According to a Reuters report, the initiative aims to improve compatibility between Google’s Tensor Processing Units and PyTorch, the most widely used software framework for building AI models.
- Google aims to remove a significant obstacle that has limited the broader adoption of its chips by outside developers.
- Internally dubbed “TorchTPU,” the project aims to make TPUs easier to use for companies that have already built AI systems with PyTorch.
- Google is also considering open-sourcing parts of TorchTPU to speed adoption and attract developers.
Google parent Alphabet Inc. (GOOGL, GOOG) is reportedly stepping up efforts to strengthen its in-house artificial intelligence chips as it looks to loosen Nvidia’s (NVDA) grip on the fast-growing AI market.
According to a Reuters report, the initiative aims to improve compatibility between Google’s Tensor Processing Units, or TPUs, and PyTorch, the most widely used software framework for building AI models.
Closing The Software Gap
Google aims to remove a major obstacle that has limited broader adoption of its chips by outside developers, who rely on PyTorch rather than Google’s preferred internal tools, stated the report.
Internally dubbed “TorchTPU,” the project focuses on making TPUs easier to use for companies that have already built AI systems using PyTorch. While Google’s chips have gained traction inside the company and within Google Cloud, many customers have struggled with the added engineering work required to shift away from Nvidia-friendly software environments.
Alphabet Inc.’s Class A stock traded over 2% lower on Wednesday afternoon. On Stocktwits, retail sentiment around the stock remained in ‘bearish’ territory amid ‘low’ message volume levels.
Strategic Push Against Nvidia
Google’s renewed emphasis on PyTorch support reflects growing pressure to show returns on its massive AI investments. TPUs have become an increasingly important part of Google Cloud’s revenue strategy as enterprise customers search for alternatives to Nvidia’s often scarce offerings.
Sources cited by the report say Google is also considering open-sourcing parts of TorchTPU to speed adoption and attract developers. PyTorch is a tool for creating deep learning models. It is written in Python, which makes it easy for many developers to learn and work with.
PyTorch, developed by Meta Platforms (META), stands out because it works well with GPUs and uses automatic differentiation to calculate changes during training. It also allows models to be changed while they are running.
GOOGL stock has gained over 58% year-to-date.
For updates and corrections, email newsroom[at]stocktwits[dot]com.<
