TECH NEWS – Nvidia’s CEO made a very clever move that could prove surprisingly powerful during the Christmas season.
Jensen Huang, the CEO of Nvidia, probably gave his chip team an unexpected Christmas present when he reached an agreement with Groq, a company that manufactures specialized AI hardware. This could be Nvidia’s gateway to dominating inference-class workloads. According to CNBC, Nvidia bought Groq Inc. in a $20 billion deal, Huang’s biggest acquisition to date. This caused an uproar in the industry.
However, Groq later issued an official statement announcing a non-exclusive licensing agreement with Nvidia, granting the AI giant access to Groq’s inference technology. In an internal email, Huang said that Groq’s low-latency processors would be integrated into Nvidia’s AI factory architecture. They are hiring talent and licensing Groq’s intellectual property, but not acquiring the company.
This is a classic reverse acquisition move, similar to Microsoft’s Inflection deal in 2024. By structuring the agreement as a non-exclusive license, Nvidia avoids FTC scrutiny and remains outside the scope of the Hart-Scott-Rodino Act.
The most interesting part of the deal is the hardware. Groq’s Language Processing Units rely on deterministic execution and on-die SRAM, delivering extremely low latency and high throughput. This makes them particularly well suited for inference workloads, especially decoding.
While it remains unclear how LPUs will be integrated, one possibility is rack-scale inference systems where GPUs handle prefill and long-context tasks, while LPUs manage decoding. This could turn LPUs from a niche solution into an industry standard.
