US and China Pursue Different AI Races as Chips, Open Models and Power Shape 2026

A Year After DeepSeek, the Competition Has Split

The United States attracted $175 billion in artificial intelligence venture capital investment during 2025, compared with $6 billion in China, according to China International Capital Corp. That is nearly 30 times as much money. Yet as of January 28, 2026, the country spending far less has become a major force in freely downloadable AI models, while American companies retain the lead in advanced chips and the most capable systems.

DeepSeek helped bring that contrast into focus. In January 2025, the startup from Hangzhou released a mobile app that shook Silicon Valley by demonstrating that large language models could rival ChatGPT in performance at a fraction of the cost, using less advanced chips and a much smaller budget. Large language models are the systems behind AI tools that generate text and answer questions.

The episode became known in technology circles as the "DeepSeek moment." Its impact went beyond a technical achievement: it challenged the assumption that American spending and access to the best hardware would keep competitors at a comfortable distance. Chinese developers showed that working under resource constraints could also encourage new approaches.

Scott Singer, a fellow at the Carnegie Endowment for International Peace, described how China has connected its AI progress to a broader diplomatic message.

China is definitely positioning itself rhetorically as a multilateral, open, and development-focused global leader," he explained. "China's geopolitical positioning on AI predates the Trump administration, but it increasingly resonates with global audiences.

Washington and Beijing are now pursuing different advantages. American companies concentrate on frontier performance, meaning the highest capabilities developers can achieve. China places greater weight on spreading usable AI through industry, infrastructure and international markets. Both approaches involve software, chips and money, but they place different bets on where AI will deliver power and economic value.

Washington Builds Its Strategy Around American Technology

The Trump administration set out its response in "Winning the Race: America's AI Action Plan," published in mid-2025. The document described a contest for "global dominance in artificial intelligence" and called for removing regulatory barriers while selling American technology abroad.

The plan promoted exports of the "full AI technology stack" to allies. A technology stack is the collection of parts needed to make a system work: in this case, hardware, AI models, software and standards. Washington's approach seeks to make American technology the foundation of AI development in partner countries. China was the unstated target of an effort to isolate it from the global AI supply chain.

Nvidia, headquartered in Silicon Valley, sits at the center of the American advantage. It designs graphics processing units, or GPUs, the chips that provide much of the computing power used to train advanced AI models. American firms hold an estimated seven-month lead over Chinese competitors in cutting edge capabilities, supported by strict export controls on advanced chips and semiconductor manufacturing equipment.

The United States also maintains approximately 90 percent control of advanced AI chip markets. Huawei's Ascend series is China's strongest domestic alternative. On basic tasks such as generating text, the Ascend 910 performs at roughly 60 percent of the capability of Nvidia's older H100 chips. Its disadvantage is greater in complex model training, the process through which an AI system learns from data.

Washington's strategy combines that hardware lead with a political offer: allies can obtain American AI infrastructure while adopting American standards. The administration presents this as a security priority, warning that reliance on Chinese technology could create dependencies that an adversary might exploit. Trade disputes and demands that partners follow restrictive American export policies complicate that offer.

The H200 Decision Exposes a Dependency Dispute

In December 2025, the Trump administration partially reversed its chip restrictions by announcing that Nvidia could sell H200 processors to vetted Chinese buyers in exchange for a percentage of sales revenue. The more advanced Blackwell and Rubin chips remained excluded.

Washington's calculation was that access to older American technology could make Chinese companies dependent on it and reduce the incentive to develop domestic alternatives. Critics warned that allowing more powerful processors into China could instead strengthen its AI sector.

Alibaba, Tencent and ByteDance were among the Chinese companies that had placed orders exceeding two million H200 chips, with a value of more than $50 billion, according to industry estimates. Those figures show the scale of demand for American computing hardware even as China invests in its own replacements.

Chinese customs officials reportedly instructed domestic companies to reject shipments in the week before January 28. Beijing's apparent resistance suggests that Chinese authorities also see a strategic risk in dependence on American technology. The resulting tension runs in both directions: Chinese companies want access to powerful chips, while the Chinese government wants greater technological independence.

The key developments unfolded over several dates:

• January 2025: DeepSeek's mobile app challenged assumptions about the cost and hardware needed for competitive AI.
• Mid-2025: Washington published "Winning the Race: America's AI Action Plan," promoting American leadership and exports of a complete AI technology stack.
• December 2025: The Trump administration announced permission for H200 sales to vetted Chinese buyers, while excluding Blackwell and Rubin processors.
• The week before January 28, 2026: Chinese customs officials reportedly told domestic companies to reject shipments.
• 2026: The United Nations is hosting the Global Dialogue on AI Governance as competing national approaches continue to develop.

China Expands Its Reach Through Downloadable Models

DeepSeek's R1, Alibaba's Qwen family and models from Moonshot AI represent another route to influence. Chinese companies have released models with openly available weights, allowing developers to download, modify and deploy them freely. The weights are the numerical settings an AI system acquires during training.

This approach differs from the closed, proprietary models maintained by OpenAI and Anthropic. Developers using downloadable models can adapt them for particular tasks and operate them on computing systems of their choosing. That makes them attractive to users seeking flexibility and lower costs.

Alibaba's Qwen models have become the world's most downloaded AI systems with openly available weights, recording approximately 700 million downloads tracked by Hugging Face. The Qwen ecosystem includes nearly 400 distinct models and more than 180,000 derivative versions customized for particular applications.

China's software strategy reaches precisely the markets Washington also hopes to influence: developing economies where price matters and Western licensing fees or export restrictions can limit access. Downloadable models offer those countries a route to AI capabilities without the same commercial or political conditions attached to American technology.

Chinese President Xi Jinping linked technological progress to national rejuvenation in his 2026 New Year's address. He said Chinese technology had "reached new heights," including breakthroughs in indigenous chip development.

Beijing connects this ambition to China's "century of humiliation," the period from the Opium Wars in the middle of the 19th century until 1949, when technological backwardness left the country vulnerable to foreign powers. In that view, AI is a decisive technology in determining whether China remains exposed to stronger rivals or reverses that historical relationship.

China Wants AI in Factories, Healthcare and Government

China's "AI Plus" initiative aims to embed artificial intelligence across industry, healthcare and governance by 2035, creating what Beijing calls a "fully AI-powered" society. Its focus is diffusion: spreading working systems through many parts of the economy rather than concentrating only on the most advanced model.

The United States is pursuing ever larger data centers and massive models that require unprecedented computing power. China places more attention on putting "good enough" AI into manufacturing, robotics and physical infrastructure. A system can deliver practical value in a factory or an autonomous machine even when it trails the strongest model on more demanding tasks.

Singer described the enthusiasm he sees around practical deployment in Chinese technology hubs as "much more excitement and energy around leveraging AI as it is today." His observation contrasts that activity with a greater focus elsewhere on speculative future breakthroughs.

Robotics, smart manufacturing and autonomous systems are areas in which China appears to be pulling ahead of Western counterparts. American investment, meanwhile, concentrates more heavily on expanding software and automating knowledge work. Chinese investment flows more toward robotics and industrial applications.

The two strategies therefore measure success differently. American developers seek advances in what models can do. Chinese companies and policymakers also seek gains from placing existing capabilities into machines, production lines and public services. Implementation speed can matter as much as the sophistication of the software when the objective is widespread adoption.

Power and Construction Limit the American Expansion

American cloud providers are projected to spend $600 billion on AI infrastructure during 2026, twice their expenditure in 2024. That spending supports the pursuit of larger computing systems, but electricity supply and construction schedules place hard limits on how quickly new capacity can become usable.

Training frontier models requires enormous data centers. Those facilities strain electrical grids already operating near capacity, leaving the United States with a mismatch between access to advanced processors and access to the power and buildings needed to run them.

China faces the reverse problem. It has abundant energy resources but lacks enough advanced chips to turn that capacity into competitive AI training infrastructure. Nvidia chief executive Jensen Huang has said China has twice the electrical generation capacity of the United States.

Speaking at the Centre for Strategic and International Studies, Huang described American chip technology as "generations ahead" while pointing to China's speed in building infrastructure.

"If you want to build a data centre here in the United States from breaking ground to standing up an AI supercomputer is probably about three years. They can build a hospital in a weekend," Huang said.

Huang's comparison captures two different constraints. The United States has superior chips but faces slow construction and pressure on electricity supplies. China can deploy physical infrastructure quickly, yet access to the processors needed for the most demanding AI work remains restricted.

This is asymmetric competition: the countries have different strengths and weaknesses rather than competing under identical conditions. American leadership in advanced capabilities and research sits alongside Chinese strengths in implementation speed and applications in the physical world.

Military Adoption Adds Another Race

The United States Department of Defense has launched an "AI-first" strategy built around seven "Pace-Setting Projects." The aim is to speed deployment across warfare, intelligence and enterprise functions, meaning the administrative and organizational work that supports the military.

Projects including "Swarm Forge" and "Agent Network" seek to integrate AI into battle management and decision support. "Ender's Foundry" focuses on military simulation enabled by AI. These programs extend the competition beyond commercial models to the systems armed forces use to plan, assess information and operate.

The Pentagon faces bureaucratic delays, including accreditation timelines and data silos. Accreditation is the approval process for using a system; data silos are collections of information kept apart in ways that make sharing and combined analysis difficult.

China uses military-civil fusion to channel civilian advances into military applications. Research by Georgetown University's Center for Security and Emerging Technology examined more than 2,800 People's Liberation Army contracts. It found a defense ecosystem anchored by state owned enterprises but drawing increasingly on innovators in the private sector.

China's centralized system can mandate adoption across military and civilian institutions at the same time. That gives deployment speed a central place in the competition, alongside the quality of the underlying technology.

Autonomous weapons, electronic warfare and command and control systems are among the applications that could shape future conflicts. A Council on Foreign Relations analysis warns that 2026 is a decisive phase in the movement from experimental AI tools toward autonomous agents capable of conducting complex military operations with minimal human oversight.

Security researchers have also described Chinese state sponsored cyber operations in which AI agents autonomously executed 80 to 90 percent of intrusion workflows. An AI agent is a system that carries out a sequence of tasks, rather than simply answering a user's question. That shift makes the speed and extent of adoption a security concern as well as a commercial one.

The United Nations Faces Competing Rules

The United Nations is hosting the Global Dialogue on AI Governance in 2026, creating the first truly global forum for coordinating standards, risks and norms. No specific meeting date was given. The effort seeks common ground among governments whose preferred rules remain far apart.

The European Union promotes regulation centered on individual rights, privacy and transparency about how algorithms work. The United States favors voluntary standards intended to preserve flexibility for innovation. China promotes "inclusive cooperation" while defending state control over data and deployment.

Atlantic Council analysts expect "global governance in form but geopolitical in substance." In practice, countries may cooperate on scientific assessments of AI while continuing to pursue rival national strategies for controlling and spreading the technology.

China's 2025 Global AI Governance Action Plan called for a "diverse, open, and innovative" ecosystem involving multiple stakeholders. Beijing's language of openness and multilateral cooperation contrasts with Washington's "America First" approach and appeals particularly to developing countries facing barriers to American AI.

Nearly 80 percent of China experts surveyed by the Mercator Institute for China Studies expect Beijing to make major advances during 2026 in diversifying export markets and deepening partnerships with developing nations. That expectation connects China's AI diplomacy to its wider effort to build international commercial relationships.

Washington's competing approach is "tech-stack diplomacy": partnerships that supply AI infrastructure to allies willing to adopt American standards. Both countries are trying to shape the systems others build, not simply the models their own companies release.

Several Contests, With Different Measures of Success

The United States leads in semiconductor design, frontier model capabilities and cloud infrastructure. China dominates adoption of openly available models, applications in the physical world, manufacturing integration and the energy resources available for expanding computing capacity.

Those strengths make the idea of a single AI race less useful. The competition resembles a decathlon, with several overlapping events and different leaders. A breakthrough in model performance, a widely adopted software platform and a factory using AI effectively each represent a different kind of success.

Xiaomeng Lu, director of Geo-technology at the Eurasia Group political risk consultancy, expects the countries to dominate distinct niches during 2026 rather than one achieving comprehensive superiority.

"The US has a definite lead in AI chips, though China is catching up in LLM, and is poised to get ahead in certain AI governance areas," she noted.

LLM means large language model. Lu's assessment brings together three separate contests: the hardware used to run AI, the capabilities of the models themselves and the rules governing their use.

Countries choosing an American technology stack gain access to leading capabilities but face high costs and political conditions. Countries adopting Chinese models with openly available weights receive accessible, customizable platforms, while potentially exposing themselves to security vulnerabilities and censorship mechanisms embedded in systems such as DeepSeek.

The development stakes are large. If AI becomes the defining technology of the 21st century, the economic gap between early adopters and countries left behind could exceed the divisions created by the Industrial Revolution. A fragmented ecosystem, with critical infrastructure operating on incompatible standards, would add another difficulty for governments and businesses deciding what to build.

DeepSeek's January 2025 breakthrough demonstrated that the barriers to entry had fallen. By January 28, 2026, the competition had broadened into a struggle over chips, software adoption, electricity, military deployment and international rules. American technological leadership and Chinese diffusion can advance at the same time, shaping different parts of the global AI economy.

Quick Facts

• DeepSeek's January 2025 release showed that Chinese models could rival ChatGPT at much lower cost using less advanced chips.
• US AI venture capital investment reached $175 billion in 2025, compared with $6 billion in China, according to China International Capital Corp.
• American firms hold an estimated seven-month lead in cutting edge capabilities, while the United States maintains approximately 90 percent control of advanced AI chip markets.
• Washington authorized H200 sales to vetted Chinese buyers in December 2025 while excluding the more advanced Blackwell and Rubin processors.
• Chinese companies ordered more than two million H200 chips worth over $50 billion, according to industry estimates; customs officials subsequently reportedly told companies to reject shipments.
• Alibaba's Qwen ecosystem has approximately 700 million downloads, nearly 400 models and more than 180,000 derivative versions.
• China's AI Plus initiative targets widespread adoption across industry, healthcare and governance by 2035.
• American cloud providers are projected to spend $600 billion on AI infrastructure in 2026, twice their 2024 expenditure.
• The Pentagon's AI strategy includes seven Pace-Setting Projects, while research into China's military procurement examined more than 2,800 contracts.
• The United Nations Global Dialogue on AI Governance takes place in 2026 amid competing American, Chinese and European approaches.