Where Nvidia’s $100 Billion Dealmaking Juggernaut Will Go Next

This July, Nvidia caught wind of a looming takeover of OpenRouter, a three-year-old startup that had emerged as a popular marketplace for AI models, including many cheaper open-source ones. Databricks and Stripe had already been talking to the startup about a potential sale. Nvidia was late to the party. The chipmaker’s CEO, Jensen Huang, wanted in.

Nvidia executives told their counterparts at OpenRouter they were interested and were prepared to make a generous offer, but they needed more time to evaluate a possible transaction before doing so, said a person with knowledge of the discussion. OpenRouter’s founders didn’t want to wait. Ultimately, Nvidia—whose interest in the startup hasn’t previously been reported—didn’t make a formal offer, and Stripe prevailed with an $8 billion bid for OpenRouter.

Nvidia’s dealmakers quickly moved on to other prizes. Over the next two months, they inked more than $140 billion worth of deals—including a $105 billion credit guarantee. The company held nearly $100 billion worth of equity investments and $25 billion in future investment commitments as of late July. More deals are likely.

Despite jitters among investors that Nvidia and other tech giants are stretching themselves financially with their AI commitments, the chip designer is expected to extend its historic dealmaking spree in the coming months. The firm is scouting for investments in or acquisitions of startups that make robots, self-driving car technology and AI models that run directly on local devices—like smartphones, home computers and other devices, according to bankers, lawyers and investors who have worked with Nvidia.

In one previously unreported example, Nvidia has discussed investing another $1 billion in humanoid robot maker Figure, according to a person familiar with the matter. Figure is trying to raise more financing at a valuation of around $38 billion before the new money. (Nvidia is already an investor in Figure, which a year ago raised more than $1 billion at that valuation.)

Nvidia could also take a swing at startups that boost its own efforts to develop its Nemotron open-source models, as well as those working on other AI applications, those people said. Helping advance Nemotron was one reason it agreed to pay $6 billion to license software and hire employees from Poolside, whose staff this year launched the Laguna open-weight AI models.

The investment push is overseen by a leader who lives in a constant state of unease about Nvidia’s position in the world. In the past, Huang has said he is consumed with anxiety about Nvidia’s dominance slipping—either because of cooling demand for AI or the emergence of compelling alternatives to its chips.

Armed with $99 billion in cash and securities and a gusher of cash flow, he wants to use Nvidia’s financial heft to ensure a future with thousands of prosperous AI models, not just a small handful. That could help the chipmaker avoid a scenario where OpenAI and a couple of other big customers hold outsized sway over its financial performance. For the six months ending in July, three customers were responsible for 44% of total sales, Nvidia has disclosed.

“If I were in the Jensen war room, I’d be doing everything I can to tilt the scales in favor of a world where there are thousands of models for thousands of different use cases,” said an investor in AI infrastructure and applications with knowledge of Nvidia’s dealmaking.

That explains why Nvidia’s dealmaking engine is in high gear.

Recently, the chip designer was in talks to invest about $2.5 billion in Thinking Machines Lab, the AI lab led by ex-OpenAI Chief Technology Officer Mira Murati, The Information reported. The investment, if finalized, would add to a string of such deals for model makers including Anthropic, Elon Musk’s xAI and the maker of open-source models, Reflection AI.

And it’s also emerged as an essential backer of costly data center projects, agreeing to pour $3 billion into SB Energy, which is developing a massive data center for OpenAI, as well as guaranteeing $105 billion in OpenAI leases for the project.

Huang said in a blog post that the SB Energy-OpenAI data center campus could hold some $600 billion worth of Nvidia compute, and that Nvidia is choosing to back OpenAI in the deal because “frontier AI labs have extraordinary demand for training and inference compute, but many are growing faster than their balance sheets and long-term credit profiles can support.”

While Nvidia let OpenRouter get away, Huang in other cases moves swiftly when a company he wants to buy has another suitor.

Take its recent acquisition of Hugging Face. In the last few years, the ten-year-old startup had emerged as a popular repository of open-source AI models and regularly received interest from would-be acquirers, according to bankers and people close to the company. In fact, Nvidia’s corporate development executives had always been interested in making an investment in Hugging Face, said people familiar with the matter.

Then, after OpenAI’s agents hacked Hugging Face early this summer, OpenAI held early discussions with Hugging Face to invest $100 million in the startup, according to a person with knowledge of the plan. Around July, other rivals, such as Salesforce, an investor in Hugging Face, expressed interest in buying the startup as well.

Hugging Face co-founder Clem Delangue approached Huang, telling him about the potential offers. Huang worked to get the deal done fast, reassuring Delangue that Nvidia would be the only partner Delangue could trust to keep Hugging Face’s community of open-weight AI models up and running, said the people. Huang’s offer of $12.9 billion was also hard to turn down, amounting to more than 80 times the startup’s annualized revenue of $150 million.

“Throughout the life of Hugging Face, we always got quite a lot of offers [for] investments [and] acquisitions that we turned down in the past,” Delangue said in a press briefing. “But this summer, the planets aligned.”

Home Computing

In the coming months, Nvidia could be looking to do more deals that help it expand its graphic processing units beyond data centers and into homes, where users may increasingly run AI locally on computers and other smaller devices, said people who work with Nvidia.

Demand for such local AI has jumped this year thanks to the proliferation of AI agents—the software that performs multistep tasks involved in ordering plane tickets or organizing email inboxes. As a result, people using these agents have snapped up Apple’s Mac mini computers, which have proven to be well-suited to handling such tasks.

For its part, Nvidia has come up with new products—such as the DGX Spark computers—that are also designed to run agents locally. Some of Nvidia’s recent deals have reflected a desire to expand its offerings in the category.

In June, engineers at Perplexity, a startup that began as an AI-powered search engine and has expanded to develop an AI agent called Perplexity Computer, gave a presentation to Nvidia’s engineers that showed how they could run Perplexity’s software on two DGX Spark computers, according to people with knowledge of the presentation.

The presentation excited Huang when he heard about it—and kicked off a summer of discussions between the two companies about a possible deal. Aravind Srinivas, Perplexity’s co-founder and CEO, suggested to Huang that Nvidia could buy Perplexity outright, according to people briefed on the matter.

The companies’ leaders then moved on to discuss a license-and-hire deal in which Nvidia could pay Perplexity at least $20 billion or possibly even more for rights to Perplexity’s technology, according to people familiar with the matter.

They ultimately landed on a partnership that was announced in late August. As part of their agreement, Perplexity released a new app, called Portable Computer, that was customized to run its AI agent better on DGX Sparks. And Nvidia, in a deal first reported by The Information, made plans to invest around $3 billion in an equity round that would value the startup at $35 billion before the investment, according to a person with knowledge of the deal.

Financial Risks

Like most tech companies, Nvidia has a staff of corporate development specialists that oversee its dealmaking. In the chipmaker’s case, that team is led by former HPE and Oracle executive Vishal Bhagwati. Despite that, Huang is often directly involved in the nitty-gritty of the company’s deals, hashing out their prices himself and leading negotiations for high-stakes acquisitions such as Hugging Face.

Huang also regularly meets with startup founders, investors and executives of private equity-backed companies to discuss how they or their companies use Nvidia products, and how Nvidia might be able to help. Such informal canvassing is a technique that other executives, such as Microsoft CEO Satya Nadella, have also used to stay abreast of the technological zeitgeist.

In the last few months, Huang has had to deal with a new reality that threatened to put a damper on Nvidia’s dealmaking ambitions: The possibility that even the chipmaker might be reaching the upper limit of how much it can comfortably spend on investments, acquisitions and backstopping of big projects.

The financing for a massive data center project developed by SoftBank’s SB Energy on federally controlled land in Ohio illustrated some of those fears. Earlier in the summer, SoftBank and Nvidia first discussed the possibility of the chip giant providing a whopping $250 billion in credit support on behalf of OpenAI, which planned to lease the data center to run and train its models on Nvidia chips. (Nvidia also invested $30 billion in OpenAI’s recent fundraise; the last $10 billion transacted on October 1.)

Nvidia’s credit default swap spreads widened in August, reflecting some investors’ worries that it was taking on too much risk. Huang took notice, frequently asking colleagues about the spreads.

Nvidia ended up guaranteeing only $105 billion worth of credit support for the first half of the project—still huge but less than half of the initial size discussed. The project was divided into phases, buying Nvidia more time, probably years, to decide whether it wants to support the second phase.

At the same time, Nvidia went ahead with another investment, agreeing to invest $3 billion in SB Energy leading up to and during its IPO, The Information first reported.

Huang has stressed that in order for AI to keep advancing, more companies need to get involved in financing the creation of chips, data centers and power. In early August, he pulled together half a dozen Wall Street firms, including Blackstone, Apollo Global Management and Goldman Sachs, to finance $500 billion of its hardware. Nvidia said it might provide a backstop for up to 25% of the total financing in some related deals.

But Huang has downplayed the risks that big-pocketed spenders like his own company are going overboard.

“People are starting to recognize that wherever I invest, it’s not a bad place to invest because I’m an informed investor. I’m not taking any risks. We’re not smart like you guys,” Huang told the crowd at the Goldman Sachs annual tech conference the bank hosted in San Francisco this September. “I need a sure thing.”