NVIDIA has reached an agreement to acquire the assets of AI chip startup Groq for approximately $20 billion. This marks NVIDIA's largest move to date and demonstrates the company's strategy of bringing potential competitors into its fold before they pose a threat to market dominance.
This latest licensing agreement from chip giant NVIDIA repeats a similar deal that occurred just three months ago. This situation strengthens the narrative that decentralized artificial intelligence infrastructure could be the only real alternative to NVIDIA's increasing dominance.
Threefold increase in three months connected to Trump Jr.
The deal was finalized just three months after Groq raised $750 million with a valuation of $6.9 billion. Giants like BlackRock, Samsung, Cisco, and 1789 Capital, which includes Donald Trump Jr., also participated in this round. NVIDIA is acquiring all of Groq's assets except for its cloud computing business. However, Groq describes the transaction as a 'non-exclusive licensing agreement.'
Groq's CEO and former Google engineer Jonathan Ross, one of the creators of Google's Tensor Processing Unit, will join NVIDIA along with company president Sunny Madra and other senior executives. The company will continue its independent operations under the leadership of its new CEO, CFO Simon Edwards.
Repeated Game Plan
The Groq agreement stands out as a continuation of a strategy that NVIDIA implemented just three months ago. In September, the company paid over $900 million to transfer Enfabrica's CEO and employees, licensing the startup's technology. Since both transactions progressed through licensing structures rather than direct acquisitions, the aim is to avoid the competitive authority barriers that blocked NVIDIA's $40 billion Arm Holdings bid in 2022.
The Kobeissi Letter summarized NVIDIA's approach as follows: 'We will buy you before you have the power to compete with us.'
Technological Advantage and Competitive Pressure
Groq's Language Processing Unit uses on-chip SRAM instead of external DRAM. The company claims that this architecture increases energy efficiency by 10 times. While such a structure offers significant advantages in real-time inference, it restricts model sizes: NVIDIA will now evaluate this trade within its own vast ecosystem.
The timing is striking. Google recently unveiled its seventh-generation TPU, code-named Ironwood, and managed to place the Gemini 3 model at the top of benchmark lists by training it entirely on TPUs. NVIDIA responded through X: 'We are very happy with Google's success... NVIDIA is a generation ahead of the industry— the only platform capable of running every AI model.' If established giants in the industry start issuing such reassuring messages, know that competitive pressure is increasing.
Potential Outcomes for Decentralized Artificial Intelligence
Although this agreement does not have a direct effect on the cryptocurrency market, it strengthens the narrative that supports decentralized artificial intelligence computing projects. Platforms like io.net position themselves as alternatives to centralized infrastructure.
Jack Collier, the Head of Growth at io.net, made the following statement to BeInCrypto: 'Whether you have a data center or your own laptop; you can connect the GPU power you have to this network and contribute, receiving fair compensation through tokenomics for it.' The platform claims that its corporate clients, such as Leonardo.ai and UC Berkeley, have achieved significant cost advantages.
However, the gap between narrative and reality remains substantial. NVIDIA's acquisition of Groq’s low-latency technology further enhances its technical superiority, making it difficult for alternatives to offer competitive performance.
This process raises questions about the development of independent artificial intelligence chips. Another NVIDIA competitor preparing for an IPO, Cerebras Systems, may face similar pressure.
