Wall Street analysts are broadly positive on the latest deal between Nvidia and AI inference chip company Groq. Cantor, for example, believes the transaction carries both “offensive” and “defensive” strategic value. The firm reiterates Nvidia as its top pick and maintains an Overweight rating.

Bank of America points out that while Nvidia’s high-priced acquisition of Groq was somewhat unexpected, it effectively turns a potential ASIC technology threat into part of Nvidia’s own competitive moat. From a long-term perspective, this strategic move is highly valuable, so BofA maintains its Buy rating on the stock.

Nvidia has already been working with Groq on certain inference acceleration tasks. Nvidia sees a real opportunity here and believes that bringing Groq in-house as an internal team, rather than keeping it as an external partner, will be more beneficial.

Nvidia is already dominant in AI training and time-sensitive inference and integrating Groq’s low-latency, high-efficiency inference technology into Nvidia’s full-stack systems will help it expand its share in the inference market—especially in the next stage of AI infrastructure build-out, covering real-time workloads such as robotics and autonomous driving.

Nvidia is well aware that although GPUs reign supreme in AI training, the explosive growth in inference demand may call for more specialized chips.

Introducing different types of hardware will add complexity to future GPU/LPU roadmaps and pricing, but Nvidia can leverage its strong balance sheet and platform status to offer customers more options, while conceptually neutralizing the competitive threat from Groq and other dedicated ASIC chips.

In the long run, if this potential Groq deal is ultimately completed, its strategic significance could be comparable to Nvidia’s 2020 acquisition of Mellanox, which has since become a foundational part of its AI expansion moat.

Market Commentary:

In February 2024, Groq launched a brand-new AI chip that it claimed delivered the world’s fastest inference performance. Running large models on Groq’s chips can be up to ten times faster—or even more—than on Nvidia’s GPUs. These chips are mainly used to speed up inference tasks for large language models and have been viewed by the market as one of the potential alternatives to Nvidia GPUs.