Openai could start producing your own fleas next year

WhileElon Musk has made a bought offer at $ 100 billion for Openai (which should not be accepted), the firm has its own plans. Thus, according to an exclusive report by Reuters, Sam Altman's company is about to finalize the design of its very first personalized chip dedicated to the training of artificial intelligence, thus entering into direct competition with the king of the Category: NVIDIA.

Homemade fleas to stop dependence in Nvidia

According to Reutersthe manufacturing process would currently be in the “mid-tape-out” stage, which actually corresponds to the phase preceding their massive production. In addition, this chip will be manufactured by Taiwai Semi Conductor Company (or TSMC) using 3NM technology. For the occasion, the new chip should adopt a systolic network architecture as well as a large -band memory, similar to the latest AI AI ACS from Nvidia: it is the architecture favored to benefit from good efficiency and Good performance in the treatment of calculations necessary for the training of artificial intelligence models.

For mass production, however, it will be necessary to show a little patience: Reuters speaks in particular of the year 2026. Besides, despite this investment in time and money, there is no guarantee that this first chip turns out Completely functional: for the time being, Openai has only a small team dedicated to this project. Only 40 people are indeed part of this flea design team. When the latter are ready, they should initially be deployed in the Openai infrastructure, benefiting from a limited role.

The fact that Optai decides to create her own chips is not in itself surprising: the firm has envisaged a solution of this kind for a good time already, in order to reduce the dependence of Openai to the GPU of Nvidia. There is no doubt the investment of $ 500 billion in the United States government in the Stargate project should help the business hold this goal.