Ddos 
The AI startup Groq, best known for its ultra-fast inference chips known as LPUs, has announced that it has signed a “non-exclusive licensing agreement” with NVIDIA covering inference technologies.
Even more striking is the personnel shift that accompanies the deal. Groq’s driving force—founder Jonathan Ross, a former core developer of Google’s TPU—will lead part of Groq’s senior technical team directly into NVIDIA. While this is not a conventional acquisition, the effective transfer of both talent and technology is poised to have a profound impact on the AI inference landscape.
According to the announcement, the agreement is explicitly non-exclusive, meaning Groq will continue to operate as an independent company. The leadership changes, however, are substantial. Key figures—including CEO Jonathan Ross and COO and President Sunny Madra—will move to NVIDIA, where they will oversee the development and deployment of the licensed technologies. Former CFO Simon Edwards will step in as Groq’s new CEO, ensuring continuity of operations, while the company’s GroqCloud service remains unaffected.
Founded in 2016 and headquartered in Mountain View, California, Groq was established by Jonathan Ross, one of the original members of Google’s Tensor Processing Unit (TPU) team.
Unlike NVIDIA’s GPUs, which rely on massively parallel computation for graphics and AI workloads, Groq developed a specialized processor called the LPU (Language Processing Unit). Built on a deterministic architecture, the LPU is optimized specifically for the inference phase of large language models. Its token generation speeds have frequently drawn industry attention, with Ross once remarking, “Everyone else relies on GPUs; our edge lies in custom silicon.” In essence, while this deal is framed as a licensing arrangement, it more closely resembles a carefully targeted acqui-hire.
NVIDIA already enjoys near-total dominance in AI training, but in inference, Groq’s LPU architecture has demonstrated superior efficiency and lower latency than traditional GPUs. Through this partnership, NVIDIA not only secures access to Groq’s core inference technology but also brings Jonathan Ross—the person who understands it best—into its own ranks.
The strategy delivers two clear advantages:
- Avoiding antitrust scrutiny: A full acquisition of Groq would almost certainly have triggered resistance from regulators worldwide. By combining licensing with talent recruitment, NVIDIA gains the technology and expertise it wants while preserving Groq’s nominal independence—a notably shrewd maneuver.
- Strengthening its inference position: As AI applications mature, demand for inference capacity is expected to outpace training. This move allows NVIDIA to rapidly reinforce its technological moat in inference, limiting the growth of Groq or other ASIC challengers.
As for Groq’s future without its founder, the company insists that business will continue as usual. Yet with its visionary leader gone, whether this former unicorn can sustain its innovative momentum remains an open—and pressing—question.
Related Posts:
- Jonathan Blow thinks C++ is a terrible language
- Google open Tensor Processing Unit AI Chip to other companies
- AI’s New Attack Vector: How Real-Time Bots Are Straining Websites
- Canadian Hacker Indicted for $65 Million DeFi Exploit
- Critical Triton Flaws (CVSS 9.8) Expose AI Servers to Remote Takeover – Patch Now!
Donate