Chinese AI labs are producing open-source models that rival Western competitors in reported efficiency, Beijing is filing hundreds of generative AI services for domestic deployment, and Chinese models have captured roughly 15% of global usage. Yet U.S. export controls continue to tighten, chip shortages persist, and earnings forecasts for major Chinese tech firms have swung from growth to decline. The question facing investors, policymakers, and technologists is whether China’s AI push represents a durable shift in global tech power or an expensive bet with uncertain returns.
Open-Source Models Close the Gap
Two model families illustrate how Chinese labs have narrowed the distance with American rivals. The DeepSeek-V3 series, described in a technical report on its Mixture-of-Experts design, activates only a fraction of its parameters for any given task, reducing the compute needed per query and promising lower inference costs. Its authors detail training token counts, hardware budgets, and scaling strategies that suggest competitive efficiency relative to frontier U.S. models, even if those claims have not yet been fully validated by independent benchmarks. Separately, Alibaba’s Qwen2.5 family documents significant increases in pretraining dataset scale and post-training refinements across multiple parameter sizes, signaling that Chinese internet platforms are willing to invest in full-stack AI capabilities rather than relying solely on licensed foreign systems.
That engineering effort is translating into measurable adoption. Chinese AI models reached approximately 15% global share by November 2025, a rise driven largely by DeepSeek’s open-source distribution and aggressive community outreach, according to TrendForce’s reporting on usage metrics. One analyst told CNBC that most of the world’s population could be running on a Chinese software stack within five to ten years, assuming cost advantages outweigh concerns about data governance and geopolitical risk. That assumption deserves scrutiny: open-source availability lowers barriers to experimentation and deployment, but it does not erase the trust deficit many governments and enterprises feel toward Chinese-origin software, especially for sensitive workloads in defense, critical infrastructure, or political communication.
Export Controls and the Chip Bottleneck
Washington’s primary tool for slowing China’s AI progress is semiconductor export restrictions. The U.S. Department of Commerce’s Bureau of Industry and Security has issued an export-control package explicitly designed to restrict China’s capability to produce advanced semiconductors for military applications, with rules that directly affect AI accelerators and high-bandwidth memory. A subsequent round of measures strengthened foundry due-diligence requirements and expanded entity listings to prevent diversion of advanced chips to the PRC through third countries or shell companies. Reporting from DW and other outlets underscores a structural constraint: China possesses deep pools of AI talent, massive data, and ample energy, but its domestically produced hardware still trails the cutting edge, and U.S. controls limit access to the latest Nvidia and AMD designs that power frontier-scale training runs.
Beijing is responding with infrastructure rather than waiting for immediate chip parity. A RAND analysis notes that China is building a National Integrated Computing Network intended to pool private and public computing resources into a shared infrastructure layer, effectively virtualizing scattered data centers into a coordinated national asset. The aim is to maximize the output of whatever hardware is available, even if individual chips lag behind U.S. and Taiwanese offerings, by improving utilization, scheduling, and access for priority projects. The United States has authorized some downgraded Nvidia chip sales to China, but the broader restriction architecture remains intact, leaving Chinese labs to balance frontier ambitions with second-best hardware. Whether domestic chip substitution and pooled infrastructure can compensate for the performance gap at the very top end is the central technical uncertainty in this competition, and it will shape how far Chinese models can push context length, multimodality, and agentic behavior.
Regulation, Governance, and Global Ambitions
China is not just building models; it is building rules. Beijing’s Interim Measures for Generative AI Services govern all public-facing generative systems in mainland China, requiring lawful training data sources, content governance aligned with socialist core values, labeling of synthetic media, and security assessments with algorithm filing for services that could influence public opinion. By the end of 2024, official figures from the Cyberspace Administration of China showed that 302 generative AI services had completed national filing, with 238 new filings in 2024 alone, and 105 apps or functions had completed local registration for calling filed models via API. That regulatory throughput signals a state apparatus actively managing AI deployment rather than simply promoting it, embedding political and security priorities into the technical rollout of chatbots, copilots, and content tools.
Internationally, Beijing frames AI as a development opportunity rather than a purely competitive race. China’s foreign ministry has outlined an capacity-building initiative that offers workshops, education resources, literacy programs, and data-sharing arrangements to developing countries seeking to adopt AI. A separate governance action plan released through the World Artificial Intelligence Conference process emphasizes national sovereignty, respect for different social systems, and a balance between development and safety. These documents aim to shape international AI norms before Western-led frameworks become defaults, giving countries in the Global South an alternative template that is less focused on rights-based regulation and more on state-led coordination and infrastructure support.
Profitability, Hype, and Strategic Risk
Behind the technical progress and diplomatic outreach lies a thornier question: can China’s AI boom generate sustainable returns? Analysts cited by Bloomberg note that consensus earnings expectations for major Chinese internet and hardware firms have shifted from double-digit growth to outright declines, even as capital expenditures on AI infrastructure surge. One recent newsletter described China’s AI future as a long road to small profit, arguing that heavy investment in data centers, model training, and subsidies has not yet translated into commensurate revenue growth for platform companies or cloud providers. Price competition among model vendors, generous free tiers to attract developers, and the difficulty of monetizing consumer chatbots all weigh on margins, raising the risk that AI becomes another low-return infrastructure race rather than a profit engine.
This gap between investment and payoff is not unique to China, but Chinese firms face additional headwinds. Domestic regulation constrains some of the most lucrative use cases, such as personalized political advertising or unfiltered social feeds, while global clients remain wary of embedding Chinese AI deeply into mission-critical systems. An analysis on OODA Loop characterizes China’s AI surge as a mix of genuine capability and substantial hype, noting that impressive demos coexist with structural constraints around chips, capital efficiency, and international trust. For investors, that means headline-grabbing model releases and state-backed initiatives do not automatically translate into equity upside, particularly when export controls and domestic competition compress pricing power.
How Much Power Really Shifts?
The strategic stakes extend beyond corporate balance sheets. A report summarized by Intellectia argues that a global tech landscape shift is underway as cost considerations increasingly dominate technology choices, with enterprises in emerging markets more willing to adopt non-Western stacks if they are cheaper and “good enough.” In that context, China’s bet on open-source models, subsidized infrastructure, and bundled cloud-plus-application offerings could win substantial share in regions where regulatory alignment with the West is weak and budget constraints are tight. If Chinese vendors can combine competitive performance with lower total cost of ownership, they may not need to match U.S. frontier models parameter-for-parameter to become the default in education, local government, and small-business software across large parts of Asia, Africa, and Latin America.
Yet the durability of any power shift will depend on factors that go beyond benchmark scores or the number of filed services. Export controls will continue to shape the ceiling of Chinese model capabilities, especially for agentic AI that requires large, continually updated models with strong reasoning and tool-use skills. Domestic governance will influence how creative, autonomous, and politically sensitive applications can become within China’s borders, affecting the feedback loops that often drive rapid improvement. International partners will weigh offers of training, infrastructure, and low-cost models against concerns about surveillance, data localization, and long-term dependence on a single supplier. China’s AI surge is therefore best understood not as an inevitable takeover or a passing bubble, but as a contested, path-dependent experiment, one that could leave the country with a powerful, state-shaped AI ecosystem that is globally influential, modestly profitable, and permanently constrained at the very frontier by forces outside Beijing’s direct control.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
