By taking Taiwan, China could erase America’s best hope of keeping its AI edge.
For decades, a fragile equilibrium has held the peace in the Taiwan Strait. Despite China’s conviction that the self-governed island must one day be united with the mainland, Beijing has held to a policy of “peaceful reunification” built upon threats and coercion, but not full-scale war. Key to this has been the specter of the U.S. military coming to Taiwan’s defense, as well as Beijing’s belief that the Taiwanese public may eventually be peacefully compelled to accept a political settlement.
These dynamics have shifted over the past several years. The Taiwanese public increasingly rejects unification, narrowing the possibility that a conciliatory government emerges in Taipei. China’s breakneck military modernization has tilted the balance of power across the strait toward Beijing; Chinese President Xi Jinping has ordered the People’s Liberation Army (PLA) to be prepared to invade Taiwan by 2027. And political dysfunction in Washington, coupled with public fatigue over military adventurism abroad, has raised doubts about the United States’s willingness to risk war with China to defend the island.
Now, Taiwan is at the center of another brewing geopolitical storm. The United States and China are locked in a competition for artificial intelligence (AI) supremacy. Many believe that whichever country develops superior technology could gain decisive economic and military advantages over the other. AI is built on a complex stack of technologies, the most critical of which are the advanced semiconductors needed to train and run the world’s leading AI models.
Virtually all of these chips are produced in Taiwan.
For years, Taiwan’s dominance in global chip production has been thought to provide the island a degree of protection—a “silicon shield” discouraging Chinese military action that could disrupt crucial technological supply chains. But as the United States has increasingly cut off China’s access to Taiwan’s leading-edge semiconductors through export controls, this dynamic could now be shifting.
Most analysts agree that the United States holds a narrow lead over China in developing cutting-edge AI models, but this lead is largely built upon its access to superior Taiwan-made chips. As the AI competition intensifies and U.S.-China relations continue to deteriorate, fears that the United States is on the verge of gaining an insurmountable AI-enabled advantage could create a dangerous incentive for China to accelerate its long-standing efforts to bring Taiwan under its control and stanch the flow of leading-edge chips to Washington.
While political considerations will remain the primary factor shaping China’s Taiwan calculus, Beijing may view going on the offensive as its best shot at accomplishing two strategic objectives at once: achieving “reunification” and leveling the AI playing field.
***
Unexpectedly rapid technological advances in recent years have convinced many U.S. power players that AI could “become the most powerful and strategic technology in history.” Fueled in large part by soaring rhetoric from Silicon Valley titans such as Sam Altman, Dario Amodei, and Elon Musk, many U.S. policymakers now believe that artificial general intelligence (AGI)—a loosely defined term for AI systems that match or exceed human intelligence across a range of tasks—may arrive at some point in the next few years.
Senior officials from across the political spectrum have painted a dark picture of what it would mean should China reach AGI first. In one interview, a former special adviser to President Biden warned, “I would not want to live in a world in which China has that capability … and the United States does not.” President Trump’s own “AI czar,” David Sacks, has echoed this view, stating bluntly that “neither [the United States nor China] can take the risk that the other one will develop a decisive advantage [in AI].”
Mainstream discourse around the AI competition is now increasingly framed around zero-sum analogies borrowed from the nuclear age, calling to mind a strategic arms race. Influential reports published in the past year have promoted policies such as a “Manhattan Project-like” effort to secure U.S. AI dominance and a “Mutually Assured AI Malfunction” concept to foster strategic stability.
The existential character of these assessments has led the United States not only to boost its own AI development but also to actively hinder China’s progress. The Biden administration imposed a series of sweeping export controls that cut off China’s access to advanced semiconductors and chipmaking tools. Officials framed these restrictions as national security imperatives to deny China the computing power necessary to train and run frontier AI models.
The Trump administration has since relaxed some export restrictions, but its overall strategy remains focused on cementing U.S. AI dominance. Vice President Vance noted at the 2025 Paris AI Summit that the administration would seek to “close pathways to adversaries attaining AI capabilities that threaten all of our people.” Meanwhile, Trump has promoted ambitious plans to accelerate U.S. AI development by boosting data center construction, adopting the logic of AI “scaling,” which holds that concentrating ever greater quantities of advanced semiconductors will produce increasingly powerful models. Since Trump came into office, U.S. tech companies have announced hundreds of billions of dollars in data center investments—such as OpenAI’s Stargate project and xAI’s Colossus—with the goal of reaching AGI.
In China, there is no doubt that AI is a strategic priority, as the country is building a world-class AI ecosystem to close the gap with the United States. Chinese labs such as DeepSeek and Alibaba have produced competitive foundation models that lag their U.S. competitors by only a few months, and the People’s Liberation Army (PLA) has been hard at work integrating AI into autonomous systems, operational planning, and battlefield simulations. China is also rapidly scaling domestic chip production and constructing data centers to support its AI efforts, so far with mixed success.
The country’s opaque political system makes it difficult to gauge how top Chinese leaders view AGI timelines; it is debatable whether Xi and other top leaders believe AGI is imminent. But even if they currently do not, accelerating progress in U.S. labs, continued export controls on advanced semiconductors for training AI models, and escalating rhetoric in Washington and Silicon Valley could force a reassessment. If Chinese leaders conclude that the United States is nearing a true AI breakthrough and decide that their current pace and approach to AI innovation are insufficient, then Beijing could look for ways to slow U.S. AI progress—just as Washington has sought to delay China’s.
***
China has already begun to take more aggressive actions to strike at the heart of sensitive U.S. supply chains in response to widening U.S. technology restrictions. Its unprecedented use of export controls on rare earth metals and magnets in recent months have signaled a growing appetite to escalate with Washington, yet its economic leverage may diminish over time. As Beijing’s perceptions of strategic risk grow, it could consider taking even more drastic steps to level the playing field.
AI models depend on three fundamental building blocks: data, algorithms, and computing power. While China has an ample supply of data collected from its more than 1.1 billion internet users, as well as nearly half of the world’s top AI researchers, its limited access to advanced chips needed to train and run large AI models has emerged as a major constraint.
Leading-edge AI chips, also known as graphics processing units (GPUs), are highly complex, expensive, and in short supply. Nearly all of the most powerful GPUs are designed by U.S. firms such as Nvidia and fabricated by Taiwan’s TSMC. U.S. export controls have gradually cut off pathways for these chips to reach China, creating a chokepoint that Chinese companies cannot easily bypass, even with sophisticated chip smuggling networks.
Chinese firms have made some progress in developing domestic alternatives, but their chips still lag behind in performance and efficiency. Moreover, they are difficult to produce at scale without access to advanced lithography equipment, exports of which the United States also restricts. As AI models grow larger and more compute-intensive, this hardware gap is becoming one of China’s most significant disadvantages in its technological competition with the United States. Without further export control relief from the Trump administration, this disparity will likely grow as new generations of even more powerful U.S. chips come online.
There is growing evidence that worsening chip shortages have already begun to slow China’s AI development. The CEO of Huawei Cloud acknowledged as much in 2024, admitting that “nobody will deny that we are facing limited computing power in China.” And DeepSeek’s CEO has noted that lack of access to advanced chips is the company’s most significant challenge. The company, which shocked the AI world in late 2024 with its release of its R1 model that was on par with leading competitors despite being trained on far fewer GPUs, was forced to delay the release of its most recent model due to inadequate compute resources.
Still, the United States’s compute advantage remains precariously reliant on its unchecked access to GPUs that are produced in Taiwan. Taiwan’s TSMC is one of a small handful of companies capable of manufacturing the world’s most advanced semiconductors, and the only one currently producing chips at the scale and precision required for frontier AI models. The vast majority of Nvidia’s most powerful GPUs, including the H100 and the latest Blackwell series, are fabricated by TSMC in Taiwan.
Recognizing this vulnerability, the United States has pushed TSMC to establish chip manufacturing facilities on American soil. In late 2024, the company opened an advanced fabrication plant in Arizona and began producing 4-nanometer chips (including those used in Nvidia’s GPUs) there earlier this year. It has also announced ambitious plans to scale up its footprint in the United States over the coming decade, committing over $100 billion to build new leading-edge fabs, packaging facilities, and an advanced research and development (R&D) center.
Despite these commitments, production at TSMC’s Arizona facilities will remain deeply entangled with its Taiwanese ecosystem for years to come. Until key processes such as advanced packaging and R&D are fully integrated in Arizona, even chips produced at U.S. facilities will need to be sent back to Taiwan before they can be provided to U.S. customers. Though TSMC is reportedly accelerating its Arizona facility construction, it is unlikely that the United States will achieve any meaningful degree of independence from Taiwan’s semiconductor ecosystem before 2030.
For China, the United States’s continued reliance on Taiwanese fabs might represent a shrinking window of opportunity. If TSMC’s overseas facilities become sufficiently independent and capable of meeting more U.S. demand, the strategic benefit of severing U.S. access to fabs in Taiwan through military action will decline. The incentive, then, is to take action while such a disruption would more significantly slow U.S. AI development, giving China the opportunity to catch up or pull ahead before the United States consolidates its lead.
***
The convergence of political and technological timelines is adding fuel to an already combustible mix. Xi’s order for the PLA to be ready for a Taiwan invasion by 2027 aligns with projections from Silicon Valley titans that AGI will arrive within the next few years.
That same window is politically charged. In the United States, a contentious 2028 presidential election cycle will likely focus public attention inward. In Taiwan, voters will go to the polls that same year, potentially reelecting a president whom Beijing has labeled a “peace destroyer.” And in China, the 21st Party Congress in late 2027 may be Xi’s last as leader, and it could signal his final chance to resolve the Taiwan question on his terms. On their own, each of these events would create an unpredictable political environment. Together, they could create a highly volatile backdrop in which a technological race plays out.
It is worth reiterating that Taiwan’s status is fundamentally a political issue for China. Communist Party lore suggests that China’s rise will be complete only when the mainland “reunifies” Taiwan, and the party has staked its legitimacy on eventually resolving the issue. These political factors will undoubtedly be the most important considerations in Beijing’s calculus, even if the AI competition brings additional incentives to act.
There are still many reasons why Xi may not be eager to take such a risk in the short term. China’s economy is fragile, as youth unemployment, real estate distress, and declining investor confidence pose major challenges. Furthermore, there are lingering doubts about the PLA’s actual readiness, especially given recent corruption scandals and limited combat experience.
Moreover, Beijing may be betting that the AGI hype is overblown. Instead of reaching for AGI, China appears to be focusing on integrating AI into various applications, such as robotics, in order to realize military and economic advantages. While China is pouring resources into domestic large language model (LLM) developers and chip companies, such as Deepseek and Huawei, the Chinese AI ecosystem appears to be experimenting with various non-LLM paths to advanced AI, such as “brain-inspired” approaches.
Yet the risk remains that should Chinese leaders come to believe Beijing is falling irrevocably behind Washington in AI, the pressure to intervene and disrupt U.S. momentum could intensify. In that case, U.S. dependence on Taiwan for AI chips would transform from the island’s greatest asset to a dangerous liability.
Ultimately, what matters most are the perceptions of top leaders in the United States and China. The sprint to AGI, even amid technological uncertainties, is raising the stakes in an already hyper-securitized relationship. The more both sides treat advanced AI as a decisive strategic imperative, the greater the risk that Taiwan’s once-vaunted “silicon shield” could become a target.
– Aidan Powers-Riggs, Sam Bresnick, Published courtesy of Lawfare.
