Google (GOOG) (GOOGL) is working to lessen Nvidia’s (NVDA) advantage with its CUDA software platform, with some help from Meta Platforms (META), Reuters reported.
The search giant is attempting to get its tensor processing units, or TPUs, run AI software framework PyTorch more smoothly, the news outlet added, citing people familiar with the matter. PyTorch has long been associated with Nvidia’s CUDA platform, since it was released in 2016. Meta is the most creator of PyTorch.
Google is also weighing whether to open-source parts of the software to increase uptake from customers, the sources said. The company is devoting more focus, resources and increased importance on the initiative, internally known as TorchTPU, the sources added.
Last month, it was reported that Google and Meta were in negotiations that would see Meta use Google’s TPUs as the social networking giant looks to decrease its reliance on Nvidia and its GPUs.
Google, Meta and Nvidia did not immediately respond to a request for comment from Seeking Alpha.
Google and Nvidia shares tumbled around 3% each in early trading on Wednesday, while Meta shares were modestly lower.