For many in Washington and Silicon Valley, the race to artificial general intelligence has become the defining test of U.S. technological leadership. But the scramble to dominate the frontier has obscured a parallel—and in many ways more immediate—contest: the race to deploy and embed AI systems across the globe and thereby secure market share, technological influence, and political leverage. In much of the world, the outcome of that race will not hinge on who builds the most powerful model, but on who shows up with reliable infrastructure, tailored applications, and financing to match.
The scramble to dominate the AI frontier has obscured a parallel contest: the race to deploy AI systems across the globe and thereby secure market share, technological influence, and political leverage.
While America leads at the frontier, in this broader contest its position is less secure. “Good enough” and cheaper Chinese alternatives may increasingly become the default choice in many parts of the world, especially the Global South. In the weeks following its release, Chinese-owned DeepSeek became the most downloaded mobile app in 140 markets, including Brazil and India, and now has more than 125 million global downloads. Chinese companies are rolling out models in local languages in key emerging markets, part of a “going global” strategy in deploying cloud infrastructure abroad. Eventually, China may be able to produce competitive AI chips at scale, at which point it will likely seek to deploy them globally. Taken together, these developments reflect a broader strategic imperative of the Chinese Communist Party: During the Politburo’s April study session, Xi Jinping called on the country to “vigorously engage in international cooperation on AI” and “help Global South countries enhance their technological capabilities.”
This trend carries profound implications for U.S. national security and global influence. These technologies will fundamentally reshape policing, education, healthcare, legal services, governance systems, and regulatory regimes across much of the world, conferring vast soft and hard power on the countries that develop and diffuse them. Yet Washington has devoted relatively few resources to this problem. In its international AI policy, it has emphasized protective measures—above all, export controls—aimed at constraining China’s AI advances; it has paid only selective attention to overseas promotion, concentrating on a handful of wealthy dictatorships seeking to play a major role in frontier AI development, such as the United Arab Emirates and Saudi Arabia.
Meanwhile, U.S. AI strategy has mostly overlooked a far larger group of nations across the Global South with growing technological needs. These countries, including Brazil, Indonesia, Kenya, and others, wield meaningful regional influence. They will play a significant role in determining the contours of the global AI ecosystem and the extent of U.S. soft and hard power within it. In the absence of a tangible counteroffer from the United States, these countries may pay little heed to American complaints about their deepening technological relationships with China. Their focus will be on the concrete economic problems the United States can help them solve in specific local contexts.
In its first months in office, Donald Trump’s administration has dismantled many of the institutions and programs that traditionally marshalled U.S. resources, diplomatic capital, and technical expertise toward competing in the Global South. But as it works on its replacement for the AI diffusion rule, the administration now has an opportunity to build a more focused and effective strategy for promoting U.S. technology overseas.
That should begin with prioritizing the cloud and model layers of the AI stack and concentrating U.S. government support on a handful of strategically important emerging markets, including Brazil, India, Indonesia, Kenya, and Nigeria, where demand for advanced AI infrastructure is likely to grow and American businesses already have some foothold. Washington should help U.S. companies deliver tailored offerings to these countries—such as high-quality data centers linked to public-sector AI applications for delivering healthcare, education, or government services—while securing narrow but meaningful commitments in return, including alignment on export controls, data center security standards, and removal of Chinese networking hardware from sensitive systems. To support these efforts, Washington should unlock additional sources of strategic capital through reforms to its export promotion institutions, including reauthorizing, expanding, and streamlining the Export-Import Bank (EXIM) and the Development Finance Corporation.
Done right, these efforts could position American AI technologies at the center of a forward-looking international engagement strategy—one that recognizes that technology promotion is as critical to long-term security as technology protection.
The Case for an AI Export Promotion Agenda
The case for a more proactive U.S. strategy to promote its AI technologies worldwide rests on a few basic propositions. For starters, by expanding AI infrastructure worldwide, American companies can help unlock substantial development benefits in the Global South, enabling advances in areas such as healthcare, education, and agriculture. In turn, a global U.S.-led AI ecosystem would allow American companies to secure market share, including in some of the world’s most populous and rapidly growing countries, generate export revenue, and funnel those earnings back into domestic R&D—fuel for staying ahead in frontier‑model development.
Embedding U.S. technical standards in overseas deployments will also give Washington a seat at the table as other states draft rules on data security, privacy, and model safety. In facilitating this diffusion, U.S. companies could help foster beneficial AI norms around the world, for example by strengthening data center security requirements and establishing safeguards against misuse. And becoming the backbone of global AI infrastructure will afford Washington greater visibility into AI development―and, when necessary, a lever to deny adversaries access.
More importantly, the global spread of U.S. AI systems would serve as a powerful source of American soft power. As digital technologies become ever more central to daily life, AI ecosystems will influence the way people understand and interpret the world. Chinese models need not parrot official propaganda to shift perceptions: Even subtle model design choices—echoing TikTok’s algorithmic skew—could systematically paint China in a favorable light while vilifying the United States. Issues of censorship and information control apply not only to Chinese models hosted on the cloud, but also to open-source ones that can be downloaded and modified. Ultimately, such models could shape global public attitudes if they come to dominate information landscapes worldwide, especially in fields such as education.
For the purposes of a U.S. export promotion agenda, the most critical parts of the AI stack to target are the cloud layer and the model layer. The cloud layer—including physical data centers often directly owned by cloud hyperscalers—provides the foundation for AI training and deployment, while the model layer enables the development of AI applications. By focusing on these layers, the United States can help local actors, both public and private, access products that are central to their economic and developmental agendas. U.S.-enabled cloud access also provides on-demand computing power that can be limited or restricted as needed, making it a more targeted and flexible tool for oversight. Crucially, computing power may be a lasting asset: Switching away from an established cloud and AI ecosystem with both digital and physical aspects requires investments in new infrastructure, software, and workforce. This means that the initial supplier of these assets may be able to serve as the enduring foundation for other nations’ AI ecosystems, especially if there are strong network effects for AI-enabled software applications, and become a platform on which U.S. companies can introduce additional products, both software and hardware.
Today, the United States enjoys a critical window to entrench its leadership in global AI infrastructure.
Today, the United States enjoys a critical window to entrench its leadership in global AI infrastructure. It currently is the only country capable of exporting high quantities of advanced AI chips. Due to export controls on advanced chipmaking tools, Chinese firms can currently produce a limited volume of these chips, and at lower quality than their American counterparts. For now, as the administration recognizes, China cannot produce enough chips for its domestic purposes, let alone for export. This is partly why U.S. companies dominate global cloud infrastructure—Amazon, Microsoft, and Google account for 59 percent of the world’s hyperscale data center capacity. Chinese companies generally do not invest as much in overseas data centers as they do in data infrastructure at home, and they focus on selective markets such as Southeast Asia. The United States should seize this moment to deploy its cloud and model solutions globally—deepening integration before Chinese alternatives can scale.
How the United States Could Nevertheless Lose the Export Promotion Race
Yet for all the United States’ technological superiority at the frontier, China could nevertheless outpace it in AI export promotion in two key ways: by offering more affordable products and by catering to local political concerns about digital sovereignty.
China’s most direct route to success in the global deployment of its technology has always been through price, with some estimates suggesting that Chinese digital offerings cost up to 30 to 40 percent less than their foreign competitors’ rates. In Southeast Asia, for instance, Chinese cloud providers typically price their services 20 to 40 percent below those of American competitors across various product categories. They do this for a few reasons. Many Chinese technology companies—above all Huawei—enjoy tens of billions of dollars in government backing in the form of grants, credit facilities, tax breaks, and other forms of financial assistance. Fierce competition and limited differentiation among Chinese products also mean that firms often must compete on price, leading to price wars that further drive down product costs.
Chinese AI models also fit this trend. They are closing the performance gap with leading U.S. rivals while offering significantly lower prices. Chinese open-weight models reinforce this dynamic. DeepSeek and Alibaba’s models—whose performances rival and sometimes surpass many American models—allow anyone to download, run, and modify them while only paying for the required computing power. As the analyst Kevin Xu has pointed out, most of the best models coming out of China are open-source with permissive community licenses. By contrast, nearly all similar American AI models are closed-source, which can restrict user flexibility and raise costs. Even Meta, the United States’ most prominent champion of open-source models, only offers a corporate license for its open-source model, Llama, and is reportedly weighing a shift toward closed-source model development amid corporate dissatisfaction.
Beyond the direct cost of products, Beijing helps lower the overall economic burden of technology infrastructure development—including data centers and telecom infrastructure—for recipient nations through robust state-backed financing, drastically outspending the U.S. government in this realm. This Chinese aid generally comes with few (explicit) strings; China’s state-run policy banks typically offer loans without conditionality, imposing few or no expectations of host country economic or political reforms (though they may come with implicit conditions regarding a nation’s foreign policy). And China’s “deal team packages,” which present full-suite digital infrastructure solutions, can be a “magic formula” for local government officials.
Chinese AI firms are also well positioned to address the digital sovereignty priorities of countries in the Global South.
In addition to affordability, Chinese AI firms are also well positioned to address the digital sovereignty priorities of countries in the Global South. Many of these governments have pushed for the build-out of data centers on local soil to ensure data privacy and to retain control over digital ecosystems. However, limited financial resources and technical capabilities often force them to turn to international partners for assistance in building and financing these projects. Such partnerships can be a hard sell to American companies and development financing institutions wary of high upfront costs, lack of key inputs such as skilled labor, and insufficient demand. But their Chinese counterparts may be more willing to make inroads. Chinese banks, for example, have historically been willing to stomach more risk than their Western counterparts, often lending to borrowers with a high likelihood of non-repayment (though China may now be recalibrating its global development initiatives in response to soaring debts from borrower countries and increasing global scrutiny over corruption and environmental harm). And as access to U.S. and other Western markets becomes increasingly uncertain, Chinese technology companies view expansion into Global South markets as an important way to sustain growth.
Local governments also want AI models attuned to their local needs and priorities, whether linguistic or cultural. Chinese companies have recognized and acted on this trend—Chinese smartphones, for example, have penetrated markets across the Global South with tailored features. Alibaba’s Qwen3 models reportedly support 119 languages and dialects, including some that lie outside most of the world’s AI training data, such as Bengali, Burmese, and Urdu. By comparison, Meta’s open-source AI model Llama4 only covers twelve languages. It’s hard to say whether the Chinese models are truly so much more capable in world languages, or whether their developers are simply more willing to make bold claims in an effort to market to the Global South. But even the latter explanation suggests a level of commercial focus on emerging markets that will play to China’s advantage in much of the world.
These dynamics could create a landscape in which American AI products continue to lead at the frontier, but China’s AI offerings nevertheless dominate in much of the developing world. Importantly, Chinese models don’t need to outperform American ones to do so. “Good enough” models can power use cases ranging from government platforms to educational programs. And once a model is trained, top-tier chips are not necessary for some deployments—DeepSeek, for example, can run on Huawei’s Ascend chips, which lag behind Nvidia’s best hardware but are good enough for many routine inference tasks. Even before China can produce significant volumes of the most high-performing AI chips, it will thus likely be able to compete in many global markets. Once established, its infrastructure will be difficult to dislodge, especially if China pairs the building of data centers with the requirement to use Chinese standards, creating harmonization challenges for international competitors. And as it gains ground globally, large quantities of data and revenue will flow back into Chinese firms, fueling further technological growth.
An Export Promotion Agenda
For now, the United States has a window of opportunity to spread U.S. AI technology in emerging markets where Chinese companies might otherwise entrench themselves. As the Trump administration develops its replacement for the diffusion framework, it should continue to work to mitigate many of the national security risks associated with unrestricted AI chip exports. But to compete more effectively with China’s digital offerings, this emphasis on controls and safeguards needs to be paired with a more positive and sustained focus on how to promote U.S. AI exports in the Global South. U.S. tech companies will take the lead here: In many cases, they have natural incentives to reach foreign markets, and companies like OpenAI have begun to roll out initiatives to build data centers in places like the Asia-Pacific region and to help countries customize their products for local languages and needs.
But Washington can play a more proactive role facilitating and accelerating these efforts than it has to date. To do so, it should implement five key strategies:
1. Pick Smart Battles and Tailor AI Offerings
The United States does not have the public resources to compete with China everywhere—nor would it be worth the political and economic capital to try to mobilize the sums needed to do so. Washington will thus need to prioritize key regions of the world and key layers of the AI stack. Rather than attempting to build up cloud infrastructure in countries across the entire Global South, for example, the U.S. government should concentrate its efforts on a handful of strategic hubs. Nations like Brazil, India, Indonesia, and Kenya have growing AI aspirations and robust digital economies, but they lack the capital and resources to build completely indigenous AI ecosystems. (Africa and Latin America currently have almost no AI computing hubs, according to data compiled by a team of Oxford University researchers.) There are clear synergies between the needs of these emerging hubs and U.S. capabilities and objectives.
The United States does not have the public resources to compete with China everywhere—nor would it be worth the political and economic capital to try to mobilize the sums needed to do so.
For one, American companies are well poised to compete in these nations, where the desire for high-quality, cutting-edge AI data centers—something only the United States can currently offer—may outweigh concerns about affordability. Moreover, thanks to their more established electricity and connectivity infrastructure, these countries are well positioned to host data centers that can deliver benefits to not only local populations but also broader regions. Such countries are already likely targets for hyperscaler investment, which means the U.S. government may be able to catalyze meaningful private capital flows with fewer public resources. While U.S. hyperscalers have established a presence in some emerging AI hubs like Indonesia, they have neglected other equally important markets. Nigeria, for example, is expected to become the world’s third-most populous country by 2050 and is currently home to over 19,000 tech startups, but American hyperscalers have yet to establish full-scale data centers there (the one exception is a limited local zone facility launched by AWS in Lagos in 2023).
To be sure, a focused approach on a small group of major emerging economies has its downsides. Many of these potential AI hubs tend to be larger countries that are inclined to hedge between the United States and China, making it more challenging to secure exclusive alignment with the U.S. tech stack. Brazil, for example, has steadily strengthened its diplomatic and technological ties with China, and while Indonesia views the United States as its preferred security partner, it has long embraced China for investment and infrastructure; Chinese tech companies like Huawei have helped build the country’s telecoms networks and subsea cable infrastructure.
But if the United States refuses to accept coexistence with some degree of Chinese technological presence in target countries, it may find that it has to limit itself to deploying AI infrastructure in smaller states like Costa Rica or Uruguay—useful diplomatic partners to be sure, especially in multilateral forums such as the United Nations and the Organization of American States, but far less influential players than major economies like Indonesia or Nigeria. At this stage in the AI competition, ceding major markets is unlikely to be the right approach.
Washington should also help U.S. tech companies tailor their offerings for individual countries. This includes linking data center build-outs with technical and implementation assistance at the model and application level, especially in the public sector, to solve local economic problems—for example, in the health care sector or other public service software projects. Diplomats at the State Department, the U.S. Commercial Service, and the U.S. Trade and Development Agency can help unlock partnerships between U.S. tech companies and local actors to solve specific needs, including by aiding companies in navigating local regulatory requirements and identifying reliable partners.
2. Unlock Strategic Capital
To bolster its AI companies on the global stage, Washington needs a coordinated effort to mobilize capital to deliver genuine economic value for developing and emerging economies. The U.S. government has long recognized the need to compete with China in the Global South on technology, but its efforts to promote U.S. technology overseas remain rigid, under-resourced, and fragmented, divided across multiple agencies including EXIM, the Development Finance Corporation (DFC), the U.S. Trade and Development Agency, and the Department of Commerce. Washington should empower a single point person at the White House to coordinate these institutions, with a specific focus on promoting U.S. AI exports. Furthermore, both the DFC and EXIM will be up for congressional budget reauthorization in the next two years, presenting opportunities to expand the scale of funding for AI infrastructure.
The U.S. government should also continue to update its export support programs to make them more flexible and better suited for advancing geopolitical goals. Washington should raise the cap on the DFC’s lending authority, direct it to prioritize AI and digital technologies, and scrap rigid income thresholds that can limit the agency’s flexibility in picking projects to support. EXIM, meanwhile, has made progress in this direction through its China and Transformational Exports Program (CTEP), which is designed to help U.S. companies compete with Chinese firms and strengthen U.S. leadership in ten strategic sectors, including AI. But as currently structured, CTEP does not have the scale or adaptability needed to compete with the level of support China provides for its exporters.
To strengthen CTEP’s impact, the U.S. government should revise it to scale up lower-cost, concessional financing—meaning loans with more favorable terms, such as lower interest rates or longer repayment periods—for data center projects in key regions. Congress should also consider excluding CTEP transactions from EXIM’s overall default rate cap (a measure of overdue payments), or raising that cap, to give the program more room to support higher-risk but potentially higher-impact projects.
In addition, EXIM could loosen its domestic content rules for AI-related projects abroad. Currently, at least 51 percent of a project’s value must come from U.S. goods or services (unless alternative criteria are met), a threshold that exporters and lenders say can make it hard to compete globally in complex, multi-country supply chains. Finally, the United States should expand and redirect CTEP’s capacity to more effectively channel financing toward digital infrastructure. Between 2020 and 2023, 88 percent of CTEP’s support went to just two sectors—renewable energy and wireless communications—leaving less than 12 percent for the other eight sectors, including AI.
Finally, Washington should consider working with multilateral institutions and external actors. In 2024, the World Bank authorized $117.5 billion in loans, grants, equity investments, and guarantees to partner countries and private businesses, compared to just $8.4 billion by EXIM. In addition to leveraging its greater financial capacity, the World Bank could address baseline infrastructure gaps, making data center deployment more feasible in emerging markets. Blended finance models that draw on these partners would enable U.S. technology deployment at both speed and scale, while also adding multilateral legitimacy to its projects. The United States should consider continuing to work on infrastructure projects in the Global South with states like the UAE, which can bring significant capital and strong diplomatic relationships to the table.
To be sure, working with external partners is not without its flaws. The UAE, for example, has been linked to destabilizing activities in many parts of Africa, including support for armed groups in Libya and Sudan, while its infrastructure firms have faced allegations of corruption and human rights violations while operating overseas. Multilateral organizations, meanwhile, are typically bound by mandates focused on development outcomes, not strategic competition, raising questions about how far Washington can push them to align with U.S. geopolitical objectives, especially when it comes to echoing the U.S. line on China. But for all their limitations, external partnerships will often unlock more resources than the United States can mobilize on its own.
3. Bargain Smartly
At the moment, the United States has considerable leverage: the world wants the United States’ AI offerings. But the United States should use that leverage wisely. Driving too hard a bargain—such as demanding full decoupling from the Chinese economy or sweeping trade concessions—is, in most cases, unlikely to bear fruit. Many countries will be reluctant to fully sever tech ties with China, especially if the United States is unable to offer viable alternatives across the full range of sectors from which it expects decoupling. Chinese firms are already embedded in some emerging markets’ tech ecosystems, from telecommunications infrastructure to data centers and smart cities. And a hardline approach could erode goodwill and diminish the appeal of U.S. partnerships.
In most cases, a more productive approach will involve limiting U.S. demands to the tech domain. Washington should push for alignment on export control enforcement and adherence to data center security protocols. It should require local companies receiving large quantities of AI chips to strip out any Chinese networking technology and cut financial or ownership ties with U.S. adversaries. And it could lobby for limits on the use of Huawei products in the public sector. But making more maximal demands of countries if they wish to access U.S. AI products is likely to backfire.
4. Embrace Sub-frontier Open-Source
Open-source AI models significantly lower the cost of AI adoption for many countries in the Global South. The Chinese government has been notably supportive of open-source AI—Foreign Minister Wang Yi, for example, emphasized its role in helping the Global South develop while speaking at the UN in September. Of course, open-source models also come with risks: Chinese researchers, for example, used Llama to develop AI models for military purposes, and analysts do not currently know the extent of cyber and biological risks associated with the most sophisticated models. U.S. tech companies should thus think very hard before open-sourcing models at the frontier. But once a Chinese open-source model is released, U.S. companies should immediately open-source their own model of equivalent (or marginally better) quality. OpenAI’s recent decision to release an open-weight LLM in the near future is a positive step in this direction, and Washington could help incentivize and encourage such moves.
5. Fewer Lectures, More Problem-Solving
To succeed in Washington, an AI export promotion strategy will need to be framed around competition with China. But while this may resonate on Capitol Hill, it will do little to win friends and influence in the Global South. In fact, an exclusive focus on China risks undermining the United States’ strategic appeal to states who, in the economic realm at least, have consistently resisted efforts by both Washington and Beijing to force a zero-sum choice in a great-power competition. In much of Asia, Latin America, and Africa, the United States will not gain much traction by simply demanding that countries forestall their deepening economic and technological relationships with Beijing. For many of these states, in the absence of a radically improved U.S. commercial and diplomatic presence in these regions, that is an unrealistic expectation and an unpersuasive pitch. It signals that the United States views these countries only as pieces to maneuver on the chessboard of U.S.-Chinese competition, not as autonomous partners of interest on their own terms.
The United States would do better to focus on its own counteroffer: how it can use its technological arsenal to speak to specific local economic and political interests.
The United States would do better to focus on its own counteroffer: how it can use its technological arsenal to speak to specific local economic and political interests. As Carnegie’s Evan Feigenbaum put it, “Washington has to outperform the Chinese competition, not just belittle and whine about it.” U.S. lectures about the exploitative, transactional, and untrustworthy nature of Chinese power have never been especially compelling to many audiences in the Global South, even in places where suspicion of Chinese intentions runs deep. As the United States zeroes out foreign assistance, cracks down on foreign students, restricts entry for citizens of over fifty countries across the Global South, and imposes punitive tariffs on key allies, swing states, and developing economies worldwide, Washington’s lectures will sound increasingly tone deaf to many global audiences. Unpredictability, unreliability, and the overt exercise of coercive economic power are unlikely to be the most effective ways to attract swing states and potential allies to join a U.S. technological ecosystem.
The United States is locked in a global economic and technological race with a serious competitor; winning will require showing up, country by country, with a real alternative.
