CCTV names ten Chinese AI startups, with Shuimu Molecule the lone AI pharma pick

CCTV has named its list of ten "AI startup steeds," with ten companies emerging from a field of 1,200 entrants across robotics, smart manufacturing and life sciences. Only one of them is an AI drug-discovery company: Shuimu Molecule.

The company is barely three years old. It was incubated by Tsinghua University's Institute for AI Industry Research (AIR), founded by academician Zhang Ya-Qin, with professor Nie Zaiqing as chief scientist and Xing Jie as COO. It runs two tracks at once: a conversational drug R&D platform called ChatDD, and its own pipeline of innovative drugs for autoimmune disease and cancer.

The sector's core bottleneck is transferring expert judgment. Drug development is governed by a rough "rule of three tens": about ten years, more than a billion dollars and a success rate below 10%. Models can predict structures, generate molecules and rank candidates, but researchers often cannot see why a ranking came out the way it did, and have no easy channel to feed their own concerns back into the model.

Shuimu Molecule frames the field in four generations: TMDD, which rests on experiment and expert instinct; CADD, where computers assist; AIDD, which applies AI to property prediction and molecule generation; and ChatDD, which the company proposed in 2023. In ChatDD, an expert states a question and a goal in natural language, the model decomposes the task and calls tools, and results come back for another round of discussion.

Underneath sits the model layer. BioMedGPT, advanced jointly with Tsinghua AIR in 2023, aligns natural language with biological data such as molecules and proteins; BioMedGPT-10B is described as the first open-source, commercially usable 10-billion-parameter multimodal biomedical model. MutaPLM, a NeurIPS 2024 paper, maps textual descriptions of mutation effects onto protein mutations, both explaining what a change might do and proposing candidate mutations from a stated goal.

Unlike teams that only sell tooling, Shuimu Molecule runs the platform inside its own programs, using it for R&D decisions, molecular design and project operations. The company describes this as a two-wheel model in which the platform and the pipeline validate each other.

The more interesting shift is in how competition is framed. The article argues the benchmark is moving from raw prediction capability to who can organize an entire research workflow, in effect becoming the "agent OS" entry point for drug discovery, with overseas efforts such as GPT-Rosalind and Claude Science exploring similar interaction models. CCTV's list is real public validation for Chinese AI pharma, but clinical progress and success rates remain the dividing line.