Jensen Huang, co-founder and CEO of Nvidia Corp., speaks during a news conference in Taipei on May 21, 2025.
I-hwa Cheng | Afp | Getty Images
Replacing Nvidia is a tall order. While Chinese competitors are years behind the company’s cutting-edge technology, many analysts and insiders warn they are catching up, thanks to U.S. export restrictions.
U.S. chip restrictions on the sale of advanced semiconductor technology, especially those used in artificial intelligence, have been rolled out over several years, with the initial aim of curbing China’s military advancement and protecting US dominance in the AI industry.
However, according to Nvidia CEO Jensen Huang, U.S. semiconductor export controls on China have been “a failure,” causing more harm to American businesses than to China.
While the goals of cutting back the Chinese military’s access to advanced U.S. technology and maintaining U.S. leadership in AI appear to have had some success on paper, loopholes and existing semiconductor stockpiles in China have complicated these aims, said Ray Wang, an independent tech and chip analyst with a focus on U.S.-China competition.
“That’s partly why we are seeing a closing of the gap between Chinese and U.S. AI capabilities,” added Wang.
A self-inflicted wound?
Leaders of Nvidia and other American chip designers have long lobbied against chip controls as they worry about losing lucrative business deals. Huang said at the annual Computex technology trade show in Taipei that Nvidia’s GPU market share in China fell to 50% from 95% over the past four years.
Indeed, chip experts say that the curbs create more harm than good for the U.S.
“The effects of the controls are twofold. They have the impact of reducing the ability of U.S. companies to access the China market and, in turn, have accelerated the efforts of the domestic industry to pursue greater innovation,” said Paul Triolo, Partner and Senior VP for China at DGA Group.
“You create competitors to your leading companies at the same time you’re cutting them off from a massive market in China,” he added.
While Washington’s most comprehensive export controls were passed during former U.S. President Joe Biden’s term in the White House, curbs on Huawei and SMIC, China’s largest chipmaker, go back to Donald Trump’s first term in office.
On April 15, Nvidia disclosed that new controls, which restricted sales of its H20 graphics processing units to China, had led to a $5.5 billion charge against its revenue.
Counter-intuitive curbs
The restrictions are expected to be a boon for the demand and development of local Nvidia alternatives like Huawei, which is working on its own AI chips. They also come against the background of Beijing mobilizing billions as part of its chip self-sufficiency campaign.
“The bottom line is, the controls have incentivized China to become self-sufficient across these supply chains in a way they never would have contemplated before,” Triolo said.
Chinese AI-related achievements, such as DeepSeek’s R1 model and news of Huawei chip progress, have led observers to question the effectiveness of chip controls.
According Wang, the independent analyst, China’s semiconductor and AI space has seen an acceleration of startups, market opportunities, and AI talent alongside the restrictions, which has clearly resulted in domestic innovations.
“I think the arguments that export controls accelerate innovation is quite valid,” Wang said.
Nivida’s Haung also noted these trends in April, telling lawmakers in Washington that the country has made enormous progress in the last several years and is right behind the U.S.
Moving goal posts?
Nvidia’s H20 chip was designed specifically to comply with existing chip controls prior to the clampdown on exports.
“We are not just talking about one export control, we are talking about a series of export controls that originate from all the way back in 2019,” said Wang, noting that the evolving policies have had a couple of different objectives.
Meanwhile, in what DGA’s Paul Trilio calls a “moving of the goalposts,” it seems that the aims of the restrictions have shifted to an intention to slow down and contain Chinese AI and semiconductor developments.
“The continued expansion of the controls, and the lack of an articulation of what the clear end game here is, has really created a lot of issues, and created a lot of collateral damage,” Trilio said, adding that it has led more people to question the policy.
In a statement earlier this month, the Information Technology & Innovation Foundation, a U.S. think tank which has received funding from various technology companies, said in a post that “the Biden administration’s export control policy for AI chips has largely been a failure since day one. Yet, year after year, it has doubled down, attempting to plug various loopholes.”
“While [the U.S. government] is certainly right to prevent U.S. companies from selling advanced AI technology to the Chinese military, cutting U.S. companies off from the entire commercial Chinese market is a cure worse than the disease,” Stephen Ezell of ITIF told CNBC in an email.
“U.S. export controls have cost NVIDIA at least $15 billion in sales, and those are revenues the company needs to be able to earn to invest in future generations of innovation.”
Brad Gerstner, Altimeter Founder and CEO, speaks at the Delivering Alpha conference in New York City on Sept. 28, 2023.
Adam Jeffery | CNBC
Investor Brad Gerstner cautioned Monday that OpenAI‘s deals with Nvidia and AMD are purely announcements, not deployments.
“Now we will see what gets delivered,” the Altimeter Capital founder told CNBC. “Ultimately, the best chips will win.”
OpenAI’s megadeal with AMD and its relentless push to expand artificial intelligence capabilities underscores the intensifying competitive landscape.
Gerstner said the deals provide “more evidence that the world will remain compute-constrained despite best efforts to bring massive supply online.”
Read more CNBC tech news
Experts say it’s also another validation of the AI arms race heating up, with AI a key element in the geopolitical race between the U.S. and China.
OpenAI’s Chinese rival DeepSeek sent shockwaves last year when it claimed to have a lower-cost AI model than its U.S. peer. And Deepseek has continued to innovate, delivering new open-sourced models using domestically made AI chips.
Last week, the U.S. government issued a report warning of DeepSeek’s national security concerns, Axios reported.
The National Institute of Standards and Technology’s Center for AI Standards and Innovation said DeepSeek provides Chinese Communist Party views more frequently than U.S. models, according to Axios.
OpenAI’s partnership with AMD is raising hopes that it is taking the right steps to increase production and build more complex AI models.
“What we’re really seeing is a world where there’s going to be absolute compute scarcity, because there’s going to be so much demand for AI services, and not just from OpenAI, really from the whole ecosystem,” OpenAI President told CNBC’s “Squawk on the Street” Monday. “And so that’s why it’s just so important for this whole industry to come together.”
The AppLovin logo arranged on a smartphone in New York, US, on Wednesday, Feb. 26, 2025.
Gabby Jones | Bloomberg | Getty Images
AppLovin shares plummeted on Monday after Bloomberg reported that the SEC has been probing the mobile advertising company over its data-collection practices.
The agency has been looking into whether the company violated agreements on pushing targeted ads to consumers, Bloomberg reported, citing people familiar with the matter. The report said that the SEC is responding to a whistleblower complained filed this year along with multiple short-seller reports, and added that neither the company nor its officials have been accused of wrongdoing.
An AppLovin spokesperson said the company doesn’t typically comment on the “existence or non-existence” of regulatory matters.
“That said, as a global public company, we regularly engage with regulators and if we get inquiries we address them in the ordinary course,” the spokesperson said in a statement. “Material developments, if any, would be disclosed through the appropriate public channels.”
The stock dropped 14% in regular trading after the report, which landed shortly before market close. It fell another 5% in extended trading.
AppLovin’s stock has been on a tear, jumping about 80% this year after soaring more than 700% in 2024. The surge has been driven by the company’s artificial intelligence technology that’s allowed it to provide better ad targeting capabilities to brands.
Last month, AppLovin was added to the S&P 500, replacing MarketAxess Holdings, at the same time that Robinhood joined the index in place of Caesars Entertainment.
AppLovin made the move into the benchmark despite a short-seller’s effort to keep it out.
In March, Fuzzy Panda Research advised the committee for the large-cap U.S. index to keep AppLovin from becoming a constituent. AppLovin shares dropped 15% in December, when the committee picked Workday to join the S&P 500.
Three notable short-seller firms, including Fuzzy Panda, have slammed AppLovin of late. The latest was Muddy Waters Research, which in March said the company’s ad tactics “systematically” violate app stores’ terms of service by “impermissibly extracting proprietary IDs from Meta, Snap, TikTok, Reddit, Google, and others.” In so doing, AppLovin is funneling targeted ads to users without their consent, Muddy Waters said.
Fuzzy Panda and Culper Research put out reports the prior month, taking aim at AppLovin’s AXON software, which drove its earnings growth and stock surge. The shares dropped 12% on Feb. 26, the day of the short reports.
After those reports were published, AppLovin CEO Adam Foroughi wrote a blog post, defending his company’s technology and practices, and taking aim at the short sellers trying to profit from AppLovin’s decline.
Figma signage appears at the New York Stock Exchange in New York as the company prepares for its shares to begin trading on July 31, 2025.
Michael Nagle | Bloomberg | Getty Images
Figma shares jumped 7% on Monday after the design software vendor’s technology was promoted by OpenAI CEO Sam Altman in an onstage demo at his company’s annual DevDay conference in San Francisco.
Altman discussed Figma’s integration into ChatGPT, which has more than 800 million monthly users. He showed how third-party applications could plug in with OpenAI’s Apps SDK, or software development framework.
“When someone’s using ChatGPT, you’ll be able to find an app by asking for it by name,” Altman said. “For example, you could sketch out a product flow for ChatGPT and then say, Figma, turn this sketch into a workable diagram. The Figma app will take over respond and complete the action.”
In addition to asking for Figma’s help by name in ChatGPT, the assistant can also suggest Figma when it’s relevant, Figma product manager Luke Zhang said in a blog post.
The rally for Figma, at its high point, was the steepest since the day of the company’s public market debut on the New York Stock Exchange in July.
Figma has been ramping up its own tools for working on app and website designs using generative AI models from OpenAI and other providers.
Subscribers to products that connect to the Apps SDK will be able to log in without leaving their ChatGPT conversations, Altman said. He said people working on products in Figma can also launch the FigJam tool to keep working on development ideas. Apps SDK is based on the Model Context Protocol, an open standard that OpenAI rival Anthropic introduced last year.
Software developers will be able to submit apps for review later in 2025, Altman said.
Over time, OpenAI will offer many ways to generate revenue through third-party integrations, Altman said. Last week, OpenAI announced a feature allowing people to buy products listed on Etsy through ChatGPT.