<div id="readability-page-1"><div><p><span>Alphabet, Google's parent company, is teaming up with Meta to develop &#34;TorchTPU,&#34; a project aimed at making Google's TPU chips fully compatible with the PyTorch framework, widely used in artificial intelligence development. This strategic initiative seeks to reduce the sector's reliance on Nvidia, which currently dominates the AI processor market thanks to its CUDA software ecosystem. Following the announcement, Nvidia's shares fell 2% during trading. The project could profoundly reshape the technological balance between hardware and software across the industry.</span></p><p><span>Until now, adoption of TPUs by external customers has been limited due to their weak compatibility with PyTorch, forcing developers to heavily adapt their code. TorchTPU aims to remove this barrier by offering native integration, which would enable Google to make its chips more attractive and accessible. The group is also considering partially open-sourcing the code to accelerate adoption by the community.</span></p><p><span>The project is being carried out in close collaboration with Meta, which oversees the development of PyTorch. Discussions are reportedly underway for Meta to gain broader access to Google's TPUs via managed services, which would help reduce its dependence on Nvidia GPUs, both costly and ubiquitous. This push is part of a broader Google strategy to democratize the use of its TPUs, notably through their direct deployment at customers and through internal appointments that strengthen the leadership of its AI division.</span></p></div></div>

谷歌-A

谷歌-C

<p>谷歌的母公司 Alphabet 正在與 Meta 合作開發 TorchTPU，旨在使谷歌的 TPU 芯片與 PyTorch 框架兼容。該倡議旨在減少對 Nvidia 人工智能處理器的依賴。公告發布後，Nvidia 的股價下跌了 2%。該項目旨在通過提供原生集成來增強 TPU 的採用，並可能部分開源代碼。Meta 可能會獲得更廣泛的谷歌 TPU 訪問權限，從而減少對 Nvidia GPU 的依賴。這是谷歌推動 TPU 使用普及和加強其人工智能部門戰略的一部分</p>

Google and Meta join forces to weaken Nvidia's dominance in AI

- Alphabet 與 Meta 合作開發 "TorchTPU"，旨在提升 TPU 與 PyTorch 的兼容性。  
- 該項目旨在減少對 Nvidia 的依賴，可能影響整個行業的技術平衡。  
- 項目還考慮部分開源代碼，以促進社區採用，併為 Meta 提供更多 TPUs 的訪問權限。

谷歌和 Meta 攜手合作，試圖削弱 Nvidia 在人工智能領域的主導地位