In addition to GPUs that handle the lion’s share of AI training, Nvidia wants to introduce a chip for running AI workloads on a daily basis. Thanks to the expertise acquired from startup Groq, the company hopes to provide AI players such as OpenAI with energy-efficient processors to run their AI services.
The chip could be unveiled next month during Nvidia GTC. According to the Wall Street Journal, OpenAI has been given early access to the new processor. On Friday evening, the news outlet reported that Nvidia wants to use this to thwart emerging competition. In the field of AI training, it seems unbeatable, although Google with its Tensor Processing Unit and AMD with its GPUs are real alternatives.
When it comes to inferencing, there are numerous options. AWS and Google Cloud, for example, have chips for this in the public cloud, while many startups are trying to provide a cheaper and more efficient alternative for inferencing. It is therefore no surprise that Nvidia itself is now developing a dedicated inferencing platform. In December, the company signed a $20 billion licensing deal with Groq, also hiring founder Jonathan Ross and president Sunny Madra. Groq’s Language Processing Units (LPUs) are built on a completely new architecture that performs inferencing with significantly less energy consumption.
The company has not yet announced exactly how Nvidia will integrate the technology. The GTC 2026 conference starts on March 16 in San Jose, so we expect more clarity then. It is noteworthy that OpenAI is an early customer, as it had been looking for faster alternatives to Nvidia’s GPUs for some time due to dissatisfaction with the inferencing speed for specific tasks, including software development. Last month, it signed a deal with Cerebras for inferencing chips. Those two deals are separate.
OpenAI as an early adopter
At the same time, OpenAI received $30 billion from Nvidia last week as part of a mega-investment totaling $110 billion. So it seems that peace has been restored between the two parties. OpenAI would like to use the new inferencing chip for Codex, its own programming tool with which it competes with Anthropic’s Claude Code. Coding is one of the most profitable use cases for generative AI, and an area in which OpenAI currently ranks second. Claude Code is, in fact, the standard for programmers working with AI, apart from solutions that companies have developed or purchased internally.