OpenAI requires more chips for training and solving artificial intelligence (AI) models. The data centers it operates are packed with its partners, including Oracle, SoftBank and CoreWave Server with nvidia chip. However, the facilities it is currently building in Texas, New Mexico, Wisconsin, Ohio or Michigan will also rely on OpenAI and Google’s integrated circuits.
In late August, the AI laboratory led by Sam Altman introduced Jalapeno, a chip optimized for inference work. OpenAI already having its own semiconductors for AI doesn’t mean it will stop buying its GPUs from Nvidia. At least right now he wasn’t sure that was his intention. Also, as I mentioned, Jalapeño is an inference chip, so Nvidia GPUs are still important in the training process of their AI models.
What is clear is that OpenAI wants to reduce dependence on companies led by Jensen Huang. Nvidia GPUs are very expensive and their availability fluctuates widely due to problems that TSMC, the company that makes them, has in responding to demand. Additionally, controlling the logic design phase of these chips will allow you to train your own models and optimize them to address your evolving inference needs.
Google, from competitor to partner
According to The Information, OpenAI is in talks with Google to buy its processors to expand its computing resources. If the deal eventually comes to fruition, Sam Altman’s lab will use Nvidia’s integrated circuits, Google’s and its own chips. This bet is very strong, and, in a way, surprising in a situation where it is not yet clear that OpenAI or other AI laboratories will be able to resolve the tension between the need to be profitable and the need to continue investing fortunes in infrastructure.
OpenAI and Google are direct competitors in the AI model market
However, there is something else we cannot ignore: OpenAI and Google are direct competitors in the AI model market. A priori we can assume that Google will not be interested in selling its TPU (Tensor Processing Unit or Tensor Processing Unit) is one of its biggest competitors, but if we analyze the current situation carefully, we will realize that It doesn’t have to be like that. OpenAI leases the chip to Google from 2025, so this potential purchase agreement is another step towards stabilizing the commercial relationship between the two companies.
For Google, the sale of its TPU processors could be a very important source of revenue. In fact, OpenAI isn’t the only AI lab trying to reduce reliance on Nvidia GPUs. Anthropologists are traveling this same path. And Chinese AI laboratories are also unable to access the most powerful Nvidia chips as a result of export restrictions imposed by the US government. Either way, if the purchase is ultimately successful, OpenAI will win. And Google won. The only one that doesn’t win is Nvidia.
Picture | TSMC
More information Information
In Xataka | These two problems have puzzled mathematicians for decades. A genius solved them with the stroke of a pen
