In early August, the king of Bhutan, Jigme Khesar Namgyel Wangchuck, traveled from the mountains of his landlocked Asian country to the headquarters of Nvidia, a maker of artificial intelligence chips in the flatlands of Silicon Valley.
King Wangchuck did a two-hour tour and listened as Jay Puri, Nvidia’s head of global business, discussed how Bhutanese investment in data centers and Nvidia chips could combine with the kingdom’s biggest natural resource, hydropower, to create new A.I. systems.
The pitch was one of dozens that Nvidia has made over the past two years to kings, presidents, sheikhs and government ministers. Many of those countries went on to pour billions of dollars into government efforts to build supercomputers or generative A.I. systems, hoping to gain a competitive foothold in what could be the century’s defining technology.
But in Washington, officials worry that Nvidia’s global sales spree could empower adversaries. Now the Biden administration is working on rules that would tighten control over A.I. chip sales and turn them into a diplomatic tool.
The proposed framework would allow U.S. allies to make unfettered purchases, adversaries would be blocked entirely, and other nations would receive quotas based on their alignment with U.S. strategic goals, according to four people familiar with the proposed restrictions, who did not have permission to speak publicly about them.
The restrictions would threaten an international expansion plan that Nvidia’s chief executive, Jensen Huang, calls “sovereign A.I.” Mr. Huang has hopscotched the globe this fall, logging over 30,000 miles in three months, and the company expects to make more than $10 billion in sales this year from countries outside the United States.
Now, interested buyers like Saudi Arabia, Malaysia and Bhutan could be caught in the contest for A.I. supremacy between the United States and China. Nvidia chip purchases may require cooperation with approved American and European cloud service operators and other assurances to the U.S. government that the technology won’t be shared with China.
Mr. Huang has kicked off a last-minute lobbying effort to defang the rules, but it may be too late.
“Every administration has made it very clear that unless the Chinese government changes its posture and the way it operates on the global stage, technology competition between the U.S. and China will continue,” said Klon Kitchen, a nonresident senior fellow at the American Enterprise Institute who focuses on national security and technology. “Companies like Nvidia should anticipate a tightening of the screws.”
‘This Tsunami of Interest’
In 2019, Sasha Ostojic, a former Nvidia executive from what is now Serbia, learned that his home country was considering A.I. chips from the Chinese company Huawei to support domestic start-ups. He encouraged Serbian officials to visit Nvidia first.
Serbian leaders left Nvidia’s headquarters persuaded that buying four Nvidia supercomputers, which cost about $150,000 each, would help accelerate their tech industry.
Some Serbians questioned the investment, said Stefan Badza, an adviser to the president of Serbia’s Parliament. But after the computers were received in 2021, Serbia’s start-up businesses swelled to 800, from 200.
The release of ChatGPT by OpenAI in late 2022 made Serbia’s move look wise. An overnight sensation powered by artificial intelligence, the chatbot created a stampede of interest in A.I. chips from Nvidia, which controls 90 percent of that market. Part of the reason countries clamored to get the chips, even as a supply shortage made purchases difficult: ChatGPT was unavailable in more than two dozen of them, including Saudi Arabia and Vietnam. If those places wanted A.I. technology, it was apparent they’d have to build it on their own.
“The unintended consequence rippled through the community that this is a really important technology that you probably don’t want to depend on a foreign country for,” said Keith Strier, who once worked for Nvidia and is now senior vice president of global A.I. markets at AMD, a rival chipmaker. “It triggered this tsunami of interest.”
In capitals across the world, there was a deepening conviction that A.I. should be treated as a utility. Its perceived importance made officials fear that they must spend money or risk falling behind rivals economically and militarily. And many wanted the added security of keeping data inside national borders, where they could guard sensitive information.
Nvidia responded to countries’ growing interest by building a division to work with governments, Mr. Puri said in an interview. He and Mr. Strier would meet with foreign leaders to discuss their computing needs, and Mr. Huang would whip up enthusiasm for investments in data centers filled with Nvidia’s chips.
The purchases have transformed the way governments approach computing. Until recently, smaller nations largely rented computing power from cloud computing giants like Microsoft and Google. It spared them the expense of paying for software updates, engineering support and semiconductor advancements. Now they’re taking on that work themselves.
Denmark was among the stampede of countries to contact Nvidia. Its leaders had first spoken to Mr. Huang in 2019, but had felt no urgency to invest, said Peder Lundquist, the chief executive of EIFO, the Export and Investment Fund of Denmark. After using ChatGPT, he said, Danish leaders understood the power of the technology and reached out to Nvidia to buy its chips.
EIFO and the Novo Nordisk Foundation, a Danish nonprofit focused on research, teamed up to invest $100 million on a supercomputer. Unveiled in October after six months of construction, it featured 1,528 A.I. chips and weighed more than 30 tons.
“We need to innovate and have a strong connection between applied research and businesses,” Mr. Lundquist said. “To excel, supercomputing power is incredibly important.”
A.I. Diplomacy
As countries lined up to buy Nvidia’s chips, the Biden administration passed rules to keep tabs on the company’s sales. It worried that nations with ties to China, which has set a goal of becoming the world’s A.I. leader by 2030, might provide Chinese researchers and companies with access to Nvidia chips, so it required that some countries get a license for purchases.
The United Arab Emirates was one of the first countries to seek a license. In 2022, it earmarked $100 billion for A.I. investments and wanted to build the region’s fastest supercomputer.
U.S. officials saw an opportunity. For almost a decade, they had struggled to persuade countries to stop using Huawei, even after warning of cybersecurity risks and the company’s ties to the Chinese Communist Party.
Licensing requirements offered a new solution in the form of Nvidia chips.
As part of an agreement that would give the use of a large number of Nvidia’s cutting-edge chips, the Emirates’ leading A.I. firm, G42, promised to relinquish its use of Huawei technology. In October, it unveiled a supercomputer filled with $30 million in Nvidia chips.
Over the summer, White House national security and Commerce Department officials began telling chip industry executives that they were developing a new A.I. policy, two people familiar with the discussions said. Administration officials said the current licensing process for the Emirates and other countries was cumbersome. They wanted to introduce rules that made it easier to build A.I. data centers abroad, while improving their ability to track chips and guard against their diversion to China.
They also wanted to make it easier for a country like the Emirates to buy Nvidia chips by replacing the licensing process with a bulk approval system that permitted sales to countries up to a preset limit.
This month, a draft of the rules began to circulate among chip lobbyists. The restrictions would provide speedier approval for data centers built by cloud computing providers from the United States and Europe, with a cap of hundreds of thousands of chips. But the process would be slower, with more scrutiny, for companies from other countries.
Mr. Huang and Tim Teter, Nvidia’s general counsel, called Biden administration officials to say the rules would hurt the company, three people familiar with the calls said. Others in the industry voiced similar concerns.
In a statement, Ken Brown, an Nvidia spokesman, said, “We fully support national security and are happy to provide any information the government needs to inform policy.” He added, “Caps on mainstream data center products would be a major shift in approach that wouldn’t reduce the risk of misuse but would threaten economic growth.”
The Commerce Department declined to comment.
The rules are expected to be unveiled in the coming weeks, the people said, and Mr. Huang has reached out to people close to the incoming Trump administration in the hope that they might roll back the rules in January.
How the Trump administration would handle this policy is unclear, said Gregory C. Allen, director of the Wadhwani A.I. Center at the Center for Strategic and International Studies. But Republican leaders in the House and Senate have attacked past Biden administration restrictions on chip technology and China as too weak.
“We’re in an A.I. race with China, and we don’t want American companies giving Gatorade and Nike shoes to opponents in that race,” Mr. Allen said. “We want them on our side and on our side exclusively.”
The post Nvidia’s Booming A.I. Business Collides With U.S.-Chinese Tensions appeared first on New York Times.
