Reports on Wednesday indicated that NVIDIA (NASDAQ: NVDA) has agreed to acquire Groq, a designer of high-performance artificial intelligence accelerator chips, in an all-cash deal valued at approximately $20 billion.
Groq most recently raised $750 million at a valuation of around $6.9 billion, underscoring the scale of the premium NVIDIA is paying for the AI chipmaker.
However, the Nvidia–Groq transaction does not appear to be a traditional acquisition, despite earlier reports from CNBC. Groq said it has entered into a non-exclusive inference technology licensing agreement with Nvidia, aimed at accelerating AI inference at global scale.
Under the terms of the agreement, Nvidia will license Groq’s inference technology, reflecting what Groq described as a shared focus on expanding access to high-performance, low-cost AI inference. The arrangement does not grant Nvidia exclusivity.
As part of the deal, Groq founder Jonathan Ross, President Sunny Madra and several members of the Groq team will join Nvidia to help advance and scale the licensed technology. Groq said it will continue to operate as an independent company, with Simon Edwards assuming the role of chief executive officer.
The structure suggests Nvidia is not acquiring Groq itself, its intellectual property, or exclusive rights to its technology, but is instead paying for licensed access to the technology alongside the addition of key personnel. Nvidia has yet to comment on the agreement.
Shares of the AI chipmaking giant were up 0.7% in thin premarket trading on Friday.
Groq was founded by former engineers who worked on Google’s Tensor Processing Unit (TPU), a chip developed to compete with Nvidia’s processors in artificial intelligence workloads.
Wall Street Weighs In on the Nvidia–Groq Deal
Bank of America analyst Vivek Arya said the deal “implies Nvidia’s recognition that while GPUs have dominated AI training, the rapid shift toward inference may require more specialized chips.”
Arya described Nvidia’s GPUs as general-purpose platforms, while Groq’s language processing units (LPUs) are positioned as more “specialized,” ASIC-like chips optimized for fast and highly predictable AI inference workloads.
He explained that LPUs rely on large amounts of on-chip SRAM to store model weights and working data, enabling extremely fast per-token access. However, this architecture comes with more limited scalability compared with Nvidia’s GPU platforms, which use high-bandwidth memory to maximize overall throughput.
Looking ahead, Arya envisions future Nvidia systems in which GPUs and LPUs coexist within the same rack, connected via NVLink.
“Longer term, we believe the potential Groq deal could be strategic, similar to Nvidia’s April 2020 Mellanox acquisition, which is now the foundation of Nvidia’s networking and AI-scaling moat,” Arya said.
Baird analyst Tristan Gerra said that while Nvidia’s GPUs are likely to “retain the majority of the AI processor market by 2030,” custom ASICs could be accretive to Nvidia’s total addressable market over time.
Similarly, Bernstein analyst Stacy Rasgon noted that “$20 billion seems expensive for a licensing deal,” particularly given the non-exclusive nature of the agreement. However, he added that the amount involved is “still pocket change for Nvidia,” given its roughly $61 billion cash balance, massive future free cash flow potential, and $4.6 trillion market capitalization — equivalent to about $0.82 per share.
“We’re inclined to give them the benefit of the doubt,” Rasgon wrote.
Soureces: Investing
THE ETERNAL SOVEREIGN 