After launching a historic wave of large-scale acquisitions, this chip giant may next set its sights on robotics, autonomous driving technology, and on-device local AI models.

In July of this year, NVIDIA heard that AI model marketplace OpenRouter was about to be acquired. The three-year-old startup had become a popular hub for AI models, listing a large number of cost-effective open-source models. At the time, Databricks and Stripe were already in talks to acquire it, and NVIDIA arrived too late. NVIDIA CEO Jensen Huang wanted in on the deal. A person familiar with the matter said NVIDIA executives expressed acquisition interest to the OpenRouter team and were prepared to make a generous offer, but needed more time to evaluate the transaction. OpenRouter's founder was unwilling to wait. In the end, NVIDIA did not submit a formal acquisition offer (the outside world had not previously reported that NVIDIA was interested in buying the company), and Stripe secured OpenRouter with an $8 billion bid. NVIDIA's M&A team quickly pivoted to other targets. Over the following two months, the company finalized deals worth more than $140 billion in total, 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 deals will follow.

Although investors are worried about the financial strain on NVIDIA and other tech giants from heavy spending on AI projects, chip industry insiders expect NVIDIA to continue this epic M&A wave in the coming months. According to multiple investment bankers, lawyers, and investors who have worked with NVIDIA, the company is scouting startups to invest in or acquire in robotics, autonomous driving technology, and AI model companies that can run directly on local devices such as phones and home computers. According to previously undisclosed information, people familiar with the matter said NVIDIA is discussing an additional $1 billion investment in humanoid robot maker Figure. Before the new funding round, Figure was valued at about $38 billion. (NVIDIA is already an existing investor in Figure.) These people said NVIDIA will also look for startups to help its self-developed Nemotron open-source large model project while expanding into other AI applications. One reason NVIDIA agreed to spend $6 billion to license Poolside's software and absorb its team was to drive Nemotron's iteration; the Poolside team launched the Laguna open-weight model this year.

This series of investment moves is led by a decision-maker who constantly worries about NVIDIA's industry position. Huang has previously admitted that he is always anxious that NVIDIA's industry advantage could erode, whether because AI demand cools or because competitors emerge that can replace NVIDIA chips, either of which would hit the company's standing. With $9.9 billion in cash and marketable securities plus a steady stream of cash flow, Huang wants to use NVIDIA's financial strength to build a future where thousands of quality AI models coexist, rather than one dominated by a handful of large models. This can avoid a risk: a few major customers such as OpenAI dictating NVIDIA's overall performance. In the six months through July, NVIDIA's three largest customers contributed 44% of total revenue. An investor who understands NVIDIA's M&A strategy and invests in AI infrastructure and applications said: "If I were in Jensen's war room, I would do everything I could to promote a world where thousands of models serve countless scenarios." That also explains why NVIDIA's M&A engine is running at full speed. Recently, NVIDIA discussed investing about $2.5 billion in Thinking Machines Lab. This AI lab was founded by former OpenAI CTO Mira Murati. If the deal closes, it will be another major investment by NVIDIA in a large-model company. NVIDIA has already invested in Anthropic, Elon Musk's xAI, and several open-source model startups.

NVIDIA has also become a core funder of large data center projects. NVIDIA discussed investing $3 billion in SB Energy, participating in the OpenAI data center project and providing a huge financing guarantee for the project. Huang wrote in a blog that the data center campus that SB Energy is developing with OpenAI could deploy about $600 billion worth of NVIDIA computing power. NVIDIA chose to support OpenAI because "frontier AI labs have extremely large demand for training and inference computing power, but many institutions have expanded faster than their balance sheets and long-term credit levels can support."

Although NVIDIA missed out on OpenRouter, Huang acts decisively in other deals once competitors emerge for a target he wants. Take the recent acquisition of Hugging Face. Investment bankers and people close to the company say this decade-old startup is already a mainstream open-source AI model repository and has long received various acquisition approaches. People familiar with the matter said senior executives in NVIDIA's corporate development department had long been interested in investing in Hugging Face. In early summer this year, after OpenAI's intelligent agent accessed the Hugging Face platform, OpenAI and Hugging Face held preliminary talks about a plan to invest $100 million in it. Around July, competitors including existing Hugging Face investor Salesforce also expressed acquisition interest. Hugging Face co-founder Clem Delangue contacted Huang to tell him he had received multiple potential acquisition offers. People familiar with the matter said Huang quickly pushed the deal forward and assured Delangue that NVIDIA was the only trustworthy partner that could keep the Hugging Face open-source model community running. Huang made a $12.9 billion offer. The startup had annualized revenue of $150 million, and the acquisition premium exceeded 80 times, making the offer hard to refuse. Delangue said at a press conference: "Along the way, Hugging Face has received a large number of investment and acquisition offers, and in the past we declined them all. But this summer, everything was in place."

People involved in NVIDIA's work in the home computing arena said that in the coming months NVIDIA will pursue more deals to push GPUs out of data centers and into home devices. More and more users will run AI locally on computers and small devices. This year, the rapid spread of AI intelligent agents has driven a surge in local AI demand. Such software can complete multi-step tasks, such as booking flights and organizing email. Many users are buying hardware suited to local AI tasks. NVIDIA has also launched new products such as DGX Spark, designed specifically to run AI intelligent agents locally. Some of NVIDIA's recent deals reflect its intention to expand its product matrix in this area. People familiar with the matter said Perplexity, a startup that began with AI search and has now developed the AI intelligent agent Perplexity Computer, demonstrated to NVIDIA engineers in June this year that Perplexity software could run on two DGX Spark devices. After hearing about the demonstration, Huang was very interested, and the two companies continued talks throughout the summer about a potential deal. People familiar with the matter said Perplexity co-founder and CEO Aravind Srinivas proposed to Huang that NVIDIA could directly acquire Perplexity. The two sides then shifted to discussing a "technology licensing plus talent absorption" plan: NVIDIA would pay at least $20 billion or even more to obtain the right to use Perplexity's technology. The two sides finally announced the partnership at the end of August. Under the agreement, Perplexity released a customized new version of its app to better run its AI intelligent agent on DGX Spark hardware. The Information was first to report the investment: NVIDIA plans to invest in Perplexity at a pre-money valuation of $35 billion.

Financial Risk Like most tech companies, NVIDIA has a corporate development team responsible for M&A. The unit is led by Vishal Bhagwati, a former executive at Hewlett Packard Enterprise and Oracle. Even so, Huang often gets deeply involved in deal details, personally setting prices and leading high-risk acquisition negotiations such as Hugging Face. Huang also frequently meets with startup founders, investors, and executives of private-equity-controlled companies to understand how they use NVIDIA products and what support NVIDIA can provide. Executives such as Microsoft CEO Nadella also use similar informal visit-and-research methods.

Over the past few months, Huang has had to face a new reality that may limit NVIDIA's M&A ambitions: even with financial strength like NVIDIA's, there may be a safety spending cap when it comes to investing, acquiring, and providing guarantees for large projects. SoftBank's SB Energy is developing a large data center project on federal land in Ohio, and this financing exposed market concerns. In early summer, SoftBank and NVIDIA initially discussed NVIDIA providing up to $250 billion in credit support for OpenAI, which plans to lease the data center to train and run models on NVIDIA chips. (NVIDIA also participated in OpenAI's new funding round, investing $3 billion, with the final $1 billion completed on October 1.) In August, NVIDIA's credit default swap (CDS) spreads widened, reflecting investor concerns that the company was taking on excessive risk. Huang also noticed the change in spreads. NVIDIA ultimately provided only a $10.5 billion credit guarantee for phase one of the project. The amount is still huge, but less than half the scale initially discussed. The project is being implemented in phases, giving NVIDIA several years to decide whether to support phase two. At the same time, NVIDIA is advancing another investment, planning to put $3 billion into SB Energy before and during its listing. Huang stressed that sustained progress in AI requires more institutions to jointly fund the construction of chips, data centers, and power infrastructure. In early August, Huang convened six Wall Street institutions, including Blackstone, Apollo, and Goldman Sachs, to raise funds for hardware projects. NVIDIA said that in some related deals it can provide backstop guarantees for up to 25% of a project's total financing. But Huang downplayed outside concerns, arguing that a well-capitalized investor like NVIDIA is not being overly aggressive. In September, Huang told the audience at Goldman Sachs' annual technology conference in San Francisco: "People are slowly realizing that wherever I invest, it won't be a bad target, because I have an information advantage. I won't take unnecessary risks. We are not as smart as everyone imagines. I pursue sure wins."