智东西（公众号：zhidxcom）
编译 | 陈骏达
编辑 | 李水青
智东西10月27日报道，今天，英伟达发布公告，称其将在Meta的股权发行中投资30亿美元（约合人民币200亿元）。同时，双方将扩大在个人智能体产品Muse及Muse系列模型上的合作，以提升模型的推理和训练性能。
▲英伟达公告（图源：X平台）
英伟达的这笔投资，是Meta更大融资计划的一部分。10月26日，Meta发布公告，称其计划进行一轮950亿美元（约合人民币6335亿元）的募资，一方面通过承销公开发行股票筹集350亿美元，随后通过按市价发行计划再筹集600亿美元，除了英伟达之外，AMD、博通也有意向投资Meta。
Meta预计，其全年资本支出将达到1350亿至1450亿美元之间（约合约合人民币9013亿至9680亿元），这轮融资将为其不断增长的资本开支提供支撑。
英伟达的公告显示，除了融资之外，Meta与英伟达计划在各个环节开展优化合作，以最大化AI基础设施性能并改善总体拥有成本。目前，Meta的Muse模型已经跑在英伟达的基础设施之上，Muse Wonder模型在今年10月发布时，取得前沿水平的基准测试表现。
此次合作也涉及近期爆火的个人智能体产品Muse。今年9月初，Meta正式发布Muse，一个具备跨网办事能力的通用智能体。上线10天后，Muse就登顶美国App Store免费应用总榜。
Meta的股价因Muse的热度而暴涨，一度触及779.82美元/股的高点，总市值逼近2万亿美元。Muse之后，个人智能体也成为了AI巨头的新战场。近期，从国外的OpenAI、Anthropic，到国内的阿里、蚂蚁、Manus、腾讯等企业，都陆续推出了类似思路的智能体产品。
此次合作中，Meta将考虑采用英伟达的OpenShell安全智能体参考架构，包括用于安全部署的OpenShell以及用于监控智能体行为的Sentry。
Meta还计划扩大英伟达基础设施的部署，包括面向下一代AI智能体的Vera Rubin和Vera CPU，以实现每瓦性能最大化。Meta也将部署英伟达Spectrum-X以太网网络和BlueField-4，以提升网络效率和吞吐量。
在上周的Meta财报电话会上，Meta创始人兼CEO马克·扎克伯格称，Muse上线约一个半月后，美国和加拿大已有超过1000万人配置Muse个人智能体，日活跃使用比例约为25%。购物是最常见的使用场景之一，超过四分之一的用户参与其中，平均客单价超过200美元。
扎克伯格称，Muse已经展现出清晰的产品与市场匹配度，未来希望将Muse发展为服务数十亿人的产品。
此次融资之际，Meta的股价连续多日下跌，截至美东时间10月26日收盘时，Meta的股价下跌1.88%至630.56美元，总市值为1.61万亿美元（约合人民币10.71万亿元）。
▲Meta股价（图源：腾讯自选股）
结语：智能体走向普通用户，应用与算力层协作愈发重要
当AI智能体开始走向普通用户，应用与算力层的协作也变得更加重要。对Meta而言，Muse能否从早期用户增长走向大规模、可持续的使用，仍有待市场检验。
对英伟达而言，智能体应用的普及，则有望进一步带动模型训练、推理和底层基础设施的需求。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Intellectual (WeChat Official Account:zhidxcom)
Compiled by Chen Shuda
The editor, Li Shuqing
He reported on 27 October that today, Weidar publish announced that it would invest $3 billion (approximately $20 billion RMB) in the distribution of Meta's shares. At the same time, the two sides will expand their cooperation on the individual AI agent products Muse and Muse series models in order to enhance their reasoning and training performance.
Announcer (source: platform X)
This investment in Weida is part of Meta's larger financing scheme. On 26 October, Metapublish announced that it planned a round of $95 billion (approximately RMB 633.5 billion) to raise $35 billion through the sale of publicly issued shares, followed by an additional $60 billion through the market-based distribution scheme, and that AMD and Chase, in addition to Weida, had the intention to invest in Meta.
Meta expects its annual capital expenditure to reach between $135.0 billion and $14.5 billion (approximately 90.13 billion to 96.8 billion yuan yuan), a round that will support its growing capital expenditure.
In addition to financing, the announcement by Ying Weida shows that Meta and Ying Wei Da plan are working at various stages to optimize AI infrastructure performance and improve overall ownership costs. At present, Meta's Muse model is already on top of the infrastructure in Inverda, and the Muse Wonder model achieved a front-line baseline test performance in October of this year when publish.
This cooperation also involved the recent explosion of the individual AI agent product Muse. At the beginning of September this year, Meta officially publishMuse, a universal AI agent with a cross-network capacity. After 10 days on the line, Muse will be on the US App Store free application list.
Meta ' s share price surged because of the heat of Muse, reaching the high point of $779.82/equity at one time, with a total market value of nearly $2 trillion. After Muse, the individual AI agent became the new battlefield for the AI giant. Recently, AI agent products with similar lines have been launched from Open AI, Anthropic abroad to Ari, Ant, Manus, Tencent enterprises in the country.
In this cooperation, Meta will consider the AI agent reference structure for OpenShell security in Weaverda, including OpenShell for security deployment, and Sentry for monitoring AI agent behaviour.
Meta also plans to expand the deployment of the British Wedda infrastructure, including Vera Rubin and Vera CPU, for the next generation of AIAI agent, to maximize their per-watt performance. Meta will also deploy the Spectrum-X Ethernet network and Bluefield-4 in Weiday to improve network efficiency and throughput.
At last week’s Meta Press conference, the founder and CEO Mark Zuckerberg claimed that, about a month and a half after Muse was on the line, more than 10 million people in the United States and Canada had been deployed with a personal Muse Z42Z, with an active daily use rate of about 25 per cent. Shopping is one of the most common use scenarios, in which more than one quarter of users participate, with an average per capita cost of more than $200.
According to Zuckerberg, Muse has demonstrated a clear match between products and markets and hopes to develop it into a product that serves billions of people in the future.
At the time of the financing, Meta ' s stock prices had fallen for many days, falling by 1.88 per cent to $630.56 as at 26 October of Eastern United States time, with a total market value of $1.61 trillion (approximately RMB 10,71 trillion).
▲Meta share price (source: Tencent selected share)
Conclusion: AI agent moves to a common user, and application and algorithm collaboration becomes more important
As AIAI agent begins to reach ordinary users, the collaboration between applications and the algorithm becomes even more important. For Meta, the ability of Muse to move from early user growth to large-scale, sustainable use remains to be tested in the market.
For Weeda, the spread of AI agent applications is expected to further leverage model training, reasoning and bottom infrastructure needs.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.