The Ai Divide

The AI Divide

How U.S.-Chinese Competition Could Leave Most Countries Behind

Sam Winter-Levy and Anton Leicht

February 10, 2026

• More by Sam Winter-Levy
• More by Anton Leicht

&
Download

Print Subscribe to unlock this feature or Sign in.

Sign in and to read later

Close

Article link: https://www.foreignaffairs.com/united-states/ai-dividehttps://www.foreignaffairs.com/united-states/ai-divide

Copy

The future of artificial intelligence will be controlled by the United States and China. The two countries employ 70 percent of the world’s top machine learning researchers, command 90 percent of global computing power, and attract the vast majority of AI investment—more than twice the combined total of every other state combined. In past technological revolutions, powers that were not at the frontier could gradually adopt new capabilities and catch up. But the AI revolution will be different, locking those countries into a strategic trap that could consign much of the world to technological vassalage.

This trap particularly affects what might be called the AI middle powers: countries such as France, India, and the United Kingdom, which have substantial state capacity and economic resources but lack the scale, capital, energy, and computing power to build frontier AI systems on their own. These powers face three principal challenges. First, their access to frontier AI capabilities is subject to the whims of policymakers in Washington and Beijing. Second, they remain exposed to AI’s disruptive effects—including job losses, social upheaval, and the expansion of AI-enhanced cybercrime—whether or not they in its benefits. Third, they lack the leverage and the policy tools necessary to shape AI’s development or manage its consequences.

Enduring marginalization is not inevitable. But avoiding it will require the middle powers to retain access to frontier AI capabilities and identify what economic and strategic value they can offer to a world transformed by AI systems. Different paths to these goals remain open: some middle powers may choose to align themselves with the United States or China, some may attempt to play Washington and Beijing against each other to extract concessions, and others may mount an ambitious attempt at technological sovereignty. But all of them will ultimately have to reckon with what a global AI economy might look —and where they might find leverage within it.

STUCK IN THE MIDDLE

Today, most middle powers access AI the same way individuals do: through commercial products and interfaces provided by foreign companies. But that access is less stable than it appears. Unmanufactured goods, weapons, or energy, AI capabilities cannot be stockpiled. Every query to a frontier AI system depends on real-time access to infrastructure controlled by a handful of Silicon Valley firms—and, ultimately, by the U.S. government, which can restrict access to American AI systems through export controls or sanctions. This dominance could change: China is rapidly building out its own infrastructure, though it remains a few years behind. And efficiency gains might eventually democratize access: smaller models require less computing power, which could lower the barrier to entry. But so far, each new generation of frontier systems has grown more demanding to run, not less.

If the United States or China decided to cut off a middle power’s access to systems based in its territory, the consequences would currently be limited because few essential services—hospitals, power grids, or military systems, for example—depend on frontier AI to function. But as artificial intelligence becomes more deeply embedded in critical economic and national security infrastructure, those stakes will rise. And this risk is not hypothetical. In other areas, both the United States and China have proved willing to weaponize other countries’ dependence in search of leverage. China has repeatedly used export restrictions on rare earths elements and critical minerals as a coercive tool, while the United States has leveraged European reliance on U.S. security guarantees to extract trade concessions and pressure Denmark over Greenland. There is little reason to think that Beijing and Washington will show more restraint when it comes to AI.

The United States and China command 90 percent of global computing power.

Some middle powers have attempted to address this vulnerability by building data centers that can run foreign models within their borders, giving them greater control over those systems and the data they process. Recently unveiled projects in the European Union, South Korea, and the United Kingdom all include provisions for greater access to sovereign computing resources for running foreign AI systems. Because these arrangements preserve revenue streams for U.S. companies while binding allies more closely to American infrastructure, U.S. companies as well as the U.S. government are promoting them through private initiatives such as OpenAI’s Stargate data centers and policy directives encouraging AI exports.

But data center buildouts are extraordinarily expensive and may not be economically viable for every country attempting them. Countries with high AI adoption rates, including Germany and the United Kingdom, face steep energy costs and local resistance to major new infrastructure projects, making large-scale data centers a hard political sell. And even if countries build out local data centers, their dependence problem will persist. AI models require continuous updates to remain competitive, meaning that any version running on a sovereign data center will quickly fall behind without ongoing access to the provider. The computing hardware itself—advanced chips and networking infrastructure that come from U.S.-controlled supply chains—requires regular maintenance and replacement. Complete sovereignty over AI infrastructure, in other words, remains elusive.

Sovereignty over the models themselves is harder still. France has attempted to build its own AI systems through its company Mistral, and Canada has attempted the same with Cohere. These are designed to reduce dependence on imports and provide a fallback if access is cut off. But U.S. and Chinese frontier systems remain more capable than anything these local champions have developed, and the gap shows no sign of narrowing. As long as that remains the case, major businesses and national security agencies will prefer the highest-quality offerings from frontier developers.

FALLING BEHIND

Whether or not they have access to frontier AI systems, middle powers will remain vulnerable to AI’s downsides, often more so than the great powers themselves. The harms associated with advanced AI will not respect borders: criminal misuse, military deployment by adversaries, labor displacement, and corrosive social effects will reach countries irrespective of whether they host leading AI labs.

Defending against these harms is considerably harder without direct access to frontier capabilities. Creating effective cyber- and biosecurity defenses, for example, increasingly depends on maintaining close relationships with AI developers and reliable access to computing power. In the military domain, middle powers that lack AI capabilities may find their legacy systems outmatched, not just by great powers but also by smaller states or nonstate actors that gain access to advanced AI tools. These aggressors could gain meaningful AI capabilities even without formal access to frontier systems if, for example, great powers arm proxies to shift regional balances, attackers steal what they need through cyberattacks, or sub-frontier models prove sufficient for offense even where they fall short for defense—a particular concern when it comes to biological weapons or cybersecurity. More broadly, any serious policy response to the full spectrum of AI-related risks presupposes a clear-eyed understanding of the technology and its implications on the part of government officials—an understanding that is difficult to develop at a distance.

Unmanufactured goods, weapons, or energy, AI capabilities cannot be stockpiled.

Middle powers also face structural disadvantages in shaping AI’s trajectory, even within their own borders. They will have little leverage to wield over development and deployment decisions, and limited flexibility in their policy options. Their regulatory input carries scant weight with the companies and jurisdictions driving the technology forward, and their tax revenues from AI-related economic activity remain uncertain. From the periphery, they may also struggle to adopt the technology fast enough to keep their workforces competitive with AI-augmented labor abroad.

Staying out of AI’s orbit entirely—forgoing both its risks and its benefits—is not a viable strategy. This is because the effects of artificial intelligence are inescapable: it will reshape the productivity of economic rivals, the capabilities of adversaries, and the expectations of citizens whether or not their governments participate. A middle power might hope that staying on the sidelines will at least mean avoiding AI’s dangers along with its rewards. But suffering the risks while missing the benefits is entirely possible. For middle powers, that outcome represents the central danger.

The middle powers’ vulnerability stems in part from the unique nature of AI development. China and the United States are able to benefit from a virtuous circle whereby access to computing power allows their domestic champions to train better models, which then attracts more users and generates more revenue, which in turn funds investment in still more computing power. Each advantage compounds the next. Less transformative technologies have already demonstrated a milder version of this pattern—Europe’s failure to capture the economic gains of the Internet goes a long way toward explaining its relative decline against the United States. Countries that fall behind on AI may find it even harder to catch up.

BREAKING OUT

To escape this trap, middle powers face two imperatives: they must secure access to frontier AI capabilities and find an economic foothold in the emerging AI order. Access will be critical, and there are three realistic strategies for maintaining it.

The first is bandwagoning: allying closely with either the United States or China to ensure access to an AI ecosystem. The United Kingdom appears to be pursuing this path, seeking ever closer alignment with the United States’ AI industry through regulatory cooperation and infrastructure deals. This strategy is particularly appealing in a securitized environment in which access requires great-power patronage. But it is not without risks. It can deepen a one-sided reliance, foreclose access to competing providers, and leave a country even more vulnerable if its patron’s goodwill fades—a risk that has only grown more salient as the Trump administration uses technology access as a bargaining chip for unrelated trade concessions. A country cut off after becoming dependent would face cascading disruptions: businesses reliant on AI-powered services could lose functionality overnight, government agencies could find critical systems degraded, and the scramble to find alternatives could take months or years—or push middle powers toward alternative providers offering exploitative terms.

The second strategy is hedging:using domestic market scale and economic leverage to extract favorable terms from both great powers and spur competition between them. Malaysia and much of Southeast Asia are perhaps hewing closest to this approach, courting AI infrastructure investment and partnerships from both the United States and China. As long as systems remain interoperable and markets stay open, hedging promises cheap and stable access. But if the environment grows more securitized or the world divides into blocs, access could be lost.

The AI revolution could consign much of the world to technological vassalage.

The third strategy is to mount a genuine sovereignty effort,building enough computing capacity and investing enough capital to have a realistic chance of developing domestic frontier capabilities. France and several other countries profess to be pursuing this path, but their ambitions fall short of the frontier in energy, compute, and talent. A successful sovereign effort could theoretically help a country avoid many aspects of the middle power trap, but it carries risks. An underfunded attempt could leave a country stranded in an unprofitable second tier, close to the frontier but never reaching it. And even reaching the frontier offers no guarantee of returns if value ultimately accrues elsewhere in the supply chain.

The success of any of these strategies will depend above all on how much frontier capabilities matter in comparison with second-best alternatives. That in turn depends on whether small performance gaps prove decisive in AI-accelerated economic competition and whether the technical paradigm continues to allow so-called fast ing—the efficient imitation that lets middle powers trail the frontier by a year or so. Beyond these technical uncertainties, it remains unclear whether the Trump administration’s pro-export stance will hold as AI models become national security assets, or whether great powers will eventually securitize their AI industries and pull frontier capabilities off the open market. Given this uncertainty, middle powers should be cautious about locking themselves into any single path. They must monitor AI technology—and how Beijing and Washington perceive it—closely enough to pivot if conditions change. And in the meantime, they must remain risk-averse. They should not commit to any path that precludes access to imported frontier capabilities.

The United States, meanwhile, should make bandwagoning as attractive as possible in order to maximize its market , revenue, and leverage over AI’s development. But such an effort will conflict with many middle powers’ sovereignty ambitions. To resolve this tension, Washington must make bandwagoning beneficial and nonthreatening by forgoing aggressive attempts to leverage AI access, promoting exports that deliver meaningful capabilities, and implementing security standards that allow even highly sensitive frontier systems to be d with allies.

FINDING A ROLE

With access secured, middle powers face a second challenge. They must find durable economic and strategic niches that can guarantee their relevance as AI’s importance grows. One possible niche is upstream inputs—the tools and resources required to build artificial intelligence systems, from semiconductor manufacturing equipment to high-quality training data. Another is downstream bottlenecks—capacities that translate AI performance into real-world impact, such as robotics, and advanced manufacturing. And a third is automation-resistant sectors—health care, artisanal production, and tourism, for example—in which human presence or physical locality is a major source of value.

Not all niches confer equal leverage. The positions that matter most are those that great powers cannot easily replicate or route around. The Dutch semiconductor equipment company ASML already occupies such a position. The United States cannot quickly build its own extreme ultraviolet lithography capability, and every advanced chip fabricator in the world needs ongoing access to ASML’s machines and expertise. Such capabilities remain valuable because they are of continuous use rather than capable of being exhausted in a single transaction. A country sitting on valuable training data, by contrast, may find that once it has been sold or scraped, its leverage disappears. India’s vast stores of professional services data may prove strategically useful if New Delhi retains control and conditions access on lasting partnerships. Or this data could be sold off piecemeal to U.S. or Chinese AI companies and lost as a source of leverage entirely.

Middle powers’ leverage depends on assets they have not yet secured and could easily squander.

Not every middle power can control an input as critical as ASML’s. But many can position themselves at the other end of the chain, where AI capabilities must be translated into real-world impact. The race for AI supremacy is absorbing U.S. talent, capital, and government focus to a degree that leaves gaps elsewhere. As systems grow more capable, the bottleneck will increasingly shift to deployment—to the manufacturing, robotics, and applied R & D capacity needed to convert models’ performance into economic and strategic value. These are areas in which U.S. capacity is limited and Chinese capacity is vast, because of decades of offshoring from the United States and sustained industrial investment by Beijing. Middle powers with strength in these areas—including South Korea and Japan in advanced manufacturing and robotics, Germany in precision engineering, and India in pharmaceutical R & D—may find themselves newly valuable to the United States as it looks for allied scale to counter China’s industrial base.

Middle powers should therefore prioritize niches that great powers cannot do without: the inputs required to build AI systems or the capacities required to put them to use. That means resisting pressure to sell off or offshore strategic assets, as has happened when European robotics and semiconductor firms were acquired by Chinese investors or when promising AI and biotech startups were lured to the United States by deeper capital markets. Silicon Valley companies will come calling in pursuit of exclusive data access, and foreign governments will seek control over key links in the supply chain. Middle powers should be wary: they should not evaluate these assets on near-term returns alone, but on their potential to serve as a source of leverage and a strategic foothold in a changing world economy. In part, that means strategic screening of foreign investments to ensure that critical domestic industries aren’t hollowed out or offshored. But it also requires active industrial policy to counter the subsidies of rival great powers. Middle powers cannot match that spending across the board, but they can succeed in safeguarding a narrow set of strategic sectors if they identify them early. They will also need to ease taxation and regulation for sectors essential to their competitive positioning.

The harms associated with advanced AI will not respect borders.

The United States also has an interest in shaping how middle powers position themselves. Beijing threatens to outcompete Washington in manufacturing capacity, supply chain integration, and speed of deployment. To counter this, the United States needs the capabilities of a broad alliance. Through bilateral engagement, export promotion, and trade deals that credibly commit both sides to lasting technology access, the United States can incentivize middle powers to align their economic positioning with U.S. strategic priorities. This can help well-positioned allies expand bottlenecks and receive enduring access to AI capabilities in an alliance founded on mutual technological dependence.

But this vision requires both sides to act strategically. On AI policy, the Trump administration has taken some steps in the right direction: its Pax Silica initiative, which exchanges access to U.S. technology for allied contributions to the semiconductor supply chain, and its executive orders promoting AI exports suggest an awareness of the stakes. But Washington has also shown a willingness to weaponize technology access for unrelated trade concessions, as the United Kingdom discovered in December when a technology deal was reportedly suspended over British food safety standards. The Trump administration’s broader hostility toward allies has left partners questioning whether the United States can be a reliable anchor for any alliance. Middle powers, for their part, must recognize that their leverage depends on assets they have not yet secured and could easily squander.

If navigated well, middle powers can turn their strategic predicament into an efficient division of labor backed by hard mutual leverage. On their current trajectory, however, they are headed for a trap in which they will bear the brunt of AI-driven disruption while capturing few of the benefits. The outcome would be dire: two great AI powers barreling toward a technological revolution with most of the world’s computing power and talent, leaving most of the world’s citizens behind.

Loading…
Please enable JavaScript for this site to function properly.

You are reading a free article

Subscribe to Foreign Affairs to get unlimited access.

• Paywall-free reading of new articles and over a century of archives
• Six issues a year in print and online, plus audio articles
• Unlock access to the Foreign Affairs app for reading on the go

Subscribe

Already a subscriber? Sign In