With an eye on America, China is beefing up its AI-controlled labs

China’s March five-year plan called for faster exploration of AI-driven science as part of a push to “seize the commanding heights in the industrial application of AI”. The government is building an AI-for-science “stack” that links specialised datasets, models and compute to autonomous labs. In April the Chinese Academy of Sciences launched ScienceOne (Panshi, or “Bedrock”), available to hundreds of universities and laboratories; it includes agents trained on academic reports and non-text data and, state media say, can perform hundreds of tasks such as working out amino-acid sequences. A University of Hong Kong researcher who uses it praises the bundle of datasets, agents and compute: “I don’t see anything like it in the US.” The article says China’s scientific models still tend to lag America’s, and that China’s labs are unlikely for now to produce big scientific breakthroughs, but could supercharge manufacturing and reduce dependence on Western technology. Xtalpi, a drug-discovery firm, runs an autonomous Shenzhen lab that Li Qiang visited; last year it generated millions of possible molecules and whittled them down to hundreds before testing remaining candidates for a form of stomach cancer on lab-grown tissue. It now serves 17 of the world’s 20 largest pharmaceutical companies. Goldman Sachs says that last year about half of all drugs licensed globally, by value, were from China, up from 21% in 2024. In March Eli Lilly secured rights to develop and sell potential drugs from Insilico Medicine, a Hong Kong-listed firm that uses automated labs, in a deal that could be worth up to $2.75bn. In materials science China publishes four times as many papers as America, according to Nature; Xtalpi says a joint facility with JinkoSolar will test 1,000 solar-cell prototypes a day, versus one or two for an average human researcher. America is pursuing a similar government-led effort, the “Genesis Mission”, unveiled late last year. A difference the paper stresses: American automated labs raised about $1bn of venture capital last year, often for speculative bets such as room-temperature superconductors, whereas Chinese firms tend to pursue research that can quickly generate profit. Zhou Bowen of Shanghai AI Lab is among those who think models must first master the sciences, and that automated labs will help them do so.