The Real Stakes of the AI Race

A sense that global technology competition is becoming a zero-sum game, and that the remainder of the twenty-first century will be made in the winner’s image, pervades in Washington, Beijing, and boardrooms worldwide. This angst feeds ambitious industrial policies, precautionary regulations, and multibillion-dollar investments. Yet even as governments and private industry race for supremacy in artificial intelligence, none of them possess a clear vision of what “winning” looks like or what geopolitical returns their investments will yield.

Much more than computing dominance is at stake; the struggle for AI primacy between the United States, China, middle powers, and Big Tech is fundamentally a competition over whose vision of the world order will reign supreme. For the United States, AI is a new frontier on which it must maintain its global technological dominance. As U.S. policymakers deploy a regulatory arsenal to cripple China’s technology development and stay ahead, China is mobilizing the power of the state to close the gap. At the same time, middle powers trying to avoid coming under the shadow of either superpower, along with tech companies devoted to the global diffusion of technology through open markets, see AI development paving the path to a multipolar world.

Already, U.S. technology controls and China’s escalatory response are creating a snowball effect: as the United States, bent on preserving tech primacy, resorts to more aggressive measures to throttle Chinese AI development, a cornered China will hunt for leverage over the United States, including, potentially, in security flash points such as the Taiwan Strait. Middle powers and tech companies, meanwhile, will strive to build AI systems and applications outside the confines of great-power rivalry—both looking to stake a claim to the new technological order even as they risk getting caught in the crossfire. As the tenor of the global AI race grows existential, the high-stakes wagers of tech titans and world powers risk geopolitical combustion.

PROTECT THE LEAD

The United States is building its AI strategy on the assumption that it can preserve its hegemony offensively, through a rate of technological innovation that outpaces the rest of the world, and defensively, through far-reaching technology controls aimed at hobbling China, its biggest geopolitical challenger. Export controls and investment restrictions are designed to cut off the flow of goods, capital, and technological know-how from Beijing. Washington’s strategy assumes that China is in structural economic decline, its statist approach stifling the animal spirits of an economy at the mercy of the Communist Party and its general secretary, Xi Jinping.

It also puts enormous stock in the the pervasiveness and competitiveness of U.S. technology. Washington is betting that partners who may resist aligning with its protectionist strategy will ultimately cast their lot with U.S. technology and the promise of Western AI innovation rather than gamble on China closing the gap and risk running afoul of U.S. sanctions. After all, U.S. technology and intellectual property dominate every level of the AI industry. Nvidia’s AI accelerators increase computing power by orders of magnitude, driving the AI revolution; U.S. cloud service providers, such as Amazon Web Services, Microsoft Azure, and Google Cloud Platform, distribute massive computing resources and cloud infrastructure; and companies such as Google, Meta, OpenAI, Anthropic, and xAI have developed the foundational AI models that companies all over the world will rely on to fine-tune their AI applications.

The United States holds unique leverage in this early stage in the development of AI. Large-scale AI models, particularly in generative AI, which uses patterns learned from existing data to create new content, require immense computational power and vast amounts of data, the resources for which few companies outside the United States possess. The United States’ lead in this industry is not an accident; it has attracted more than $328 billion in investment over the past five years to promote AI development and foster a culture in which risk-taking is rewarded. Building on the Biden administration’s suite of industrial policies expanding domestic chip manufacturing, the incoming Trump administration has touted an “AI Manhattan Project” to turbocharge a U.S.-led AI industrial revolution.

U.S. tech hegemony is not impregnable, however. The development and diffusion of AI technology could dilute the United States’ current advantages. Generative AI development to date has largely been focused on obtaining the huge amounts of computational power, data, and energy required to train large-scale AI models, for which access to the world’s most advanced chips has been critical. U.S. dominance of the production of these chips through companies such as Nvidia and AMD has enabled the U.S. government to tighten export controls to severely restrict that access. The United States is trying to hobble China’s domestic AI development, in particular, by cutting off its supply of high-end chips and the equipment and components needed to manufacture advanced node semiconductors.

U.S. tech hegemony is not impregnable.

But Washington finds itself in an awkward phase of that strategy. Despite a barrage of U.S.-led chip controls, Huawei’s HiSilicon and SMIC, China’s leading chip designer and manufacturer, respectively, leading China’s push for self-reliance, are producing high-performance AI accelerators used in data centers, such as the Ascend 910C processor, which is nearly as advanced as Nvidia’s H100 and A100 chips. Still, Chinese production has been far less efficient than that of the Taiwan Semiconductor Manufacturing Company (TSMC), which uses state-of-the-art manufacturing technology to produce cutting-edge chips.

Chinese engineers are being flooded with state resources to innovate their way out of U.S.-imposed constraints on chip production. AI development is shifting away from maximization of computational power for training large-scale models and toward optimization to get pretrained models to generate more sophisticated responses to queries. Nvidia and other leading companies are already focusing less on shrinking transistor nodes at the atomic level and more on performance gains across the full AI system, from the design of the chips themselves to the cooling systems used to integrate the hardware in data centers. When launching the first round of chip controls in October 2022, U.S. policymakers assumed that process node miniaturization was the fundamental chokepoint to “freeze in place” Chinese chip production. But with Huawei leading China’s self-reliance efforts, China has demonstrated that it still has the process engineering ability, manpower, and sheer tenacity to keep up with industry advances in performance optimization.

Concern about the United States’ ability to create a wide enough lead over China in AI development will lead to blunter controls, less patience for partners that do not align their own policies with U.S. export restrictions, and swifter application of extraterritorial measures. In December, the U.S. Department of Commerce released a package of semiconductor controls that exemplifies this approach, using far-reaching restrictions to force partners into alignment and to throttle Chinese production. But if Chinese companies are still able to scale their computing power, and gaps in partner alignment on chip controls remain, Washington will likely adopt blocking sanctions on Huawei and expand trade restrictions to include China’s AI tech champions. The potential blow to the Chinese economy and Beijing’s geopolitical ambition would heighten the stakes of the AI and chip war.

PLAYING CATCH-UP

China assumes that it is on equal geopolitical footing with the United States in a bipolar world. For Beijing, AI development is both the path to preserving parity with Washington and the antidote to its domestic economic challenges. China has intrinsic strengths: the sheer size of its population and the country’s massive industries give it an enormous well of data to draw from in training AI models and the opportunity to pioneer AI applications for manufacturing. Thanks in part to their large pools of employees willing to work long hours, Chinese tech champions such as Alibaba, Baidu, DeepSeek AI, Huawei, and Tencent are competitive with U.S. peers in large language model development and cloud infrastructure. China has also put enormous resources into optimizing power generation for data centers; it is now the fastest developer of energy generation in the world. Beijing is betting that high levels of state funding for AI-powered industries will pay off, especially if overreaching trade and technology controls alienate U.S. partners, drive them into China’s arms, and open up new markets for Chinese technology. From China’s perspective, this strategy is the country’s best (and only) hope to arrest recent economic malaise and to avoid subordination to the United States.

But China also faces significant vulnerabilities, most notably in its ability to develop the AI hardware needed to keep pace with the United States. As U.S. technologists create ever more efficient AI chips and dynamic AI models, China’s leading foundries will have to devote considerable resources just to maintain their chip production capabilities as tightening U.S. controls restrict their access to key components. This diverted energy will make it more difficult to keep pace with U.S. innovation. Chinese AI developers will face a dual challenge of seeking breakthrough innovation amid stifling U.S. tech controls while also complying with the nebulous Communist Party mandate that AI models “uphold socialist values.”

U.S.-led chip controls have not been impenetrable to date, but more aggressive U.S. controls could reduce the ability of foreign toolmakers, chip designers, and foundries to maintain a foothold in China’s chip market. If the United States builds on current de-risking momentum and coordinates with partners effectively, China could find itself increasingly excluded from emerging U.S.-led technology and trade blocs that reserve access to data and critical digital infrastructure for like-minded countries. Any potential reorganization of trade blocs would be disastrous for China’s ability to expand to foreign markets, especially at a time when its tech giants are relying on overseas growth to compensate for the structural decline of the domestic economy.

If China struggles to keep pace in chip development, and Beijing fears that this will make it fall further behind the United States in the AI race, the state could direct Chinese tech juggernauts to centralize and scale their computing resources. But such a step would expose China further to U.S. tech controls. For now, Huawei’s HiSilicon is the country’s preeminent AI chip designer. But Huawei is also the Chinese tech company most heavily sanctioned by the United States. Huawei’s chips are manufactured by SMIC, China’s state-owned foundry, which, like Huawei, is now subject to U.S. foreign direct product rule restrictions. These are designed to prevent components from any country from going to SMIC plants where advanced chips are being manufactured. If the government orders Chinese AI model developers to pool their resources to support the state’s AI development, they, too, would become prime targets for severe U.S. sanctions that would undermine their efforts to expand their market shares outside China. The United States has already expanded export control categories to include the gray area of firms engaged in military “support” activities. The justification for broader controls is already baked into U.S. rules and is waiting to be deployed.

GOLDEN TICKET?

Middle powers and Big Tech leaders have yet another outlook on the global AI competition. Many of them see the diffusion of technology as enabling the new era of multipolarity. With the United States and China locked in a contest for AI primacy, countries such as France, Saudi Arabia, Singapore, and the United Arab Emirates have started to build sovereign AI systems that make use of their national strengths: control over data access, intimate knowledge of their respective economies, and expertise in their respective languages and cultures to mitigate bias. The UAE, for example, may deploy up to $100 billion over the next few years as part of the country’s goal to be a global AI leader by 2031. These gains reflect middle powers’ status anxiety, as well as their recognition of the opportunity to carve out niches for themselves amid U.S.-Chinese competition.

Nvidia’s founder, Jensen Huang, has been quick to acknowledge the mix of anxiety and aspiration among middle powers. He has toured Canada, France, India, Japan, Malaysia, Singapore, and Vietnam, delivering the same seductive message: every country needs its own sovereign AI to reflect its language, culture, and ambitions. Meta, meanwhile, has open-sourced its Llama large language model, sharing its blueprint for building the model with the world. Open-sourcing makes it easier for the technology to diffuse across borders. It also may erode the competitive advantages of Meta’s closed-model rivals, including OpenAI’s GPT, Anthropic’s Claude, and Google’s Gemini. Unlike their competitors, which have erected proprietary walls around their technology to protect their innovation, open-source advocates such as Meta hope an expanded ecosystem of developers will encourage widespread adoption of and innovation on their own platform.

For most countries, however, achieving AI sovereignty is more complicated than tech leaders make it out to be. Middle powers can build AI infrastructure, but they will likely be doing so with U.S. semiconductors, cloud infrastructure, and a heavy reliance on U.S. talent and U.S.-origin AI foundation models. The ubiquity of U.S. technology enables Washington to impose conditions on partners, such as excluding Chinese firms from their supply chains and preventing Chinese access to U.S.-built AI systems, on the grounds that Chinese companies pose national security risks.

Finding the optimal level of AI regulation will also be a challenge. Many tech leaders prefer a light touch on AI regulation to clear the path for private industry to innovate without getting bogged down by heavy compliance requirements, but their appeals may not always align with U.S. government priorities to impose restrictions that block out China and countries willing to work with China. For example, open-source business models alarm security-minded policymakers who fear that unrestricted access could enable adversaries to more easily develop or exploit sensitive technology. Meanwhile, middle powers lacking a strong domestic tech base will try to assert their relevance by passing regulation that creates a de facto global standard for AI development. In August, the EU AI Act came into force as the world’s first comprehensive AI law. It was met with significant criticism, including from the former European Central Bank president Mario Draghi, who warned that the law’s “precautionary approach” to regulation would be onerous, and advised that deregulation is urgently needed to “close the innovation gap with the U.S. and China.”

CHIPS ON THE TABLE

With nothing less than competing visions for the world order at stake, competitors will do all they can to gain the upper hand. Unintended consequences will follow. U.S. regulators may be overzealous in their controls. If Washington pursues proposed quotas on the number of U.S.-made AI chips other countries are able to procure, for example, it may undermine its leverage to lock in strategic dependencies for U.S. technology in fast-growing markets. U.S. officials might also be underestimating the momentum of China’s technology innovation. Overconfident in the efficacy of export controls, they could prematurely see inefficiencies in SMIC chip production as evidence of their policies’ success, while domestic Chinese suppliers make strides with heavy state backing and Chinese foundation model developers such as DeepSeekAI and Alibaba pool enough high-end computing power to rival their U.S. peers.

If Washington comes to believe that the AI technology gap between the two superpowers is closing, it will push for blunter measures. This would likely entail imposing full blocking sanctions on Huawei and more draconian restrictions on Chinese AI tech champions. If China’s tech giants are denied markets abroad while its statist economic model is faltering at home, a cornered Beijing may interpret the United States’ moves as an existential threat, creating the conditions for the technology war to potentially spill into the security sphere.

China’s leadership could respond to intensified trade restrictions with escalatory moves elsewhere. Assuming that Washington would try to avoid direct war, Beijing could test its rival’s limits with more serious gray-zone maneuvers in the Taiwan Strait. For instance, China could establish a customs quarantine that effectively asserts the mainland’s writ over Taiwanese ports of entry and challenges Taiwan’s political sovereignty. Such a step would have dramatic consequences, not only for the future of semiconductor supply chains as assets and investment flee Taiwan but also for the balance of power in the Indo-Pacific.

For now, the unbridled optimism of technologists will keep the AI revolution moving forward, in hopes that the spread of the technology will transform humanity. At the same time, the anxiety of policymakers, who see economic and security risks lurking at every corner, will continue to drive efforts to channel AI competition toward zero-sum geopolitical goals. All recognize that power will accrue to those who hold the keys to AI development and deployment. As countries and tech giants jockey for position, the geopolitical tempest that ensues may overshadow the transformative potential of the technology. 

Loading…