美国科技巨头Meta在宣布大幅提高人工智能基础设施投资计划后，获得了来自英伟达的30亿美元资金支持。双方于10月27日确认，将进一步深化围绕Muse人工智能助手及相关模型系列的合作关系，同时扩大硬件和软件层面的协同开发。
根据双方公布的信息，英伟达将向Meta的股票发行计划投资30亿美元。此前Meta宣布计划通过公开募股和后续售股项目筹集资金，用于加速人工智能基础设施建设。随后Meta将整体融资目标从950亿美元提高至975亿美元，成为公司历史上规模最大的融资计划之一。
英伟达表示，此次合作重点在于提升Muse及其相关模型的训练和推理性能，同时降低运行成本。双方计划针对人工智能工作流展开更深层次优化，以提高计算效率并改善总体拥有成本。
作为合作的一部分，Meta将进一步扩大英伟达新一代AI基础设施的部署规模，包括Vera Rubin系统和Vera CPU，同时采用Spectrum-X以太网网络方案以及BlueField-4数据处理技术，以提升数据传输效率、网络吞吐量和单位能耗下的计算性能。
除硬件合作外，双方还在评估进一步采用英伟达OpenShell参考架构的可能性。相关合作将包括利用OpenShell实现AI代理的安全部署，以及借助Sentry工具对人工智能代理行为进行监测和管理，从而提升系统的安全性和可控性。
英伟达首席执行官黄仁勋表示，AI代理正在代表一种新的计算范式。这类软件不仅能够理解用户目标，还能够进行推理、调用工具并代替用户完成任务。他认为，这种转变正在改变人们使用计算机的方式，使用户从操作应用程序转向与智能代理协同工作。
Meta首席执行官马克·扎克伯格则表示，Muse是Meta推动“个人超级智能”战略的核心组成部分，公司希望借助这一平台将先进人工智能能力带给数十亿用户，并与英伟达共同打造面向未来的个人智能代理与前沿模型。
据介绍，Muse被Meta视为下一代个人AI助手，可在用户授权的前提下访问电子邮件、日历、支付服务、健康应用、购物平台以及智能家居系统等多种服务，并能够执行发送邮件、安排旅行等实际任务。用户可以随时撤销相关权限。
Meta表示，Muse及其模型家族自发布以来已在多项基准测试中取得领先表现，同时用户下载量持续增长。不过公司并未披露具体数据。
此次投资也再次引发市场对于“循环融资”模式的讨论。由于Meta未来将继续大规模采购英伟达AI硬件，外界认为芯片供应商向客户提供资金支持，再由客户采购更多芯片产品，正在成为人工智能产业中越来越常见的商业模式。
Meta预计其本年度资本支出将在1350亿美元至1450亿美元之间，绝大部分将用于建设人工智能基础设施。与此同时，公司最新财报显示，其季度营收同比增长26%，达到646亿美元。分析人士认为，随着AI竞争持续升温，Meta与英伟达之间的合作关系还将进一步深化，而双方正在押注AI代理将成为继移动互联网之后的新一代计算平台。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Following the announcement of the United States technology giant Meta ' s programme to significantly boost AII investment, it received $3 billion in financial support from Weeda. The parties confirmed on 27 October that they would further deepen their cooperation around the Muse manual intelligence assistant and associated model series, while expanding the synergistic development of hardware and software levels.
According to the information published by both parties, Ying Weida will invest $3 billion in Meta ' s stock distribution scheme. Earlier, Meta had announced plans to raise funds through the Open Collection and Follow-Up Sale Unit projects to accelerate the construction of artificial intelligence infrastructure. Meta subsequently increased the overall financing target from $95 billion to $97.5 billion, becoming one of the largest financing schemes in company history.
According to Ying Weida, the focus of this collaboration was on upgrading the training and reasoning performance of the Muse and its associated models, while reducing operational costs. The parties plan to further optimize artificial intelligence workflow to improve computing efficiency and overall ownership costs.
As part of this cooperation, Meta will further scale up the deployment of a new generation of AI infrastructure in Weida, including the Vera Rubin system and Vera CPU, with the Spectrum-X Ethernet network programme and Bluefield-4 data-processing technology to improve data transmission efficiency, network throughput and computing performance under unit energy consumption.
In addition to hardware cooperation, the parties are assessing the possibility of further adopting the OpenShell reference architecture in British Wida. Cooperation will include the use of OpenShell to achieve the safe deployment of AI agents and the monitoring and management of artificial intelligence agents through the use of the Setry tool, thereby enhancing system security and control.
Chief Executive Officer Wong In-hoon of Weida indicated that AI was representing a new model of calculation. Such software not only understands user objectives, but also provides reasoning, calls tools and substitutes for user assignments. In his view, that shift was changing the way people used computers and shifting users from operating applications to working with smart agents.
The CEO of Meta, Mark Zuckerberg, for his part, stated that Muse is a central component of Meta's strategy to promote “personal super-intelligence”, a platform through which companies want to bring advanced artificial intelligence capabilities to billions of users and to work with Young Wida to develop future-oriented personal intelligence agents and front-line models.
It was described that Muse is considered by Meta as the next generation of personal AI assistants, who can access a wide range of services such as e-mail, calendars, payment services, health applications, shopping platforms and smart home systems, subject to user authorization, and can perform practical tasks such as sending mail, arranging travel, etc. Users can withdraw permission at any time.
Meta states that Muse and its model family have been leading in many benchmark tests since publish, while user downloads have continued to grow. However, the company did not disclose specific data.
This investment also rekindled the market's discussion of the “revolving finance” model. As Meta will continue to procure British-Wida-AI hardware on a large scale in the future, it is believed that chip providers provide financial support to clients and that the purchase of additional chip products by clients is becoming an increasingly common business model in the AI industry.
Meta expects that its capital expenditure in the current year will range from $135.0 billion to $14.5 billion, the bulk of which will be spent on building artificial intelligence infrastructure. At the same time, the company ' s latest financial statements show a 26 per cent increase in quarterly revenue over the same period, to $64.6 billion. The analysts believe that as competition continues to rise in AI, the relationship between Meta and Weeda will further deepen, while both parties are betting on AI agents as a new generation of computing platforms after the mobile Internet.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.