Jensen Huang has gone on a spending spree, investing 65.5 billion yuan in just 7 days.
NVIDIA's overseas investments are accelerating again.
According to Crunchbase statistics, from 2020 to 2024, NVIDIA made 2, 10, 9, 47, and 56 investments in startups each year respectively. This year is the most investment - intensive year since its establishment. For example, within just one week in mid - September, it officially announced five overseas investments with a total amount exceeding $9.2 billion (approximately 65.5 billion RMB).
Moreover, these five investments do not include the $100 billion investment in OpenAI. As mentioned in the article "Jensen Huang Goes All - In with $710 Billion" by Touzhongwang, this "stepping on one's own feet" style of investment is an American - style industrial investment. Large companies cooperate to build an AI industrial ecosystem. Five investments in one week, especially with significant amounts, show an aggressive pace.
Five Investments: Betting on AI Infrastructure and Robotics
Among these five investments, the most significant one is the $5 billion investment in Intel. There have been many analyses and reports in the market, so there is no need to elaborate further. From the perspective of business layout, this helps NVIDIA reduce its dependence on TSMC. Meanwhile, cooperation with Intel may spark new ideas at the product level. After all, in the minds of many enterprise customers, Intel's products and brand have a good reputation.
In addition, NVIDIA spent $900 million to acquire Enfabrica, an American AI network chip startup. This is a comprehensive acquisition. It not only recruited the founder and CEO Rochan Sankar and his core team but also obtained the company's key technology license.
Image from the official website
In 2020, NVIDIA acquired the network giant Mellanox for $7 billion, filling the technological gap in the high - speed network interconnection field. Five years later, NVIDIA turned its attention to the network chip field again, targeting the current bottleneck in AI memory. Enfabrica's products not only bring innovative solutions to this industry problem but also achieve both "cost reduction" and "efficiency improvement", especially suitable for memory - intensive scenarios such as long prompts, large context windows, and multiple AI agents.
The founders of this company, Rochan Sankar and Shrijeet Mukherjee, are also industry veterans. They have a profound industry background and have worked at Broadcom and Cisco respectively. The rare technological innovation and the valuable industry background advantage explain why NVIDIA didn't hesitate to acquire the entire team.
The other two financings were achieved under the same background. In September, Jensen Huang visited the UK and promised to invest £2 billion (approximately $2.6 billion) in the UK to enhance the construction of the UK's artificial intelligence startup ecosystem.
During this visit, Jensen Huang listed eight UK startups, including Revolut, the artificial intelligence video company Synthesia, and the autonomous transportation group Oxa, and told each of them, "I will invest in your next round of financing."
However, the first ones to be selected were the autonomous driving company Wayve and the AI infrastructure company Nscale.
Wayve said that NVIDIA will participate in its next - round strategic financing of $500 million, and the two parties have already signed a letter of intent. This is not NVIDIA's first investment in Wayve. In May last year, NVIDIA participated in Wayve's Series C financing of $1.05 billion.
Public information shows that Wayve was founded in 2017. Its autonomous driving system has attracted much attention and favor from investors because it uses a self - learning rather than rule - based autonomous driving software method. The company has been using NVIDIA's systems since 2018 and is regarded by Jensen Huang as a "next - trillion - dollar" company.
In contrast, NVIDIA's relationship with Nscale doesn't seem as harmonious. Nscale claims to be a "hyperscale computing platform designed for artificial intelligence". A few days ago, Nscale, which has been established for less than two years, officially announced a Series B financing of $1.1 billion. This round of financing was led by the Norwegian energy infrastructure group Aker, and many companies such as Nokia and NVIDIA also participated.
According to PitchBook data, Nscale's financing this round is the largest venture capital transaction completed in the UK so far this year and the second - largest in Europe, only after the €1.7 billion (approximately $2 billion) Series C financing completed by Mistral AI, a competitor of OpenAI, earlier this month.
Combined with previous news, NVIDIA invested $683 million in Nscale this time. Besides the equity transaction, the bigger goal is to expand the GPU capacity in the UK to 60,000 units by 2026. The hardware will be deployed in Nscale's data center. Recently, the company also announced a plan to cooperate with Microsoft to build the largest supercomputer in the UK.
Just two years ago, Nscale was not an independent entity. In May last year, it was spun off from the cryptocurrency mining infrastructure provider Arkon Energy to meet the surging demand for AI - dedicated data centers.
Like CoreWeave in the United States, Nscale repositioned its cryptocurrency roots into an AI infrastructure business, combining a large data center capacity, high - density power, and thousands of GPUs with a client software layer. Nscale initially relied on AMD hardware. As the cooperation deepened, Nscale switched to NVIDIA GPUs.
From NVIDIA's two investment moves in the UK, it can be seen that it first chose to invest in its most potential customers, and the purpose of industrial investment is obvious.
Another company that NVIDIA invested in is the embodied intelligence company Dyna Robotics. Not long ago, the company officially announced a Series A financing of $120 million, and its valuation soared from $100 million to over $600 million, a five - fold increase in half a year.
This is a company that has been established for only one year. In April this year, they released their self - developed VLA model DYNA - 1 (Dynamism v1), which is the world's first dexterous operation foundation model that can be applied in commercial scenarios. Besides the popularity of the field it is in, it also has an obvious label - an all - Chinese team, including Lindon Gao, York Yang, and Jason Ma.
Among them, Lindon Gao is the co - founder and CEO of Dyna. He graduated from the Stern School of Business at New York University and is responsible for the company's overall strategy and commercialization direction. York Yang and Jason Ma are respectively responsible for R & D and the foundation model. York studied electronic engineering at Zhejiang University as an undergraduate and then went to UCLA to pursue a master's degree in computer science. Jason is a PhD from the Department of Computer and Information Science at the University of Pennsylvania and has worked in cutting - edge artificial intelligence laboratories such as Google DeepMind, NVIDIA AI, and Meta AI.
Judging from the team configuration, Dyna indeed fits the image pursued in Silicon Valley. Moreover, they have not only conducted scientific research but also launched products and models. Looking at NVIDIA again, this is its third robotics project investment this year. The other two are Figure AI and Skild AI.
To sum up briefly, except for the investment in Intel, NVIDIA's three financings all went to existing or potential customers. This kind of cooperation based on equity investment or equity investment based on cooperation is worth pondering. The only acquisition was for technological considerations, providing new solutions for the current AI needs.
NVIDIA's New Problem: Too Much Money
Actually, not only in the UK and the United States, NVIDIA is making investments through alliances and partnerships globally, turning the strong players in various fields into close allies. Of course, a necessary prerequisite here is that NVIDIA has sufficient funds.
According to FactSet's estimate, NVIDIA generated $72 billion in free cash flow in the past four quarters and is expected to reach slightly less than $100 billion by the end of this fiscal year. This exceeds the expected free cash flow of all large technology companies this year except Apple.
Having money is one thing, but making it work is not easy.
For example, NVIDIA repurchased nearly $50 billion worth of stocks in the past four quarters and recently added another $60 billion to its share - buyback plan.
Another example is that even though R & D spending has doubled in the past two years, in the past four quarters, NVIDIA's R & D spending accounted for only a little over 9% of its revenue. This shows that there are still too many orders, and the explosive sales growth has made this expenditure seem insignificant.
The surging revenue growth is also becoming a sweet burden for NVIDIA. On the one hand, as mentioned at the beginning, how to spend this money; on the other hand, how to maintain the growth momentum of orders? Or, more grandly, how to seize the AI dividend?
Expanding investments and making small acquisitions seem to be the general solutions to these two problems.
As mentioned above, NVIDIA's investments in startups ultimately point to cooperation. So we can see NVIDIA "selecting the best" in various AI fields and regions this year. This operation is not a hasty decision but has a historical basis. Some analyses point out that for every $10 billion NVIDIA invests in OpenAI, the company will spend $35 billion on NVIDIA chips.
In the short term, this approach may result in a loss of some profits, but it ensures continuous demand and provides a lifeline for startups. It's a worthwhile deal.
Jensen Huang said that NVIDIA's positioning is no longer just a chip company. It focuses on the entire AI infrastructure and seeks various ways to cooperate with all parties. He also humorously said, "We don't ask anyone to buy everything from us. My only request is to buy something from us."
Besides the purpose of industrial investment, NVIDIA is also using large - scale investments to make up for regrets and prevent missing opportunities.
Screenshot of Jensen Huang's interview
In a recent interview, Jensen Huang recalled his relationship with OpenAI and admitted that he had left deep regrets in investment decisions in the early years. He mentioned that when OpenAI invited NVIDIA to participate in early - stage investment, NVIDIA had limited funds and didn't bet enough: "They invited us to invest at that time, but we were too poor to invest enough. I should have given them all my money."
Who wouldn't want to be an investor in a company that is expected to become the next trillion - dollar company?
Speaking of small - scale mergers and acquisitions, for NVIDIA, any large - scale acquisition faces strict supervision and review, so this is only an option of last resort.
In contrast, Jensen Huang prefers a flat organizational structure. He has a large number of direct subordinates, which makes small - scale supplementary acquisitions within the company more popular than large - scale acquisitions, like the Enfabrica case.
Since this year, NVIDIA has been continuously looking for new customers and markets globally, including promoting the construction of "AI factories" in fields such as industry and seizing the opportunity of sovereign AI to build AI infrastructure in many places around the world. These actions also guarantee the sales of the next - generation chip platforms to some extent.
It's lonely at the top.
What lies in front of NVIDIA, a company with a $4 - trillion market value, is not only technology, customers, and growth but also a new subject about fund allocation and utilization.
This article is from the WeChat public account "Dongshisitiaojie Capital" (ID: DsstCapital). The author is Zhang Xue, and it is published by 36Kr with authorization.