Yes, the reference is to China—the “nanosecond behind” competitor in the global AI race. There is mounting evidence that increasingly supports Jensen Huang’s claim that the gap is not only closing, but closing fast. The reality is that Chinese AI is advancing at a pace that has outstripped even the most optimistic forecasts (Steiber & Teece 2026).

A glance at the most downloaded and highly rated models on Hugging Face in February 2026 tells a clear story: Chinese labs are now consistently claiming a significant share of the platform’s top-ranked positions. In some weeks, that presence seems unmistakable, with four of the five most prominent training models on this platform originating from Chinese labs. The shift became more evident when Alibaba’s Qwen overtook Meta’s Llama last September to become the most downloaded family of LLM on the platform.

Chinese AI models have effectively closed the gap. In some cases, they moved ahead of global peers not only in technical performance but also in actual adoption. What this really means is that the shift is already underway within leading organizations. For example, Qwen models are being deployed for customer service at Airbnb, used by Amazon for robotic simulation, and selected by Apple to handle Siri’s AI requests in the Chinese market. The Allen Institute adopted it to build multimodal systems, and both Nvidia and Perplexity have cited Qwen in parts of their own development efforts.

The Qwen panic rattles Silicon Valley, and it now redefines the AI race. Similar to Qwen, other models from Chinese labs, such as Deepseek, Minimax, iFLYTEK, ERNIE (Baidu), Doubao (ByteDance), GLM-4.5 (Zhipu AI), Hunyuan (Tencent), and Kimi (Moonshot), are gaining presence among an array of Fortune 500 companies. China released 1,500+ LLMs last year, equivalent to 40% of the total models released globally by 2025. China now boasts 16 of the world’s top 25 AI models when ranked on core capabilities such as reasoning, knowledge, mathematics, and coding

Not only in the West, but also in the Global South, the use of Chinese platforms is rapidly increasing. For example, the usage of DeepSeek in Africa is two to four times higher than in any other region, with 18% market share in Ethiopia, 17% in Zimbabwe, and with even larger shares in countries where U.S. technology is restricted. The rapidly growing AI industry in China was estimated at $160–$170 billion last year, with 5,300 AI enterprises, the most AI patents, and the most AI adopters in the world.

Over the past twelve months, this study has conducted 40 in-depth interviews with global AI experts closely engaged with China’s AI ecosystem. The ongoing research suggests that Chinese models are steadily narrowing the gap in the global AI race. It is not through a single breakthrough, but through a coordinated set of innovation and management initiatives, discussed below.

How is China Narrowing the AI Gap?

Open Weight Models

Models from Chinese labs follow a distinctly open approach to innovation. They are typically open-sourced, offered at little or no cost, and designed to be readily customized by developers who want direct access to the underlying code. The expectation is that adoption will extend well beyond China, with developers and firms in other countries building a wide range of applications on top of these Chinese models. For example, Pinterest uses DeepSeek to provide its AI-powered shopping assistance services largely for two reasons: performance economics and capability. These models are freely downloadable and customizable, delivering about 30 percent higher accuracy than leading off-the-shelf options, and cutting costs sharply—sometimes by as much as 90 percent compared with proprietary alternatives. Hence, Chinese AI models are the default choice for global AI systems.

Innovation Ecosystem

China has successfully built its scientific and innovation ecosystem around three pillars: project-based funding, STEM education, and incentives (e.g., preferential procurement, subsidized land, rent-free premises, tax incentives) for private ventures. For example, local governments in Shanghai, Guangdong, and Beijing compete to provide subsidized land to start-ups, and the semiconductor industry enjoys a whopping 120% tax relief on R&D. This ecosystem reflects a distinctly engineering-led, state-coordinated innovation approach to solving physical and social challenges—one that contrasts sharply with the U.S. system, where legalism and a more adversarial regulatory climate dominate, as Dan Wang argues in his book Breakneck (Yamakawa & Davenport, 2025).

AI Talent Cultivation

China’s sustained emphasis on AI skills has been central to building deep AI capabilities. More than 535 universities in China offer AI majors, combining machine learning, computer vision, cognitive reasoning, natural language processing, robotics, and multi-agent systems into a single, scalable skill set. As Tsinghua University and Zhejiang University now produce engineers on par with Stanford University and MIT, the talent gap that once shaped the AI race has largely disappeared, with more PhDs in Science, more AI research output, and more than 53,000 AI researchers. Overall, a tightly coordinated project funding and a deeper talent pool encourage the growth of new ventures and create a self-sustaining ecosystem that advances both research breakthroughs and industrial adoption.

State-Led Diffusion

China’s priority lies less in symbolic breakthroughs (e.g., AGI or human-level cognitive abilities) and more in diffusing AI broadly across the economy and society. The government identifies open source as the key strategy for both domestic and international diffusion.  Its AI+ Initiative, launched in August 2025, aims to embed AI across nearly 90% of the economy by 2030. Its regulatory approach is comparatively enabling as it focuses on business productivity rather than irrelevant benchmarks, as Jeffrey Ding notes in his book “Technology and the Rise of Great Powers: How Diffusion Shapes Economic Competition” . That is, commercial AI activity is permitted so long as it operates within clearly articulated and approved boundaries. It protects new ventures from unnecessary lawsuits when they introduce their innovations in the market. For example, in serving markets, Pony.ai, which operates robotaxis across China, prioritizes a positive regulatory environment. Similarly, Apollo GO and WeRide have been hugely beneficial in this positive environment and, after successful trials, they are now competing in Europe and the Middle East.

Infrastructure

China’s advanced ICT infrastructure is based on three pillars: expansive 5G networks, high-capacity data centers, and mature cloud computing ecosystems. That foundation is reinforced by the world’s most comprehensive renewable energy supply chain, which is expected to deliver roughly 60% of all new renewable capacity installed globally between now and 2030. Although it lags behind in chips and computing power, its homegrown alternatives, such as Huawei’s Ascend 910C, now deliver roughly 76% of Nvidia’s H200 processing power and about two-thirds of its memory bandwidth. In addition, the CloudMatrix 384 is emerging as a domestic alternative to Nvidia’s GB200 NVL72. It is an AI system designed to compete at scale while relying on entirely homegrown chipsets. Baidu has advanced this progress by unveiling M100 AI chips and planning to build a supernode capable of supporting millions of chips by 2030. It signals a long-term push toward infrastructure-scale AI capability.

Market Orientation

Chinese companies are moving quickly from models to markets, rolling out AI applications at scale for actual use. Manufacturing, e-commerce, and robotics are already embedding these systems into main operations, not experimental pilots. According to Open Router, a marketplace for AI models, the global use of Chinese models has skyrocketed to nearly 30% in late 2025, up from a miniscule 1.2% in late 2024. Most firms apply the classic formula for business sustainability: reducing Capex and R&D costs, offering competitive services at a fraction of OpenAI’s costs, and capturing a broad customer base. For example, DeepSeek built its V3 model for roughly $6 million, while Meta has spent tens of billions on AI with far fewer visible returns. Due to a thinner investor base, the new generation of AI unicorns in China—StepFun, Zhipu AI, Moonshot, Minimax, 01.AI, and Baichuan —have no choice but to achieve market performance quickly.

Opportunities for Managers and Leaders

The Chinese innovation ecosystem offers several lessons for managers, business leaders, and policymakers worldwide on implementing AI. From the factory floor to the operating room to the supply chain, we propose a set of practical strategies to solve real-life problems and transform societies.

Chase the Model

It is already evident that Chinese labs develop efficient models using far fewer resources. In practice, that means managers should chase and embed the best models directly into organizations’ core missions and workflows. It is the only way to achieve productivity across inventory management, education, and healthcare. For example, Alibaba reports that its Qwen3-Max-Thinking model outperformed leading U.S. competitors, and that the model has been adopted by Airbnb to develop its customer service agents.

Tap the Talent

The ongoing AI war has made it clear that breakthrough innovation ultimately depends on AI talent. For example, Meta undertook a multibillion-dollar hiring push last year, deliberately recruiting AI researchers to accelerate its ambitions in superintelligent systems. Similarly, the recent AI summit in India attracted around $200 billion in investment in AI, chasing mostly AI talent, such as data scientists, cloud engineers, designers, security specialists, etc. As AI adoption accelerates across autonomous driving, smart manufacturing, and healthcare, leaders seek talent with broader, cross-disciplinary skill sets from anywhere in the world.

Build the Collaboration

Managers should be ready for a future in which dominant innovations and industrial ecosystems increasingly determine standards, platforms, and strategic priorities to address societal needs. Managers should pursue open, responsible collaboration with leading AI labs across countries, paired with clear safeguards. For example, AMD, Cisco and HUMAIN collaborate with Saudi Arabia to build a robust AI infrastructure, Microsoft partners with Japan to boost skills and cybersecurity and the U.S. State Department partners with major tech firms (e.g., Amazon, Google, IBM, Meta, OpenAI and others) to address sustainable development goals in the developing world.

Learn the Infrastructure

Managers and policymakers can learn from the infrastructure layer of Chinese AI, which intertwines semiconductors, cloud platforms, data centers, and the energy systems that sustain them. The key takeaway for managers is that permanent value is likely to be created through a robust infrastructure, regardless of which models or countries ultimately lead the AI race.  A deeper understanding of the cost-effectiveness of Chinese infrastructure can also protect firms from the AI bubble, as the financial commitments to build AI infrastructure currently stand at $1 trillion.

Open the Gate of Diffusion

Open weight models and the rate of diffusion are directly correlated. It means the model parameters are publicly available, permitting developers to fine-tune them efficiently for specific applications. Managers of proprietary models can consider how to reduce costs and lower entry barriers. Diffusion occurs when free lancers to tech titans can autonomously run the model and adapt to new use cases while owning the data. For example, Alibaba has open-sourced nearly 400 Qwen models, catalyzing an ecosystem that has already produced more than 180,000 derivative versions. It clearly shows how open weights can accelerate diffusion at scale.

Check the Safety

While open-weight models offer the benefits of reproducibility and efficiency, it is critical for managers to assess AI safety, as these models are susceptible to data breaches and evidence of bypassing guardrails. We call for continued monitoring of how these models are deployed and adapted in real-world use cases, alongside systematic collection of evidence on their performance, reliability, and safety. This might help address risks such as autonomy risks (e.g., rogue AI), economic disruption risks, and misuse risks, as Amodei discusses in his essay “The Adolescence of Technology”.

The fate of the evolving AI race will be determined by leaders’ dynamic capabilities to implement AI across sectors through an open, multilateral, and development-focused ecosystem. Embedding AI into everyday life is a practical way to address societal needs. The future of AI need not be framed around a particular country. Treating it instead as a shared global race for human progress may be the more effective way to move the technology—and its benefits—forward.

References

1. Annika Steiber and David J. Teece, “The Dynamic Capabilities of Chinese Companies: Leadership in an Age of Disruptive Innovation, Emerging AI, and Global Competition.
2. Yasuhiro Yamakawa and Thomas H. Davenport, “Your AI Strategy Needs to Expand Beyond the U.S. and China,” Harvard Business Review, December 2, 2025.