against Nvidia GPUs and other accelerators, executing a leadership transition after key founders departed, and scaling a capital-intensive chip-plus-cloud business. The Nvidia relationship is both validation and a complication for independence.

For AIpedia readers, Groq matters when inference speed and cost on open models are the priority, weighed against GPU-based providers and other inference clouds.