Where is the US-China AI 'race' heading in 2026?

One year ago this week, Silicon Valley and Wall Street were shocked by the release of China's DeepSeek mobile app, which rivaled US-based large language models like ChatGPT by showing comparable performance on key benchmarks at a fraction of the cost while using less-advanced chips.

DeepSeek opened a new chapter in the US-China rivalry, with the world recognizing the competitiveness of Chinese AI models, and Beijing pouring more resources into developing its own AI ecosystem.

In its AI action plan released several months after DeepSeek's emergence, President Donald Trump's administration laid out the stakes in stark terms: "The United States is in a race to achieve global dominance in artificial intelligence (AI)."

The plan, titled "Winning the Race: America's AI Action Plan," called for rolling back regulatory barriers that allegedly hinder innovation and leveraging the dominance of US tech around the world.

To all countries "willing to join America's AI alliance," Washington said the US must begin by "exporting its full AI technology stack," which includes "hardware, models, software, applications, and standards" to "stop ... strategic rivals from making our allies dependent on foreign adversary technology," in a veiled reference to China.

Around the same time, Beijing published its "Global AI Governance Action Plan," using a less adversarial tone, calling for the creation of a "diverse, open, and innovative" AI ecosystem that would fully leverage the role of "multiple stakeholders" and "jointly promote international exchanges and dialogue on AI governance."

"China is definitely positioning itself rhetorically as a multilateral, open, and development-focused global leader," Scott Singer, fellow in the Technology and International Affairs Program at the Carnegie Endowment for International Peace, told DW.

"China's geopolitical positioning on AI predates the Trump administration, but it increasingly resonates with global audiences with the Trump administration's focus on AI dominance and America first," he added.

"If the question is, 'is China pursuing a strategy of AI dominance,' certainly not. It is going to seize leadership through the opportunity to fill in the gaps that the US has left in retreating from many of the spaces in the international order," said Singer, whose work centers on global AI development and governance with a focus on China.

US dominates in computing

Despite the popularity of DeepSeek and other Chinese large-language models (LLMs), which compare in capability with OpenAI or Gemini on simple tasks like creating text, translation and running chatbots, US-based tech titans are set to continue dominance in computing power needed to train frontier AI models.

The most advanced AI chips are produced by Silicon Valley-based Nvidia, and the US is right now far ahead of China on scaling data centers.

China has pools of AI talent, massive amounts of data and energy resources, but it is missing the most high-tech chips needed for computing power. US export controls on advanced AI chips and semiconductor kit have maintained this advantage.

Huawei's Ascend series is arguably China's most advanced family of AI chips. It includes models like Ascend 910, which testing by DeepSeek showed to have 60% of the performance of Nvidia's older H100 chips on executing text generation, responding to chatbot queries or classifying images.

However, the Chinese chips still fall far short on the more demanding task of training models or ecosystem development. Huawei is also unable to produce as many chips as Nvidia, so clustering them together to improve performance becomes more difficult and costly.

In December 2025, the Trump administration announced a major reversal by allowing the sale of Nvida's H200 series to vetted Chinese buyers in exchange for a cut of the sales.

In holding back Nvidia's most advanced Blackwell and Rubin chips, the Trump administration is betting it can preserve the lead while nudging China toward developing a reliance on older US-based tech. Critics warn that the H200 is still highly capable, and could erode the current US advantage in computing.

Chinese firms like Alibaba, Tencent and ByteDance have already ordered more than 2 million of the H200 chips at an estimated value topping $50 billion (around €43 billion). 

However, right now, the chips are being blocked from entering China. Last week, the Reuters news agency reported that Chinese customs officials had been instructed to tell Chinese companies not to buy the chips. It is unclear how long the chips will be blocked, or if it amounts to a formal ban.

Chinese firms throw open weight around

China's leadership is prioritizing building semiconductor self-sufficiency.

But even without the most advanced AI chips, "good enough" Chinese AI is taking off, with open-weight Chinese models like DeepSeek, Moonshot AI and Alibaba's Qwen allowing developers to roll out AI applications at a lower cost and with fewer entry barriers.

"Open weight" means the parameters, known as "weights," are public, allowing developers to efficiently fine-tune models for specific tasks, whether it be producing code or summarizing reports.

Alibaba's Qwen model family is currently the most downloaded open-weight model worldwide, with around 700 million downloads, according to data from the US-based AI collaboration platform Hugging Face reported by China's Xinhua news agency. This beat out Meta's open weight family, LLaMA.

According to the Qwen team, Alibaba has open-sourced nearly 400 Qwen models and those models have "spawned more than 180,000 derivative versions."

In his New Year's address, Chinese President Xi Jinping said that in 2025, Chinese tech had "reached new heights."

"Many large AI models have been competing in a race to the top, and breakthroughs have been achieved in the research and development of our own chips," Xi said.

A report by the National Center for AI Standards and Innovation in the US outlined the importance of open models "founded on American values," noting that "open source and open-weight models could become global standards in some areas of business and in academic research" and that the government should "create a supportive environment for open models."

At the same time, it warned about the "security shortcomings and censorship" found in popular Chinese models like DeepSeek, which "may pose a risk to application developers, to consumers, and to US national security."

"So much of the race for AI diffusion will depend on productization. How much developers can build on your product, and how much consumers like using your product, will shape the AI adoption race," Singer said.

China looks at 'physical world' AI applications 

The success of Chinese LLMs has driven Beijing's ambition toward a domestic rollout of AI that prioritizes real-world applications such as in robotics and manufacturing.

In 2026, China is set to move toward implementing the "AI Plus" initiative, which aims to embed AI across multiple sectors like industry, services, health care and governance as part of economic modernization. The long-term action plan calls for a "fully AI-powered" society by 2035.

"The US and China are placing fundamentally different bets across the AI stack," said Singer.

"US companies are leading the race to scale software capabilities and automate many tasks that can be done on a computer. China, however, is investing substantially in AI-powered robotics," he said.

"Integrating AI into the physical world is the heart and soul of their AI policy, as there is a belief it is going to solve challenges and problems today. AI-powered robotics is an area where China seems to be running ahead of Western counterparts," Singer said, adding that China's acute economic problems like youth unemployment and sluggish consumer demand are driving policymakers.

Is there a finish line?

Looking at the overall AI landscape in 2026, constraints to developing LLMs like power availability and construction times for data centers could determine the pace of development and adoption of newer AI models in the US, just as China continues to lag behind on chips in developing its own ecosystem.

As the US and China's priorities on advancing AI diverge, what could emerge moving forward is both dominating AI niches rather than one dominating the whole.

"That seems to be the trend — 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," Xiaomeng Lu, director of Geo-technology at the US-based political risk consultancy Eurasia Group, told DW.

And China is set to continue positioning itself as a champion of what it set out in its 2025 AI plan as "creating a diverse, open, and innovative ecosystem."

"That's our base case for 2026. As long as China's open-source mode stays closely behind top US models in capabilities, that's the likely approach of Chinese government to AI governance," she added.

Singer pointed out that the experience on the ground in Chinese and US tech hubs allows a sense of where the AI zeitgeist is right now.

"What's so different is the vibe at the level of the ecosystem; where if you go to Silicon Valley and you walk into San Francisco airport, you can't help but feel that there's this belief that continuing to scale the capabilities of the most advanced foundation models will yield transformative effects to the way we operate as a society and it could yield tremendous geopolitical benefits.

"Whereas if you compare that to the conversation in China, across Beijing, Shanghai, Shenzhen, Zhejiang, the areas where the technology is being developed, there is much more excitement and energy around leveraging AI as it is today."

Edited by: Keith Walker