China Hawks Discover the Cobra Effect
Vital Export Controls Are Needlessly Damaging American Progress
During the late nineteenth century, the Indian city of Delhi suffered from a plague of deadly wild cobras. British colonial technocrats responded with a market-based solution: they offered a bounty on dead snakes.
It worked. Locals hunted down cobras, the streets became safer, and deaths from neurotoxic cobra venom dropped. It worked so well that enterprising hunters began to breed cobras to collect the reward. When it became clear that farmers were gaming the bounty, the British understandably shut down the program. But once their snakes had no value, breeders turned them loose in large numbers. What seemed a reasonable solution produced an absurd and dangerous outcome: Delhi had more wild cobras after the program ended than before it began.
Some economists now refer to this parable about how market interventions can backfire as a “Cobra Effect”. It’s one of those tidy narratives that’s become shorthand for perverse incentives in public policy, regulatory failure, and algorithmic design.
Because human decisions commonly have third, fourth, or tenth-order unintended consequences, Cobra Effects are everywhere. To reduce crime, corruption, and health problems, the US outlawed alcohol in 1920. But Prohibition fueled the growth of organized crime, bootlegging, corruption, alcohol-related violence, and homebrew poisonings. To ensure affordable housing, many cities control rents. But rent control often leads to housing shortages, poor property maintenance, reduced incentives to build new housing, and higher rents in uncontrolled areas.
Cobra Effects remind us that incentives operate not in theoretical markets but in a messy world full of resourceful people. Those who propose clever ways to motivate people with money are often humbled when those people respond with surprising creativity instead of textbook rationality.
Turning Customers Into Competitors
For good reasons, both the Biden and Trump administrations have sought to curtail China’s emergence as a rival technology superpower. Starting in 2022, the Commerce Department issued a series of sweeping export control rules that restricted the sale of advanced semiconductors, semiconductor manufacturing equipment, and AI technologies to China.
By restricting access to American technology, however, China hawks in Congress and Commerce turned China from a profitable customer into a very determined competitor. US restrictions spurred China to undertake massive investments in its semiconductor supply chain. Last year, China’s leading chipmaker, SMIC, grew 27%.
China is now advancing its chipmaking faster than anyone thought possible before the controls. Semiconductor manufacturers measure progress by how fast they reduce the size of lithographic nodes used in chip fabrication. Smaller nodes produce faster chips. By 2023, SMIC was making 7-nanometer (trailing edge) chips. It is piloting 5nm (highly competitive) chips this year, and plans to prototype a 3nm (state-of-the-art) chip next year.
Chinese chips are not as good as those made with extreme ultraviolet lithography by TSMC, the world’s leading chipmaker. But it’s the trend that matters because learning effects drive semiconductor yields, meaning that with enough volume, SMIC will continue to improve rapidly. US chip controls have guaranteed that SMIC will have the volume it needs to make rapid progress.1 This is not the outcome that the China hawks had in mind.
The problem is not confined to semiconductors. Thanks in part to advances in their leading-edge chips, many AI experts now argue that Chinese AI is at parity or ahead of US efforts.2 Yet more snakes.
Cobra Effects extend beyond hardware and software. Chinese companies had come to rely on US industrial standards, which created a significant advantage for American suppliers. By cutting off China, the hawks have compelled China to develop competing standards and have enabled entire nations to consider shifting to Chinese semiconductors.
US semiconductor export controls were strongly reminiscent of 2017 restrictions placed by the US on Huawei, the Chinese maker of telecommunications equipment and consumer electronics. The US accused Huawei of building backdoors into its telecom equipment. China hawks boasted that sanctions on Huawei would destroy the company and slow the growth of the Chinese semiconductor industry.
The result? More cobras. Before the 2017 campaign, Huawei booked revenues of $92 billion. Last year, revenues were over $118 billion, and the company grew more than 20 percent annually.3
Giving Up a Lead in Graphene Semiconductors
Semiconductors are among the world’s most critical technologies because they progress rapidly and because advanced chips drive progress in almost every industry. The United States very understandably wishes to control access to advanced semiconductor technologies by its main geopolitical rival (just as it is understandable that British colonial administrators wished to eradicate cobras). One of the most critical emerging semiconductor technologies is chips built using graphene instead of silicon. As with any future technology, enormous hype surrounds graphene chips, but many experts and policymakers believe that they can be transformative.4
In January, the research consortium of Georgia Tech and the Tianjin International Center for Nanoparticles and Nanosystems announced that they had built the world’s first functional graphene-based semiconductor. Scientists heralded this announcement as one of the most significant scientific breakthroughs of the century.
But in May, the House Select Committee on the CCP, led by Representative John Moolenaar (R-MI), sent a letter to the president of Georgia Tech targeting the collaboration. The letter reminded Georgia Tech that Tianjin University has remained on the Commerce Department’s Entity List since 2020, and cannot receive US exports or technology as a result. It highlighted Georgia Tech’s use of its Defense‑sponsored research institute to fund this sensitive research with a PLA-linked Chinese institution. The committee demanded full disclosure on joint affiliations, funding, security vetting, and export compliance.
Under congressional pressure—and fearing loss of federal support—Georgia Tech immediately severed ties with Tianjin University and closed the joint research programs at both Tianjin and Shenzhen institutes. This decision effectively prevented any future US investment, research collaboration, or technology transfer related to their graphene semiconductor platform. With Georgia Tech cut off, no US-based funding could continue downstream or support scaling efforts. The decision deterred US agencies and private investors from investing in graphene technology. Collaborative follow-on studies or prototypes between US labs and Tianjin counterparts could no longer proceed.
Of course, China faced no such restrictions. Although Georgia Tech co-owned some intellectual property, its withdrawal paved the way for Chinese companies to further develop the technology. Chinese venture capital firms immediately invested several hundred million dollars to attempt to commercialize the technology in a matter of months. According to analyst Richard Vigilante,
“The hawks have deprived US semiconductor firms of many tens of billions of dollars in revenue, investment, and revolutionary intellectual property, and have driven Chinese firms forward dramatically.”The US is now at risk of ceding graphene semiconductors to China by threatening to prosecute investors who collaborate with Chinese researchers. Short-term Congressional posturing has directly undermined US scientific leadership.
Policymakers cannot avoid Cobra Effects, but we can foresee the possibility that specific approaches to restricting China will instead accelerate their rise. Completely shutting down academic and commercial channels with China may unavoidably fragment the global R&D ecosystem. But doing it in a way that slows progress in the US is an own goal worthy of Delhi cobra bounties.
Nuanced Adversity
Export controls are a relatively minor aspect of the Trump administration’s China strategy, which appears to be tailor-made for Beijing. Instead of supporting China’s neighbors, Trump just imposed 25 percent tariffs on India and threatened one hundred percent “secondary tariffs.” It slapped Taiwan with tariffs and blocked its president from visiting New York to appease Xi Jinping. He is surrendering the global south to China by destroying USAID and conducting a tariff war on the poorest countries of the world. He has doubled down on fossil fuels and surrendered progress on clean energy to China, which is now adding more renewable energy than the rest of the world combined.
Rather than surrendering to China or relying heavily on containment policies that consistently trigger Cobra Effects, the United States should adopt a more nuanced, consistent, and forward-looking approach toward China.
Coopetition. Policymakers should embrace strategic "coopetition," distinguishing clearly between critical national security risks and areas of mutual scientific or economic interest. Fields such as graphene semiconductors, where both countries have investments, may be prime candidates for managed collaboration. Technologies like wafer-scale computing, where a single US firm dominates, may not be.5 The US would establish clear, transparent guidelines for joint research that safeguard intellectual property without strangling innovation. Establishing dedicated frameworks for "safe" collaborative zones could sustain American technological leadership while reducing the risk of unintended technological isolation.
Invest. America should dramatically increase investment in domestic innovation ecosystems, particularly through public-private partnerships focused on next-generation materials like graphene. Policies that strengthen homegrown R&D, support startups, and create incentives for commercialization can position the US competitively without fueling China’s rapid advance.
Coordinate with allies. Instead of provoking trade wars with allies, the US should rally allies and partners around common international standards, regulations, and collaborative research protocols to establish global benchmarks. A unified international front makes unilateral Chinese standards less attractive to neutral third-party nations. Insulting India, Canada, and Vietnam is just a massively bad idea.
Build soft power. Expanding cultural and academic exchanges, rather than broadly restricting them, helps sustain American soft power and influence among China's intellectual elite and younger generations. Anyone involved in higher education understands that these exchanges promote deeper mutual understanding, foster personal connections, and can subtly encourage alignment with Western standards and values.
In short, positive-sum economic competition and political cooperation strengthen American leadership, reinforce our values, and reduce Cobra Effects. Trump’s trademark zero-sum confrontations do the opposite and will end badly for the United States. We should respond to China’s rise with disciplined technological vigilance, targeted and sustained investment, coordinated diplomacy, and proactive cultural engagement, not blunt decoupling or indiscriminate tariffs.
Musical Coda
As a college student more than a decade ago, Joshua Lee Turner recorded the Dire Straits classic Sultans of Swing. The recording has a cult following and has been viewed more than 13 million times.
These export controls hurt Nvidia, now the most valuable company ever created. Starting in 2022, the US restricted powerful Nvidia chips like the H100 and A100 from export to China. The company responded by designing a custom chip to comply with US export restrictions. Starting this past April, Nvidia sold the dumbed-down H20 to China, but Trump imposed new controls banning these chips as well. This left Nvidia with hundreds of millions of dollars in unsold H20 inventory and forced the company to take a $4.5 billion write-off.
Nvidia’s CEO, Jensen Huang, began to lobby Trump and other US policymakers amid trade talks between Washington and Beijing over China’s rare earth exports and US restrictions on technology exports. A month ago, the US government announced it would permit Nvidia to resume sales of the H20 chip to China. Trump lifted the H20 ban without explaining whether the Chinese acquisition of H20 chips had ever been a genuine threat to US national security.
Leading examples:
Harvard’s Belfer Center (notably Graham Allison and Eric Schmidt) argue that China now stands as a "full-spectrum peer competitor" in both commercial and national security AI, and in some application domains, “has already overtaken the United States to become the world’s undisputed No. 1.” They point to indicators like research output, talent, patents, AI adoption, and China’s national prioritization of AI as evidence that the US can no longer assume technological leadership is unassailable.
White House AI & Crypto Czar David Sacks has stated publicly that the US may be only “three to six months ahead of China in AI model development,” citing the rapid rise of Chinese models like DeepSeek, which have reached parity or near-parity with top American models on key benchmarks and daily user adoption.
Stanford University’s 2025 AI Index finds that US-led models still dominate in terms of both numbers and prestige, but Chinese models are catching up. DeepSeek’s models (R1 and V3) have posted results on par with or even surpassing their US counterparts in some open benchmarks, sparking significant concern and debate among US tech policy circles.
Cameron Johnson, a partner at the consulting firm Tidalwave Solutions based in Shanghai, who spends time inside Chinese manufacturing companies, stated on a recent Odd Lots podcast that China has surpassed the US in its deployment of AI.
Alongside Nvidia, Huawei is one of the most resilient technology companies on the planet. Ironically, both companies have now been accused of creating “backdoors” in their exported chips. This week, the Cyberspace Administration of China launched a probe into potential security risks in H20 chip, which was partly in response to US Congressional hawks demanding that the US install tracking or security features on exported chips. Both Huawei and Nvidia have publicly denied the existence of “backdoors” or tracking functions in their chips.
The essential podcast Acquired does an excellent job of distilling the histories of these two remarkable companies. Find Nvidia part one here and Huawei here.
Graphene chips are a big deal because silicon chips are approaching fundamental physical limits that threaten the continuation of Moore’s Law and traditional semiconductor scaling. Graphene semiconductors have better electrical, thermal, and mechanical properties than silicon. They support faster switching speeds and lower power consumption because electrons move through graphene with lower resistance. This enables them to dissipate heat better and scale up larger. Other technologies like transition metal dichalcogenides—TMDs are entering scaling discussions and may emerge as potential successors in post-Moore logic and photonics. Regardless of which technology prevails, poorly structured export controls will slow their development.
Wafer-scale computing is an advanced hardware approach where, instead of slicing a silicon wafer into hundreds of individual chips, manufacturers build a single, massive chip using nearly the entire wafer. Standard GPU or CPU semiconductors are cut from wafers the size of dinner plates into chips the size of postage stamps. Wafer-scale chips use the whole plate, which allows chipmakers to integrate millions of processing cores and a massive amount of memory onto a single, unified piece of silicon.
Wafer-scale chips are revolutionizing workloads in artificial intelligence (especially large language models), scientific computing, and scientific simulations by offering speed, energy efficiency, and scalability that were previously impossible using conventional multi-chip or GPU-based systems.
The first and only company in the world building AI hardware at wafer-scale is Cerebras. Their big chips hold the world's speed record in AI inference, which is likely to drive AI compute loads far more than training over the long term. Richard Vigilante argues that Congressional interference delayed the Cerebras IPO, but the company has secured a great deal of private financing despite the hawks. It may never need to go public.