NVIDIA bans sales of 4090 GPU to China: Whatever stands in your way will ultimately make you stronger
TechFlow Selected TechFlow Selected
NVIDIA bans sales of 4090 GPU to China: Whatever stands in your way will ultimately make you stronger
For China, the design and manufacturing of advanced chips is no longer an option but a necessity.
Navigating Web3 tides with focused insightsAuthor | Xue Liang Neil
The shoe has finally dropped. The U.S. export control regulations that had been quietly circulating in whispers across the Chinese market for over half a month have now officially revealed their true form.
As widely rumored, the new rules further tighten chip exports to China. By shifting the evaluation criteria from last year’s bandwidth parameters to computational power, nearly all current high-performance chips are now swept into the restrictions.
Even consumer-grade gaming GPUs like NVIDIA's RTX 4090 are affected. Although its total processing performance (TPP), calculated as computing power multiplied by bus width, is under 2700, its performance density—what the U.S. Department of Commerce calls “performance per die area” (PD)—falls precisely within the threshold defined under Category 3A090b for high-performance chips. This means even sales of graphics cards to individual gamers are now restricted. According to U.S. officials quoted by Reuters: "Orders for the most powerful consumer chips must also be reported to the Department of Commerce, even if they’re intended for use in laptops."
Moreover, even chips slightly below the performance threshold require government notification prior to export.
This implies that all high-performance chips are now subject to restrictions—including AI chips specially developed for the Chinese market, such as NVIDIA’s China-specific A800 and H800, the newly launched L40S, AMD’s MI250, and Intel’s Gaudi2, which hasn’t yet been officially deployed in China.
Despite market expectations, the severity of the new regulations still came as a shock. Late on the 16th, a screenshot detailing internal updates at NVIDIA circulated through WeChat groups. In this unverified message, NVIDIA urged anyone interested in purchasing the H800 to place orders before midnight on the 17th.
A chill has spread throughout the entire AI industry.
Beneath the Precision Cut
The U.S. Department of Commerce clearly came prepared. Beyond strictly using computing power as a uniform benchmark, the new rules introduce additional hurdles, including an expanded list of restricted destination countries (nearly 20 nations), tighter licensing requirements for semiconductor manufacturing equipment, and the addition of Chinese firms such as Biren and Molthreads to the so-called "Entity List."
Furthermore, the Department will soon initiate a process to restrict China’s access to cloud computing power—especially significant given that the world’s largest cloud service providers are almost all American companies.
In short, the updated export controls operate on two fronts: preventing China from acquiring chips via third countries, and blocking China’s development of advanced chip manufacturing capabilities.
This irrational and hysterical policy reflects America’s near-pathological attitude toward China. During the formulation of this policy, U.S. media outlets were flooded with biased reports about China’s semiconductor industry, dominating headlines. Commerce Secretary Gina Raimondo described Huawei’s new smartphone release (which occurred during her visit to China) as making her “deeply uncomfortable.”
These farcical episodes seem designed to lay the groundwork for such policies. Although the new rules have a 30-day public comment period, no changes are expected. Raimondo emphasized that these controls will be updated at least annually, signaling Washington’s determination to extend its long-arm jurisdiction indefinitely.
Industry and political circles offer divergent interpretations of this unusually strict regulation.
The Semiconductor Industry Association (SIA), representing 99% of U.S. chip companies, stated in a public release that “overly broad unilateral controls harm the U.S. semiconductor ecosystem.” In plain terms, SIA believes the government is overreaching. While “national security” has become the new political correctness—under which anyone can peddle self-interest—SIA warns that such stringent measures may push overseas customers to “look elsewhere.”
Yet some in Congress believe the policy doesn’t go far enough. Michael McCaul, Chairman of the House Foreign Affairs Committee, took time out from hearings on the Israel-Hamas conflict to issue a statement urging to “remain vigilant” against attempts by Chinese entities to circumvent these already unjustified restrictions.
Some analysts familiar with the semiconductor industry note that the export restrictions lack provisions on technology transfer and coordination with third countries, creating a major loophole.
Clearly, U.S. policymakers and their supporters are seriously concerned that not every country shares America’s pathological paranoia.
The Collective Failure of NVIDIA and Others
If opinions on the policy itself remain divided, one thing is certain and universally acknowledged: the failure of semiconductor companies in lobbying efforts.
Clearly, neither NVIDIA nor any other company managed to influence the Department of Commerce’s decision-making from a constructive standpoint. In many ways, they were caught completely off guard.
Take Intel’s Gaudi2 chip, for example. In early July, Intel held a high-profile launch event in Beijing. At the time, Sandra Rivera, Intel’s Executive Vice President overseeing AI, stated that Gaudi2 complied with U.S. export control policies. Compliance was achieved simply by reducing the number of internal Ethernet interfaces from 24 to 21—an adjustment that left the chip’s computing power identical between the Chinese and international versions.
The new export controls are likely to disrupt Intel’s entire rollout plan, further worsening its already weak financial performance.
NVIDIA, once riding high on the generative AI wave, faces a similar fate. Revenue from China has consistently accounted for roughly one-quarter of its total income. Now, this segment faces great uncertainty. In regulatory filings submitted to the U.S. Securities and Exchange Commission, NVIDIA outlined how these policies affect its sales—including restrictions on nearly all popular chips. The company noted specifically that while it can assist customers in applying for export licenses, it cannot guarantee approval or timely processing.
The semiconductor industry’s weak position in shaping this policy is now unmistakable. This is surprising, considering that chipmakers, much like internet giants, are major players in political lobbying in the United States. Government incentives and funding under the CHIPS Act largely flow to the largest few chip companies—a fact often cited as evidence of their influence on Capitol Hill.
Yet that influence appears to have vanished entirely in the realm of export controls. Upon news of the Commerce Department’s announcement, NVIDIA’s stock plunged 8%, reflecting widespread pessimism.
As one insider put it: Only the U.S. government could make an arrogant giant like NVIDIA lose so badly.
The Panopticon and Weaponized Interdependence
The division of labor in the semiconductor industry once epitomized globalized economics, and such deep interconnections among global economies were believed to benefit all.
But globalization did not make the world flatter. Instead, the specialization of supply chains ultimately centralized the free movement of production factors, with the United States leveraging its economic and technological strength to dominate this centralization.
Now, the U.S. is systematically abusing this dominance under the banner of national security. Georgetown University political scientist Abraham L. Newman describes this abuse as "weaponized interdependence"—a term that captures how interconnectedness itself becomes a strategic weapon.
Vast and complex global supply chains, where everything depends on everything else, can be destabilized by just one or two chokepoints. In other words, whoever controls these critical nodes—and leverages them strategically under an expansive definition of national security—holds near-absolute power.
When introducing the concept of “weaponized interdependence,” Newman draws a parallel to British utilitarian philosopher Jeremy Bentham’s Panopticon effect. From the central tower, guards can observe all prisoners without being seen, ensuring control and dominance—a tactic long employed by the United States.
After 9/11, counterterrorism became the centerpiece of national security. Former NSA Director Michael Hayden openly advocated exploiting America’s so-called “home-field advantage” due to its dominance in electronic communications. FBI Director Chris Wray echoed similar concerns about foreign entities gaining strategic positions in telecom networks—precisely the logic used to justify the comprehensive blacklisting of Huawei.
Today, the new battlefield of “national security” has shifted to advanced semiconductor technology.
Final Thoughts
It is said that leading Chinese AI and large model developers have already placed orders exceeding $5 billion worth of related chips. The fate of these orders remains uncertain. In the aforementioned internal NVIDIA update screenshot, the 30-day review period is seen as a grace window to accelerate chip deliveries to the Chinese market.
Of course, these are merely stopgap measures. The fundamental solution lies in comprehensively enhancing domestic chip capabilities. Indeed, this is exactly what corporate representatives—including those from NVIDIA—and the SIA have repeatedly stressed: blocking China will leave it with no choice but to strengthen its own semiconductor capabilities.
For China, designing and manufacturing advanced chips is no longer optional—it is mandatory.
Among currently available domestic alternatives for AI computing power, Huawei’s Ascend ecosystem shows some competitiveness. Many Chinese companies involved in large model training and inference rely on its solutions to varying degrees. Some even design algorithms to support multiple hardware platforms from the outset, aiming to mitigate geopolitical risks. Yet overall, domestic chip capabilities still fall far short of meeting the market’s voracious demand for computing power.
The burgeoning race among hundreds of large models demonstrates China’s strong presence in the field—an irreversible trend in China’s technological advancement. The chip bottleneck may slow us down, but it will never extinguish the spark of artificial intelligence.
Whatever stands in your way will ultimately make you stronger.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
