TPU、TensorFlow—谷歌云“增强”AI 芯片团队，与微软云竞争 AI 云份额

原文题目： 

Google’s Cloud Unit Gains Key AI Chip Team to Compete With Microsoft 

文章来源： 

Information 4 月 23 日文章 

翻译：Notion AI 

编辑：宇婷 

  Google’s org chart  ‍‍‍‍‍‍ 

Google has moved the engineering team responsible for making artificial intelligence chips into Google Cloud, a spokesperson confirmed, in a step that could make the cloud unit more competitive with its bigger rivals, Microsoft and Amazon Web Services, in selling AI-powered software to businesses. 

谷歌已将负责制造人工智能芯片的工程团队转移到了 Google Cloud 中，一位发言人证实了这一步骤，这可能会使云部门在向企业销售 AI 驱动软件方面更具竞争力，从而与其更大的竞争对手微软和亚马逊网络服务相抗衡。 

The move is the latest sign of how Google is scrambling to respond to Microsoft and OpenAI, whose successful launch of ChatGPT—a chatbot now incorporated into Microsoft’s Bing search engine—threatens to erode Google’s long-standing dominance of search. This week Google also combined its two AI research labs, DeepMind and Google Brain, in an effort to speed its AI work. 

这一举动是谷歌急于应对微软和 OpenAI 的最新迹象，后者成功推出了 ChatGPT——现在已纳入微软的必应搜索引擎——这可能会侵蚀谷歌长期以来在搜索领域的主导地位。本周，谷歌还将其两个 AI 研究实验室 DeepMind 和 Google Brain 合并，以加速其 AI 工作。 

THE TAKEAWAY Google has moved an engineering team responsible for AI chips into Google Cloud, in a step aimed at improving Google Cloud’s ability to sell the AI chips to companies that rent its servers. 

谷歌将负责 AI 芯片研发的工程团队转移到了 Google Cloud，旨在提高 Google Cloud 出售 AI 芯片给租用其服务器的公司的能力。 

Google Cloud CEO Thomas Kurian has been pressing higher-ups at Google for more resources to better equip his unit to compete with the AI-powered software now coming from OpenAI and Microsoft’s Azure cloud business and soon to come from AWS. The Microsoft cloud service offers ChatGPT-powered software that can automate certain business tasks, such as quickly creating complex websites and apps. Meanwhile, AWS announced last week plans to offer several AI models, including large-language models developed in-house and by AI startups, through its cloud. 

Google Cloud CEO Thomas Kurian 一直在向 Google 的高层争取更多资源，以更好地装备他的团队，与来自 OpenAI 的 AI 软件以及 Microsoft 的 Azure 云业务竞争，并很快就会从 AWS 中获得。Microsoft 云服务提供 ChatGPT 动力软件，可以自动化某些业务任务，例如快速创建复杂的网站和应用程序。同时，AWS 上周宣布计划通过其云提供多个 AI 模型，包括内部开发和 AI 初创企业开发的大型语言模型。 

Google has long developed chips used for AI tasks such as machine learning, known as tensor processing units, which it both uses internally and sells to customers through Google Cloud. The engineering team responsible for designing the chips had been located within Google’s core systems group, which oversees the software and hardware needed to run key Google services such as search and YouTube. In addition to designing the chips, the team develops TensorFlow, the software used to run the chips. TensorFlow supports AI elements in Google’s existing products such as web search and ad targeting. 

谷歌早已开发了用于人工智能任务（如机器学习）的芯片，称为张量处理单元，它既在内部使用，也通过 Google Cloud 出售给客户。负责设计芯片的工程团队位于 Google 的核心系统组内，该组负责运行关键的 Google 服务所需的软件和硬件，例如搜索和 YouTube。除了设计芯片外，该团队还开发了 TensorFlow，用于运行芯片的软件。TensorFlow 支持 Google 现有产品中的人工智能元素，如 Web 搜索和广告定位。 

On Thursday, Google moved the TPU and TensorFlow teams into Google Cloud, where they will be under Amin Vahdat, who reports to Kurian. The team will be combined with Google Cloud’s existing cloud AI group.  ‍ ‍ ‍ ‍ 

上周四，谷歌将 TPU 和 TensorFlow 团队转移到了谷歌云下，他们将归属于 Amin Vahdat，后者向 Kurian 汇报。该团队将与谷歌云现有的云 AI 团队合并。 

The team’s move is meant to improve Google Cloud’s ability to sell the AI chips to companies that rent its servers. It’s also intended to help Google Cloud develop more marketable AI cloud services like the kind Microsoft and OpenAI have begun to sell. Vahdat currently oversees a separate effort by Google to develop custom chips for its cloud servers that would consume less power and deliver higher processing speeds—including for some machine-learning tasks—compared to traditional servers. 

团队的举动旨在提高 Google Cloud 销售 AI 芯片给租用其服务器的公司的能力。这也旨在帮助 Google Cloud 开发更具市场营销价值的 AI 云服务，就像微软和 OpenAI 开始销售的那种。Vahdat 目前负责 Google 的另一项努力，即开发定制芯片，以便其云服务器消耗更少的电力并提供更高的处理速度，包括一些机器学习任务，相比传统服务器. 

Google Cloud already had a team dedicated to selling the chips to cloud customers, but the reorganization brings all TPU-related work under one leader. It comes as a host of companies and startups rush to develop new AI software that relies on specialized computer chips, mainly graphics processing units made by Nvidia. 

谷歌云已经有一个专门向云客户销售芯片的团队，但此次重组将所有与 TPU 相关的工作都归于一个领导者之下。这是众多公司和初创企业争相开发新的人工智能软件的时候，这些软件依赖于专门的计算机芯片，主要是由英伟达制造的图形处理单元。 

All of the major cloud providers have limited access to these chips, The Information previously reported. Microsoft is rationing the hardware among its internal teams, The Information also previously reported. At the same time, Microsoft is developing its own AI chip, known as Athena, The Information reported. 

据 The Information 之前报道，所有主要云服务提供商对这些芯片的使用都有限制。微软正在在其内部团队中限制硬件的使用。同时，微软正在开发其自己的人工智能芯片，称为 Athena。 

There’s no guarantee Google will be able to increase sales of its TPUs. Most developers prefer working with Nvidia chips because the chipmaker’s proprietary software, which AI developers need to work with the chips, is widely used. Google’s TPUs, in contrast, were built for internal purposes with software customized for Google. That could present a challenge for Google in persuading other firms to use it. Last month Google Cloud notched a win when Midjourney, a key startup developing AI that automatically generates photorealistic images, said it has been using cloud-based TPUs to train its machine-learning models. 

谷歌不能保证其 TPU 的销售额会增加。大多数开发者更喜欢使用 Nvidia 芯片，因为这家芯片制造商的专有软件是 AI 开发者需要使用的，而这种软件被广泛使用。相比之下，谷歌的 TPU 是为内部目的而建立的，其软件定制了谷歌的需求。这可能会对谷歌在说服其他公司使用它方面带来挑战。上个月，Midjourney 这个关键的 AI 开发初创企业选择了谷歌云，用于自动生成逼真图像的 AI 开发，称其正在使用基于云的 TPU 来训练其机器学习模型。 

Meanwhile, Alphabet disclosed on Thursday in a securities filing that it was adjusting how it allocates costs among its businesses, significantly reducing Google Cloud’s reported losses. Google CEO Sundar Pichai hopes Google Cloud can become the company’s third profit center after search ads and the Google Play mobile app store. Google Cloud is the biggest individual unit within Google, with around 50,000 employees. 

与此同时， Alphabet 在一份证券文件中透露，将调整其业务间的成本分配方式，大幅减少 Google Cloud 的报告损失。 Google 首席执行官桑达尔·皮查伊希望 Google Cloud 成为公司继搜索广告和 Google Play 移动应用商店之后的第三个盈利中心。 Google Cloud 是 Google 最大的单一部门，拥有约 50,000 名员工。