China and the US Find Common Ground on the Application of Artificial Intelligence in Healthcare

The differences in artificial intelligence (AI) development between the US and China were measured, not assumed. In August, the RAND Corporation center focusing on AI, security, and technology published a detailed count of companies in both countries, based on what each firm creates, sells, and patents. A total of 1181 companies were identified, with 743 based in America and 438 in China.

The main finding of the study is as follows: sixty-one percent of American AI companies specialize exclusively in software. In contrast, only twenty-six percent of Chinese firms fall into this category. The remaining companies in China are actively involved in creating physical carriers for their algorithms, including robots, vehicles, and drones.

Despite this key difference, RAND researchers found that the two industries are very similar in most aspects. Approximately 28% of American and 27% of Chinese firms use transformer architecture, which is the basis for systems like ChatGPT and DeepSeek. Furthermore, about nine out of ten American AI companies and three out of four Chinese ones sell ready-made products using AI—such as diagnostic tools or logistics control panels—instead of providing direct access to raw models.

Even the so-called 'AI arms race' between leading labs turned out to be a localized phenomenon: only about 19% of companies in each country are developers of foundational models that train general models from scratch, similar to how OpenAI or Alibaba do. Most of the other companies quietly develop software for hospitals, factories, and insurance companies.

The deep divergence between the two markets manifests in the application of their AI. Chinese firms focus on the manufacturing sector, transportation, and energy—that is, the physical economy. American companies, however, concentrate on healthcare, scientific research, and cybersecurity—which corresponds to the knowledge economy. Moreover, the share of companies dealing with humanoid robots in the Chinese sample is more than five times higher than in America: 12% of Chinese AI firms versus only 2% of American ones. At the same time, America produces significantly more startups specializing in diagnostic software.

The reasons for this division can be explained by the theory of division of labor proposed by the Scottish economist and philosopher Adam Smith. Since China maintains the world's densest production supply chain—including motors, batteries, actuators, and sensors—its AI naturally integrates into machinery. Moreover, there is state industrial policy that clearly points in this direction: in 2025, China declared 'embodied intelligence' a national priority and subsequently published a national standard system for humanoid robots.

Additionally, over half of China's AI companies in the RAND sample are part of large conglomerates such as Alibaba and Tencent, which provides them with an existing production base or fleet for implementing developments. America's advantage works in reverse. Its capital markets can invest tens of billions of dollars in a single language model hoping for a breakthrough, taking on this risk not the government, but Wall Street. America's deep experience in medicine, science, and security provides rich, high-value tasks for software AI to solve. This can be seen as a comparative advantage updated for the algorithmic age: no one is smarter, they simply specialize in what their own economy offers.

There is another quiet race running parallel to the story of robots versus chatbots—the battle over price. RAND tracked real developer traffic on OpenRouter, a gateway that directs API calls to hundreds of AI models. At the end of 2024, American models processed about nine-tenths of all traffic, while Chinese models were almost invisible. However, by April 2026, the situation had leveled out: the ratio became 49% to 47%. This shift was driven by open Chinese models, such as DeepSeek-V3 and the Qwen series, which achieve results close to the best American systems in independent benchmarks, while the overall cost per token is one-fourth or one-fifth the cost of American counterparts.

It is important to note that Chinese labs have not suddenly surpassed competitors at the highest level of intelligence; American systems still lead in raw intelligence metrics. Nevertheless, the principle of 'good enough, four times cheaper' represents the quiet, cumulative gain that changes the market from the bottom up, following the harsh logic that economist Joseph Schumpeter called 'creative destruction.'

These trends do not describe a sprint where one player wins. They demonstrate two nations solving different parts of the same problem, and it is in this gap that the most effective and economically viable opportunity for cooperation lies. These areas include testing the safety of humanoid robots when operating near people, creating common standards that allow a sensor created in Shenzhen to interact with software written in Seattle, and joint AI research for drug discovery and disease diagnosis.

Geoffrey Hinton, the 2024 Nobel laureate in Physics for creating the mathematical foundation of modern neural networks, stated directly that the technology is developing faster than most people expected, including himself, and that the risks are too great to manage solely through competition. This is an engineering fact.

The British mathematician and logician Alan Turing, writing in 1950, put it simply: 'We can see only a short distance ahead, but there is much to be done.' Seventy-five years later, one country taught its machines to speak and reason, while another taught them to walk and lift heavy objects. Ultimately, something will have to do both, and then the two parts of this story—mind and body—will finally meet.