Tokenpolitik: How the United States Can Compete with China to Build the Global AI Stack

There is a new form of great power competition that is set to define the future of international system. While the United States and China were negotiating behind the scenes to hold formal talks on artificial intelligence, President Xi Jinping made a public speech proclaiming Beijing as the leader for a new global AI order. The race is on to see who will build the network of datacenters, power plants, and models that are redefining power and international relations.

That race pits U.S. firms against state industrial policy in China that is actively promoting its national champions to build lower-cost chips and high-performance open-weight models as alternatives to U.S. technology. At the same time, Beijing has accused Washington of “AI hegemonism” amid possible U.S. investigations into how Chinese firms trained their models using knowledge distillation.

The Trump administration has identified the magnitude of the challenge and undertaken a number of initiatives to ensure the United States remains the world’s AI superpower. But these efforts have yet to coalesce into a broader grand strategy that shapes how the United States engages with the world beyond its borders.

What is missing is a clear concept of “tokenpolitik”: a grand strategy that bridges multiple agencies within the U.S. government with the private sector to create a new economic and security architecture for the twenty-first century. This strategy would focus on AI’s role in defining power and interest in the decades ahead and provide a framework for the U.S. government to back its private sector firms building datacenters, power plants, and research labs. This would allow U.S. firms to curate the data and new models necessary to ensure a free society and free market ushers in the AI era.

What Is New Is Old: Infrastructure as Strategy

Realpolitik is about power, but power has never existed only in the barrel of a gun, and has often extended to infrastructure. The French-led Suez Canal project changed the geography of trade between Europe and Asia. Britain opposed its construction, then purchased Egyptian government’s 44 percent shareholding in 1875 after the canal became essential to British commerce and imperial communications. The strategic question extended beyond building the canal to who would finance, govern, secure, and benefit from a new commercial artery.

The Cold War saw similar examples. The U.S.-backed International Telecommunications Satellite Organization (INTELSAT) and Soviet-sponsored Intersputnik system competed to extend satellite communications across the world. The wider contest involved satellites, earth stations, orbital access, service prices, standards, technical assistance, and relationships with developing countries. Communications infrastructure became a means of organizing international dependence and influence.

Artificial intelligence is producing a similar contest. Models are the most visible products, with close models from firms such as Google, OpenAI, and Anthropic competing against open-weight models from China (e.g., Kimi, DeepSeek, Qwen, and GLM). Yet, the deeper struggle concerns the layered system that makes them useful at scale: energy, chips, infrastructure, and applications. Together with models, these layers create a token supply chain that runs the AI applications increasingly common in government and the workplace.

Tokens are the language and currency of the new AI order. A token is the fundamental atomic unit that a neural network ingests, processes and generates. In large language models (LLMs), it is the smallest unit of a text, generally a word, and in large audio-language models (LAMs), it becomes the smallest discrete unit of a continuous sound wave. Tokens represent a basic fragment of text, audio, or image translated into mathematical vector representations for a neural network to process. In the context of global competition, these tokens act as the ultimate industrial output of the massive datacenters and power grids dotting the globe. Every operational constraint in modern artificial intelligence, from the real-time cost of running an inference query to the finite memory of a context window, comes down to how efficiently a system parses and predicts these units. They function as the foundational currency of this new technological era.

The great power that most effectively generates tokens through a network of state agencies and private sector firms will sit at the commanding heights of a new international system. The primary strategic question of the twenty-first century is who will build the token supply chain and wire key regions of the world for an AI future.

China Is Backing Firms Across the Full Stack

Though China’s approach is often described as state-led, private firms remain central to innovation. The state shapes the environment around them by supplying capital, subsidized compute, infrastructure, research support, diplomatic backing, and protection from foreign pressure.

Beijing is deploying industrial-policy tools across the full stack. These efforts include state-sponsored labs alongside major infrastructure investments including the $8 billion National AI Investment Fund in 2025 followed by a broader $295 billion spending plan for AI infrastructure buildout in 2026. Academic research on Chinese government venture capital similarly shows that public investment operates through a distributed system of national and local funds, helping some initially weaker firms attract private follow-on financing. The Chinese Communist Party also uses state-backed efforts to promote chip makers such as CXMT, which soared more than 466 percent on its stock market debut, YMTC, and infrastructure firms including Huawei.

Open-weight models provide a second pillar of Beijing’s strategy. They can be downloaded, locally hosted, and more easily optimized for enterprise use cases. RAND research has argued that such models function as instruments of soft power by shaping developer communities, technical education, standards, and downstream applications.

Recently, open-weight models emanating from Chinese labs have been shown to perform well in key AI benchmarks, on par with the US frontier AI models. Furthermore, prices per token are cheaper compared to the U.S.-based models. This puts extra pressure on the U.S. closed-source models as global adoption of open-weight models increases. There is now a growing influence of Chinese developers in the open-model economy. In fact, many U.S. enterprises started to experiment with these Chinese AI models, though they host these models on U.S.-based platforms.

Huawei provides a physical distribution network for these models. Its established position in telecommunications allows it to offer governments packages combining datacenters, cloud services, accelerators, networking, cooling, software, training, and maintenance. In developing economies, a technically weaker chip can remain attractive when embedded in an affordable package from a supplier that already understands the national network and relevant ministries. China, through Huawei, has been installing these relationships in markets such as Brazil, Indonesia, Kenya, and the Middle East. The diplomatic narrative centered on access, sovereignty, open development, and capacity building in the Global South complements this trend.

Washington Has Assembled Many of the Pieces

The United States starts with stronger private companies and deeper capital markets. U.S. investors accounted for more than half of global outgoing AI venture-capital investment in 2025, while infrastructure and hosting attracted the largest share of AI capital. Washington has now moved beyond the export-control-only strategy that dominated the Biden administration’s AI diffusion policy. The Trump administration, in line with the AI Action Plan, established the American AI Exports Program to promote full-stack American artificial intelligence technology packages to international allies and partners. These efforts rely on a mix of initiatives across the Departments of Commerce and State, including Pax Silica, which focuses on building “a secure, resilient, innovation-driven technology ecosystem” with allies and partners.

Furthermore, there are now growing number of bilateral agreements with key actors. The U.S.-United Arab Emirates framework links a planned regional AI cluster with security, monitoring, and investment commitments. Commerce has authorized large chip purchases by G42 and Saudi Arabia’s Humain, subject to reporting and security conditions. 

The U.S. International Development Finance Corporation (DFC) and the Export-Import Bank of the United States (EXIM) provide financing foundations for these initiatives. DFC reports that AI datacenter investment has become the leading request from partner governments and identifies telecommunications, fiber, cloud infrastructure, datacenters, generation, and grids as components of the stack. EXIM’s ExportAI initiative offers export-credit insurance, working-capital guarantees, loan guarantees, and direct lending for U.S. technology.

Still, what is missing is an overarching grand strategy that extends efforts such as Pax Silica and the AI Action Plan into a larger theory of victory for the global competition with the Chinese Communist Party. That plan must find additional ways to coordinate efforts across agencies and back U.S. firms that are building the AI stack of the future. Such a strategy will require a more deft approach to token diplomacy.

From One-Off Programs to a Grand Strategy

The United States needs tokenpolitik: a grand strategy connecting State, Commerce, Treasury, DFC, EXIM, allied governments, and private firms to make secure, affordable AI capacity available to strategic partners. This strategy should seek to outflank China’s Digital Silk Road and build out the infrastructure that will power economic growth and opportunity in the Global South while also assuring China doesn’t gain a dominant infrastructure foothold.

The White House should begin by establishing an international AI infrastructure strategy program jointly accountable to the National Security Council and National Economic Council. The Department of State should identify priority countries, negotiate operating and security arrangements, and lead embassy deal teams. Commerce should assemble private consortia, evaluate markets, coordinate technical standards, and manage licensing. These efforts need to embrace the private sector and bring in leading companies to provide key insights, since these firms will ultimately design, own, and operate the AI stack.

This plan should build on Pax Silica to envision regional orders that create a new economic and security architecture. For example, the effort could align U.S. firms with Turkish infrastructure companies and Azerbaijan energy giants to build out an AI cluster in the Caucasus. Combining this with Armenia’s recent bet on AI compute can align with the Trump Route for International Peace and Prosperity (TRIPP) as well as recent normalization of relations between Turkey and Armenia. The United States could work with Japan, including the Japan Bank for International Cooperation and firms such as NEC Corporation, to build out a cluster in Southeast Asia. These efforts would pivot from building the token supply chains envisioned by Pax Silica to promoting a U.S.-led, AI-empowered future economic and security order.

With a standing set of strategy guidance managed by the National Economic Council and National Security Council, the emphasis should turn to establishing joint infrastructure projects that support growing the AI stack in the Global South. There is now a risk of a larger global digital divide that could leave the Global South in the dark; DFC should finance host-country infrastructure such as power generation, cable landings, and datacenters required to generate tokens and power AI-led economic growth. EXIM can offer direct loans, loan guarantees, and insurance to U.S. firms so they can compete against China in building this infrastructure in the Global South and regional clusters.

Washington should also connect technology policy with financial intelligence. Treasury’s Outbound Investment Security Program already covers certain U.S. investments involving Chinese semiconductors and AI. Commerce collects information through licenses, end-use checks, red-flag guidance, and trusted-operator programs. This data should support an annual Treasury-Commerce assessment of how national funds, local-government vehicles, state banks, public procurement, and private capital finance the Chinese AI stack.

In addition, this effort should include leveraging a mix of anti–money laundering authorities and counter-smuggling efforts such as Operation Gatekeeper, which interdicted Chinese technology smuggling. Beijing uses a complex web of shell companies, falsified paperwork, disguised end users, third-country transshipment, bribes, and blackmail to obtain restricted AI hardware. The United States can use these authorities such as those under the Financial Crimes Enforcement Network to combat these tactics and issue AI-specific red flags for banks, payment processors, and even transportation firms and datacenter operators.

Congress should work to make an AI strategy durable. It should authorize multiyear DFC-EXIM financing and invest in the electricians, grid engineers, network operators, cooling specialists, and semiconductor workers needed for the buildout. It should streamline duplicative regulation while preserving competition, privacy, cybersecurity, and critical infrastructure protections.

A New Era of Tokenpolitik

Over the next generation, tokens will power economic growth and change the balance of military power. This fact makes tokens the focal point for defining interest in terms of power. Failing to embrace this reality and build a new grand strategy is a recipe for decline. 

To win the coming race, the United States needs to outmaneuver Chinese state capitalism. A new generation of strategists inside the executive and Congress, alongside broader networks connecting industry, academia, and think tanks, needs to create a blueprint for great power competition based on the source of power and interest for a new era. That plan should guide the diplomacy and development of building a new economic and security architecture.

Benjamin Jensen is director of the Futures Lab and a senior fellow for the Defense and Security Department at CSIS. Yasir Atalan is a deputy director and data fellow in the Futures Lab at CSIS.