The Justice Department’s antitrust division has already contacted Nvidia, whose AI chip market share is estimated at over 80%, to ask questions about the terms of its contracts and partnerships, according to people familiar with the matter.

The investigation is at an early phase, and government attorneys haven’t issued a subpoena to Nvidia for internal documents, the people said. The company said last week that it hadn’t received a subpoena, after a media report said that it had. The department could issue a subpoena in the coming months if it determines a fuller-fledged investigation is needed, the people said.

The Federal Trade Commission, which shares antitrust authority with the Justice Department, this year launched its own effort to examine investments that Microsoft, Amazon.com and Google’s owner, Alphabet, have made in the most compelling AI startups. The FTC’s review, which is ongoing, is looking into whether the tech giants received any special advantages over competitors by putting billions of dollars into the companies that are creating the most popular generative-AI products.

The eagerness of the Justice Department and FTC to police the AI space in its nascent phase stands in contrast to the approach the agencies took with Google, Facebook and other companies before they became the internet titans of today.

The feds previously were shy about intervening in fast-developing tech markets whose products were popular with consumers. Liberal and conservative critics have argued that the hands-off approach of the past two decades—across administrations—allowed a handful of tech giants to amass too much power. The backlash helped spur a shift toward the aggressive antitrust enforcement the Biden administration is embracing.

The Justice Department “would say, ‘We’ve watched enough Big Tech to know that we want to kick the tires,’” said Randal Picker, a law professor at the University of Chicago. “You are allowed to get market power, you’re allowed to have it. The question is always, how do you exercise it?”

A handful of the biggest technology companies are furiously jockeying for market position in AI, which requires huge capital investments that smaller competitors can’t provide. There could be a tipping point, according to antitrust enforcers, when incumbents gain control over critical aspects of AI—similar to how internet search and social media developed.

FTC Chair Lina Khan has worried that dominant tech companies could obtain new monopolies in the growing field of generative AI, systems that have humanlike abilities to converse, create media, write computer code and more.

“There are very reasonable concerns now that the critical inputs into this technology may already be controlled by a handful of companies,” Khan told the venture-capital firm and startup adviser Y Combinator last year. “We may already be seeing bottlenecks that can impede innovation and impede competition.”

Jonathan Kanter, the Justice Department’s antitrust chief, warned in May that competitive dynamics in the AI market are similar to those in other tech industries in which a single company emerged as dominant. “AI relies on massive amounts of data and computing power, which can give already-dominant firms a substantial advantage,” Kanter said in a speech at Stanford University. “Powerful network effects may enable dominant firms to control these new markets.”

Nvidia’s chips are essential to training the AI models developed by OpenAI and others. The department’s investigation comes as the company faces growing antitrust scrutiny across the globe, from the European Union and the U.K. to China and South Korea. Its French offices were raided last year by competition authorities. Nvidia said in an August securities filing that it has received requests from different enforcers for information about its sales, how it allocates its limited chip supplies and partnerships with AI companies.

Nvidia’s high market share isn’t itself illegal. The company’s growth exploded after customers found that its specialized chips, once used mostly for computer gaming graphics and then by computers that run cryptocurrency transactions, were particularly well-suited for the computational demands of AI.

To pursue an antitrust claim, the government would have to find instances in which Nvidia took steps, such as through restrictive contract terms, that harmed competition and artificially preserved or expanded its market power.

Some of Nvidia’s rivals complain that it has used its outsize market share in AI computing to its advantage.

Rodrigo Liang, the chief executive of the AI-chip startup SambaNova Systems, said that competition was growing and that there was strong demand for alternatives to Nvidia. But he added that there was a fear among some customers that Nvidia could withhold selling its AI chips, known as GPUs, to customers who explore other alternatives such as his.

“It’s more of a commercial concern tied to, if the world is supply-constrained on GPUs, do I want to take the risk of not having supply?” he said.

An Nvidia spokeswoman said the company doesn’t require exclusivity and supports all of its customers and partners.

“Our customers know that they are free to choose any combination of suppliers they deem best, and we’ll work with them no matter what they decide,” she said.

Because Nvidia’s chips are so heavily used in AI development, decisions about who gets them can have an impact on which companies succeed. Nvidia’s most advanced AI chips have long been supply-constrained, with customers sometimes waiting months for orders to be fulfilled. The company has said demand will exceed supply for its next-generation chips, code-named Blackwell and set to start shipping in the coming months.

In a statement, Nvidia said its market position is due to the high AI computing performance of its products and said it was happy to answer questions from the government about how it conducts its business.

“Nvidia wins on merit, as reflected in our benchmark results and value to customers, and customers can choose whatever solution is best for them,” the company said.

Aside from scrutinizing companies’ business practices, antitrust enforcers can challenge mergers that remove too much competition from a market. The government was slow over the years to question deals that created today’s technology giants. The FTC, for instance, allowed Facebook to acquire Instagram in 2012.

The Justice Department is investigating Nvidia’s agreement in April to buy Run:ai, a startup that offers software that organizes and streamlines the use of clusters of AI chips to make them more efficient.

The department began examining Nvidia’s deal for Run:ai after it reached a power-sharing agreement with the FTC over how the agencies should divide oversight of AI companies, people familiar with the matter said. Politico earlier reported the investigation of the Run:ai deal.

The agencies’ efforts to investigate AI companies could stretch for months or years. Merger investigations can take as long as a year, and investigations of a company’s broader conduct often stretch beyond that.

Write to Dave Michaels at [email protected] and Asa Fitch at [email protected]