Mistral AI Aims to Produce Its Own Chips to Cut Costs

Mistral AI Plans to Produce Its Own Chips

In the world of artificial intelligence, many companies are looking to produce the chips they use themselves. Among these companies, Mistral AI, a major player in AI in France, stands out with similar ambitions. Founded by Arthur Mensch, Mistral AI is following in the footsteps of industry giants by exploring the possibility of designing its own chips for AI.

A Project Inspired by American Giants

American AI companies, such as Anthropic, have already expressed their intention to produce their own chips, with potential investments reaching $30 billion. In France, Mistral AI is also considering this path. In an interview with CNBC, Arthur Mensch confirmed the company's interest in this initiative, while specifying that it is a long-term project. “Of course, it’s interesting,” he stated, emphasizing that in-house chip manufacturing could be beneficial for the company.

Reducing Token Deployment Costs

Arthur Mensch explained that in-house chip manufacturing could significantly reduce the cost of deploying tokens, a fundamental unit in AI language processing. Currently, Mistral AI relies on Nvidia for its chips, but the company is also testing other solutions. Creating in-house chips would not only provide independence from Nvidia but also help optimize costs related to token processing. According to Mensch, “We may end up making our own chips; I think that will eventually happen, but for now, we rely on Nvidia, which is an excellent partner for us, and we are testing a few solutions here and there.”

The Importance of Tokens in AI

As a reminder, a token is the basic unit of text processed in AI: it can be a word, a piece of a word, or a punctuation mark. AIs read and produce language token by token, a measure that also serves to calculate their cost and processing power. The ability to produce tailored chips could therefore have a significant impact on the efficiency and operational costs of Mistral AI.