当地时间10月7日,据The Information援引一名知情人士消息称,英伟达曾讨论向人形机器人制造商Figure再投资10亿美元,此前尚无报道披露这笔投资计划。Figure寻求更多融资,投前估值约为380亿美元。据悉,英伟达本就是Figure的投资者;一年前,Figure按同样估值融资超过10亿美元。
On October 7 local time, according to The Information citing a person familiar with the matter, Nvidia had discussed investing another $1 billion in humanoid robot manufacturer Figure, an investment plan that had not been previously reported. Figure is seeking more financing, with a pre-money valuation of about $38 billion. It is understood that Nvidia is already an investor in Figure; a year ago, Figure raised more than $1 billion at the same valuation.
黄仁勋的焦虑
主导投资布局的英伟达CEO黄仁勋,始终对公司的全球市场地位感到不安。他此前坦言,担心AI需求降温,或出现足以替代英伟达芯片的产品,使公司失去主导地位,为此深感焦虑。英伟达持有990亿美元现金和证券,现金流也十分充沛。黄仁勋希望借助公司的财力,推动数千个AI模型蓬勃发展,避免未来只有少数模型兴旺。如此一来,英伟达的财务表现就有望减少对OpenAI及其他少数大客户的依赖。英伟达披露,截至7月的六个月内,三家客户贡献了总销售额的44%。一名了解英伟达交易情况的投资者说,如果我参与黄仁勋的战略决策,我会尽一切努力,推动数千个模型服务数千种不同的应用场景。
Jensen Huang's anxiety
Nvidia CEO Jensen Huang, who leads the investment layout, has always felt uneasy about the company's global market position. He previously admitted that he worries about AI demand cooling, or the emergence of products capable of replacing Nvidia chips, causing the company to lose its dominant position, and he is deeply anxious about this. Nvidia holds $99 billion in cash and securities, and its cash flow is also very abundant. Huang hopes to use the company's financial resources to promote the flourishing of thousands of AI models, avoiding a future in which only a few models thrive. In this way, Nvidia's financial performance is expected to reduce its dependence on OpenAI and a few other major customers. Nvidia disclosed that in the six months ending in July, three customers contributed 44% of total sales. An investor familiar with Nvidia's deals said, "If I were involved in Jensen Huang's strategic decision-making, I would do everything possible to promote thousands of models serving thousands of different application scenarios."
Figure的位置
Figure是人形机器人赛道里估值最高的未上市美国公司之一。此前Figure发布Helix 2.5,在旧金山湾区租下30套住宅,让机器人在没有任何针对性训练的情况下整理客厅、铺床和折叠毛巾。420次测试成功237次,完整任务成功率56%。Figure还宣称看到了机器人版scaling law。英伟达加码Figure,对应的判断是人形机器人是Physical AI最主要的载体之一。英伟达在机器人领域已经布局了Isaac Sim仿真、Cosmos世界模型、GR00T机器人模型和Jetson Thor端侧芯片。投资Figure,一方面可以拿到前沿机器人公司的数据和反馈,另一方面可以把Figure作为英伟达机器人技术栈的标杆案例。除了Figure,英伟达还可能投资两类初创企业:帮助公司开发旗下Nemotron开源模型的企业,以及开发其他AI应用的企业。英伟达同意花费60亿美元取得Poolside的软件许可并聘用其员工,部分原因就是希望推动Nemotron的开发。
Figure's position
Figure is one of the highest-valued private US companies in the humanoid robot track. Previously, Figure released Helix 2.5, rented 30 homes in the San Francisco Bay Area, and had robots tidy living rooms, make beds, and fold towels without any targeted training. Of 420 tests, 237 succeeded, for a full-task success rate of 56%. Figure also claimed to have seen a robot version of scaling law. Nvidia increasing its bet on Figure corresponds to the judgment that humanoid robots are one of the most important carriers of Physical AI. Nvidia has already laid out Isaac Sim simulation, Cosmos world models, GR00T robot models, and Jetson Thor edge-side chips in the robotics field. Investing in Figure, on the one hand, allows access to data and feedback from a cutting-edge robot company, and on the other hand, allows Figure to serve as a benchmark case for Nvidia's robotics technology stack. In addition to Figure, Nvidia may also invest in two types of startups: companies that help the company develop its Nemotron open-source models, and companies that develop other AI applications. Nvidia agreed to spend $6 billion to obtain a software license from Poolside and hire its employees, partly because it hopes to promote the development of Nemotron.
减少对少数客户的依赖
英伟达的AI芯片业务高度依赖少数几个大客户。三家客户贡献了总销售额的44%,这个集中度意味着一旦某个大客户的采购节奏发生变化,英伟达的财务表现就会受到直接影响。黄仁勋推动数千个模型和数千种应用场景,本质上是在扩大客户基数,降低集中度风险。投资Figure这类机器人公司,对应的是英伟达对AI算力需求下一波增长点的判断。如果人形机器人真的大规模部署,每台机器人都会产生持续的算力需求。这个需求分散在成千上万家机器人公司和企业客户手里,而不是集中在几个大模型公司。
Reducing dependence on a few customers
Nvidia's AI chip business is highly dependent on a few major customers. Three customers contributed 44% of total sales. This concentration means that once the purchasing rhythm of a major customer changes, Nvidia's financial performance will be directly affected. Huang's push for thousands of models and thousands of application scenarios is essentially about expanding the customer base and reducing concentration risk. Investing in robot companies like Figure corresponds to Nvidia's judgment about the next wave of growth in AI computing power demand. If humanoid robots are truly deployed on a large scale, each robot will generate continuous computing power demand. This demand is dispersed among thousands of robot companies and enterprise customers, rather than concentrated in a few large model companies.
