How Shenzhen can win the ‘AI for science’ race

The next big competition in artificial intelligence is over science, shifting focus from the largest model to the ability to use AI for faster scientific discovery. The United States has made this shift explicit with its Genesis Mission, partnering with Japan to create an integrated discovery platform that aims to double productivity and impact of US research and innovation within a decade.

China is following suit but increasingly through its cities. Beijing's 2025 plan calls for scientific-intelligence infrastructure, at least 10 high-quality databases, more than 10 million users, and eight benchmark applications by 2027. Shanghai's "Hundred Teams, Hundred Projects" initiative offers support for up to 70% of approved investment.

Shenzhen, with the highest R&D intensity among Chinese cities, has a unique opportunity to leverage its comparative advantages in building a city-level foundation for AI-for-science. Its strong R&D ecosystem, with over 93% of research spending coming from companies, positions it well for turning AI predictions into reliable scientific tools.

Shenzhen's physical technology base, producing a significant number of industrial and service robots, drones, servers, integrated circuits, and 3D-printing equipment, can serve as tools for AI to perform scientific tasks. Additionally, Guangming Science City offers major facilities for synthetic biology, brain research, and materials genomics, serving over 200 universities, research organizations, and companies.

To succeed, Shenzhen should build a shared city-level foundation for computing, scientific data, model access, simulation, experiment scheduling, equipment interfaces, and traceability. This approach allows different scientific fields to build specialized layers on top of the foundational system. Shenzhen should prioritize automated experimentation as its signature capability, focusing on areas with both scientific facilities and industrial depth, such as materials and synthetic biology.

By demonstrating a complete loop from AI prediction to real experiments and data feedback, Shenzhen can measure success by faster, cheaper, and easier-to-replicate experiments. Lastly, Shenzhen should design for interoperability from the beginning, developing common interfaces, data formats, and testing methods alongside technology, aiming for local and international adoption.

By turning fragmented scientific resources into a reusable system, Shenzhen can achieve its goal of making AI move from prediction to experimentation, where tools built in Shenzhen can be used to move from an idea to an experiment and back again.