After initiating a historic large-scale merger and acquisition, this chip giant is next set to expand into the fields of robotics, autonomous driving technology, and edge-side local AI models.

In July this year, NVIDIA learned that the AI model trading platform OpenRouter was about to be acquired. This three-year-old startup has become a popular AI model market, with a large number of cost-effective open-source models available on the platform. At that time, Databricks and Stripe were already in negotiations for the acquisition, so NVIDIA entered the game a bit late. NVIDIA CEO Jensen Huang hopes to participate in this acquisition as well.

An informed source stated that NVIDIA executives expressed their interest in acquiring OpenRouter and were prepared to make a generous offer, but they needed more time to evaluate the transaction. The founder of OpenRouter was unwilling to wait. In the end, NVIDIA did not submit a formal acquisition bid (there had been no prior reports indicating that NVIDIA was interested in acquiring this company), and Stripe acquired OpenRouter for $8 billion.

NVIDIA's mergers and acquisitions team quickly turned their attention to other targets. In the following two months, the company finalized various transactions totaling over $140 billion, including a $10.5 billion credit guarantee. As of the end of July, NVIDIA held nearly $10 billion in equity investments, with an additional $2.5 billion in future investment commitments. More transactions are expected to be completed in the coming period.

Despite investors' concerns about NVIDIA's substantial investments in the AI project alongside other tech giants, and the increasing financial pressure, industry insiders expect NVIDIA to continue this wave of epic mergers and acquisitions in the coming months. Several investment bankers, lawyers, and investors who have worked with NVIDIA have revealed that NVIDIA is looking for startups to invest in or acquire those focusing on robotics, autonomous driving technology, as well as AI model-based solutions that can run directly on local devices such as smartphones and home computers.

A previously undisclosed piece of information indicates that sources familiar with the matter claim that NVIDIA is discussing adding an additional $1 billion in investment to the humanoid robot manufacturer Figure. Before this new round of financing, Figure was valued at approximately $38 billion. (NVIDIA is already an existing investor in Figure.)

These individuals stated that NVIDIA will also be looking for startups to assist in its own Nemotron open-source large model project, while also laying out for other AI applications. One of the reasons NVIDIA agreed to spend $6 billion to acquire Poolside software licenses and absorb its team is to drive the iteration of Nemotron; this year, the Poolside team launched the Laguna open-source weight model.

This series of investment decisions was led by a person who is constantly concerned about NVIDIA's position in the industry. Jensen Huang has previously admitted that he is always anxious about NVIDIA's competitive advantages being lost: whether it's due to a decline in demand for AI or the emergence of competitors that can replace NVIDIA's chips, both could impact the company's position.

With $9.9 billion in cash and securities in hand, along with a steady stream of cash flow, Jensen Huang hopes to leverage NVIDIA's substantial financial resources to create a future where thousands of high-quality AI models coexist, rather than a landscape dominated by a few large models. This can help avoid a risk: a situation where a handful of major clients, such as OpenAI, influence NVIDIA's overall performance. In the first half of the year ending July, NVIDIA's top three customers contributed 44% of its total revenue.

An investor who is familiar with NVIDIA's acquisition strategy and its investment in AI infrastructure and applications said, "If I were in Jensen Huang's war room, I would do everything in my power to create a world where thousands of models serve tens of thousands of scenarios."

This also explains why NVIDIA's engine is running at full speed after the acquisition.

Recently, NVIDIA has been in talks to invest approximately $2.5 billion in Thinking Machines Lab. This AI laboratory was founded by OpenAI, formerly CTO Mila Murati. If the deal goes through, this will be another major investment by NVIDIA in large model manufacturers. Previously, NVIDIA has already invested in Anthropic, Elon Musk's xAI, as well as several open-source model startups.

NVIDIA has also become a key investor in large-scale data center projects. NVIDIA is in talks to invest $3 billion in SB Energy, participate in the OpenAI data center project, and provide substantial financing guarantees for this project.

Jensen Huang wrote in his blog that the data center park, which is a collaboration between SB Energy and OpenAI, can deploy NVIDIA computing power worth approximately $600 billion. NVIDIA chose to support OpenAI because "the cutting-edge AI laboratories have an extremely high demand for training and inference computing power, but many institutions' expansion rates have exceeded the capacity of their balance sheets and long-term credit levels."

Although he missed out on OpenRouter, once other bidders emerged for his desired target, Jensen Huang was very decisive in other transactions.

Taking the recent acquisition of Hugging Face as an example, investment banks and people close to this company said that this startup, which has a history of ten years, has become a mainstream open-source AI model repository and receives various acquisition offers throughout the year. People familiar with the matter stated that executives from NVIDIA's corporate development department have always been interested in investing in Hugging Face.

At the beginning of this summer, after OpenAI's intelligent agents accessed the Hugging Face platform, OpenAI and Hugging Face began preliminary discussions, planning to invest $100 million in it. Around July, Hugging Face's existing investor, Salesforce, as well as other competitors, also expressed their interest in making a takeover bid.

Hugging Face Co-founder Clem Dranger contacted Jensen Huang, informing him that they had received multiple potential acquisition offers. According to people familiar with the matter, Jensen Huang quickly pursued the transaction, assuring Dranger that NVIDIA was the only trustworthy partner capable of maintaining the ongoing operation of the Hugging Face open-source model community. Jensen Huang offered a price of $12.9 billion; this startup has an annual revenue of $150 million, making the acquisition premium more than 80 times their annual revenue, an offer that was difficult to refuse.

De Lange stated at the press conference, "Throughout our journey, Hugging Face, we have received numerous investment and acquisition offers, which we have declined in the past. But this summer, everything is ready."

Home Computing Track

People involved in the cooperation with NVIDIA have revealed that in the coming months, NVIDIA will push forward more initiatives to bring GPU out of data centers and into home devices. More and more users will run AI locally on their computers and small devices.

This year, the rapid popularization of AI intelligent agents has led to a surge in demand for local AI. Such software can perform multi-step tasks, such as booking airline tickets and organizing emails. A large number of users have purchased hardware adapted for local AI tasks.

NVIDIA has also launched new products such as DGX Spark, which are specifically designed for running AI smart proxies locally. Some of NVIDIA's recent transactions reflect its intention to expand its product portfolio in this area.

According to informed sources, Perplexity, a startup company that began with AI search capabilities and has now developed AI intelligent agents, demonstrated to NVIDIA engineers in June of this year that its Perplexity software can run on two DGX Spark devices.

Jensen Huang became very interested after hearing the demonstration, and the two companies continued to negotiate a potential deal throughout the summer. According to people familiar with the matter, Arvind Srinivas, co-founder of Perplexity and also involved with CEO, suggested to Jensen Huang that NVIDIA could directly acquire Perplexity.

Subsequently, both parties turned to negotiate a 'technology licensing + talent acquisition' deal: NVIDIA would invest at least $20 billion or more to obtain the rights to use the technology of Perplexity.

The two parties officially announced their cooperation at the end of August. According to the agreement, Perplexity will release a customized new version of its application to better run its AI intelligent proxy on DGX Spark hardware. The Information was the first to report on this investment: NVIDIA plans to invest Perplexity, with a pre-investment valuation of 35 billion US dollars.

Financial Risk

Like most technology companies, NVIDIA has a corporate development team responsible for mergers and acquisitions (M&A) activities. This department at NVIDIA is led by Vishal Bagwati, a former executive from HP Enterprise and Oracle. Nevertheless, Jensen Huang often gets deeply involved in the details of these transactions, personally determining the prices and overseeing high-risk acquisition negotiations such as Hugging Face.

Jensen Huang also frequently meets with founders of startups, investors, and executives of private equity holding companies to understand how they use NVIDIA products and what support NVIDIA can provide. Executives at Microsoft, such as Satya Nadella, also adopt similar informal methods of field research and visits.

In the past few months, Jensen Huang has had to face a new reality that could limit NVIDIA's ambitions for mergers and acquisitions: even a company as financially strong as NVIDIA may reach its upper limit for security expenditures when it comes to investing, making acquisitions, and providing guarantees for large-scale projects.

SoftBank's SB Energy is undertaking a large-scale data center project on federal land in Ohio, and this financing has exposed market concerns. Earlier this summer, SoftBank initially discussed with NVIDIA the possibility of NVIDIA providing up to $250 billion in credit support for OpenAI. OpenAI plans to rent this data center to train and run models on NVIDIA chips. (NVIDIA also participated in a new round of fundraising for OpenAI, investing $3 billion, with the final $1 billion being delivered on October 1st.)

In August, the credit default swap (CDS) spread of NVIDIA widened, reflecting investors' concerns that the company was taking on excessive risks. Jensen Huang also noticed this change in the CDS spread.

NVIDIA ultimately provided only a credit guarantee of $1.05 billion for phase one of the project. Although this amount is still substantial, it is less than half of the initial negotiation scale. The project will be implemented in phases, giving NVIDIA several years to decide whether to provide support for the second phase of the project.

At the same time, NVIDIA is making another investment, planning to invest $3 billion in SB Energy before and during its listing period.

Jensen Huang emphasized that to continue to drive the progress of AI, more institutions need to jointly invest in building chips, data centers, and power infrastructure. At the beginning of August, Jensen Huang convened six Wall Street institutions, including Blackstone, Apollo, and Goldman Sachs, to raise funds for the hardware project. NVIDIA stated that in some related transactions, it can provide a maximum guarantee of up to 25% of the total project financing.

However, Jensen Huang alleviated external concerns, stating that investors with substantial funds like NVIDIA are not overly aggressive.

In September, Jensen Huang told the audience at Goldman Sachs' annual technology conference in San Francisco: "People are gradually realizing that wherever I invest, it's not a bad investment because I have an informational advantage. I don't take unnecessary risks. We're not as smart as people think. I pursue winning trades."

Responsible Editor: Guo Mingyu