Nvidia confirmed Tuesday it must seek specific U.S. government approval to sell certain advanced AIchips to China, a policy tightening by the Trump administration that prompted the company to announce a $5.5 billion charge and triggered declines across Asian technology stocks.
Citing national and economic security concerns, the Commerce Department stated the new export licensing requirements, affecting Nvidia’s H20 processor, AMD’s MI308 chip, and equivalents, would be in effect indefinitely. The move represents the administration’s first major expansion of semiconductor sales limits abroad according to the New York Times, intensifying the tech rivalry between the two nations and immediately impacting stock prices.
In a regulatory filing made Tuesday, April 15, the American chip giant detailed the financial impact, stating the $5.5 billion charge against current quarter revenue stems from H20 chip inventory, purchase commitments, and related reserves it can no longer sell or fulfill under the updated rules.
News of the restrictions sent Nvidia’s shares down nearly 7% in after-hours trading Tuesday, with competitor AMD seeing a 7.6% drop. The repercussions spread through the supply chain in Wednesday trading across Asia: Taiwan Semiconductor Manufacturing Corporation shares fell up to 5%, and memory maker SK Hynix dropped 3.9%. Broader market indices in Hong Kong (-1.9%), Taiwan (-2%), Japan (-1%), and South Korea also retreated as investors digested the renewed trade friction and concerns over a potential recession spurred by trade wars, sentiments echoed in a Bank of America survey showing recent investor pullback from US stocks.
Policy Whiplash Follows Reprieve Reports
This definitive restriction arrived just days after conflicting signals emerged. On April 10, reports surfaced, suggesting the administration had unexpectedly paused the implementation of planned H20 restrictions. That apparent pause reportedly followed promises from Nvidia regarding new U.S.-based AI data center investments and occurred shortly after CEO Jensen Huang attended a Mar-a-Lago dinner with President Trump the previous week. This back-and-forth highlights the complex interplay between corporate lobbying and security apparatus aims; for historical context, previous Commerce Secretary Gina Raimondo had acknowledged in late 2023 the possibility of permissible commercial sales of some dual-use AI chips within limits.
However, pressure to enact stricter controls had been building, amplified by the capabilities demonstrated by Chinese AI firm DeepSeek, allegedly using H20 chips. Senator Elizabeth Warren, in an April 14 letter to Commerce Secretary Howard Lutnick, had urged immediate action, stating, “The Commerce Department cannot further delay undertaking necessary and urgent action on the H20 to protect U.S. national security.” Responding to the earlier reports of a pause, Rep. Raja Krishnamoorthi had stated, “Export controls work, and we don’t have time to waste… Every day America fails to restrict the export of chips designed to circumvent existing controls is a day that our adversaries have to build up stockpiles to defeat us”.
Lutnick himself had previously advocated stopping Chinese companies from using American technology “to compete with us.” The Trump administration privately notified Nvidia of its final decision on Wednesday, April 9, according to the New York Times, before the company’s public filing on April 15. Confirming the action Tuesday, Commerce Department spokesman Benno Kass stated, “The Commerce Department is committed to acting on the president’s directive to safeguard our national and economic security.”
The H20 Chip: Born From Restrictions, Now Restricted
The Nvidia H20 chip exists solely because of prior U.S. efforts to limit China’s access to top-tier AI chips. After the Biden administration banned Nvidia’s A100 and H100 chips from China in October 2022, Nvidia created the less powerful A800 and H800 as compliant alternatives.
When those too were banned in late 2023, Nvidia developed the H20. Based on the same GH100 silicon as the H100, the H20 offered significantly reduced performance density – delivering around 2.9 TFLOPs per square millimeter of die size compared to the H100’s 19.4 TFLOPs/mm², according to early 2024 details. While reports suggested its FP32 performance lagged behind Huawei’s competing Ascend 910B, its interconnect speed was seen as an advantage for large AI model training clusters. Priced reportedly between $12,000-$15,000 earlier this year and drawing orders previously estimated at $16 billion, it was a sought-after, albeit downgraded, option before this latest ban.
While the $5.5 billion charge is substantial, the strategic implications for Nvidia may be larger. The company had fought to maintain a presence in China, a market that provided $17 billion in sales last year, even as its share of Nvidia’s total revenue decreased from about a fifth in fiscal 2023 to 13% last year under existing restrictions. The concern, voiced by analysts like Patrick Moorhead of Moor Insights & Strategy, is ceding the market entirely. “This kills Nvidia’s access to a key market, and they will lose traction in the country,” Moorhead stated in the NYT. “Chinese companies are just going to switch to Huawei.”
However, others believe Nvidia’s global strength provides resilience. Marc Einstein at Counterpoint Research noted that Nvidia’s China trade has “been in the crosshairs… for some time,” and suggested chip controls could become bargaining chips in broader U.S.-China trade deals. This reflects an ongoing debate, highlighted in previous industry reactions to U.S. policies. Microsoft’s Brad Smith argued earlier this year that strict controls might prove ineffective, simply pushing global customers to alternatives and possibly speeding up China’s domestic capabilities. Nvidia itself had issued similar warnings.
This latest move seems likely to fuel China’s push for semiconductor self-sufficiency, an effort already backed by initiatives like its $47.5 billion “Big Fund” and yielding results like Huawei’s domestically-processed smartphone chip and its Ascend AI chips. These Ascend chips now directly compete with Nvidia and AMD offerings and are reportedly used by firms like DeepSeek. Experts like Gregory Allen of CSIS warned earlier this year, as reported by Fortune, that “It is unrealistic to expect a lead of more than a year or two, even with extremely aggressive export controls,” regarding the US-China AI development gap.
