The conventional framing of the US-China AI competition is a race to the frontier: which country can build the most powerful models, attract the most talented researchers, and deploy the most advanced chips. By that measure, the United States retains a meaningful lead. But a McKinsey analysis published in Fortune by senior partners Joe Ngai and Nick Leung argues that this framing misses something important: China is not trying to win the same race. It is competing on entirely different terms, and on those terms, it is doing exceptionally well.
The analysis, drawn from McKinsey’s extensive research into China’s technology sector, identifies three structural advantages that distinguish China’s AI trajectory: a relentless focus on efficiency over raw capability, a deliberate open-source distribution strategy that prioritizes global adoption over monetization, and a deployment environment that has produced real-world AI integration at a scale that no other country has matched.
The Efficiency Imperative
The efficiency orientation of Chinese AI development is not a choice born of preference, it is a response to constraint. Cut off from the most advanced chips by US export controls, Chinese AI laboratories have been forced to extract more performance from less hardware. The result has been a series of architectural and training innovations that have compressed the cost of producing capable AI models to a fraction of what US frontier labs spend.
DeepSeek’s V4 model, released in April 2026, is the most prominent example. Trained at a reported cost that is orders of magnitude lower than comparable US models, it demonstrated that the efficiency gap between Chinese and American AI development had narrowed dramatically, and in some dimensions, reversed. The model’s performance on standard benchmarks was competitive with GPT-5 and Claude Sonnet 4, despite being produced with a fraction of the compute.
This efficiency advantage has direct commercial implications. Chinese model providers can offer API access at prices that US providers struggle to match without eroding their margins. The cost differential is not a temporary promotional strategy, it reflects a genuine difference in production economics.
Open Source as a Distribution Strategy
The second pillar of China’s approach is the systematic use of open-source releases for global distribution. By releasing model weights publicly, Chinese AI companies have made their technology accessible to developers worldwide without requiring those developers to send data through Chinese infrastructure or enter into commercial relationships with Chinese companies. The result is a form of soft adoption that builds familiarity and dependency without triggering the data security concerns that would otherwise limit uptake.
Alibaba’s Qwen series, DeepSeek’s V4, and MiniMax’s models are now among the most downloaded on Hugging Face globally. On OpenRouter, four of the five most popular models by usage are Chinese. This distribution footprint has been built not through marketing or enterprise sales, but through the simple mechanism of making good models available for free.
The McKinsey analysis notes that this strategy is particularly effective in markets where cost sensitivity is high, and data sovereignty concerns are lower, which describes a large portion of the global developer community outside the United States and Europe.
Deployment at Scale
The third advantage is perhaps the hardest to quantify but the most consequential in the long run: China has deployed AI at a scale and depth of integration that has no parallel elsewhere. EastFrontier has previously reported on the explosion of China’s token economy, with daily AI API calls rising from 100 billion to 140 trillion in fourteen months. That volume of real-world usage generates training data, identifies failure modes, and drives product iteration at a pace that laboratory benchmarks cannot capture.
Chinese consumers and businesses have integrated AI into workflows ranging from customer service and content creation to industrial process control and medical diagnosis. The breadth of that integration means that Chinese AI companies are receiving feedback signals from a far wider range of use cases than their US counterparts, whose deployment is more concentrated in software development and enterprise productivity tools.
The Question of Ceiling
The McKinsey analysis does not argue that China has overtaken the United States in AI capability, it argues that the question of who is “winning” depends entirely on what you are measuring. If the measure is frontier model performance on academic benchmarks, the US leads. If the measure is cost-efficient deployment at global scale, China is ahead. If the measure is the breadth and depth of real-world AI integration in the domestic economy, China has no close competitor.
The more uncomfortable implication of the analysis is that the US strategy of restricting China’s access to advanced chips may be accelerating rather than slowing the efficiency-driven approach that makes Chinese AI competitive on its own terms. Every constraint that forces Chinese engineers to do more with less produces innovations that, once made public through open-source releases, become available to the entire world. The efficiency gains that export controls were designed to prevent are instead being shared freely, and adopted widely.