NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

A Startup Eight Months Old Raised Nearly 300 Million Yuan to Simulate Whole Cells

Yuanshi Technology raised nearly 300 million yuan in a Pre-A round to build AI virtual cells that predict how cells respond to disease and drugs, eight months after its founding.

2026-11-21 · 1119 words · NeuroAI
A Startup Eight Months Old Raised Nearly 300 Million Yuan to Simulate Whole Cells

A drug team had a compound that looked excellent in a test tube and then did almost nothing in a cell. The company that announced a round this week thinks that gap is a modelling problem, not a chemistry problem. It wants to simulate the whole cell before anyone touches a pipette.

Yuanshi Technology (元示科技), founded in Shanghai in April 2026, said on November 19 that it had closed a Pre-A round of nearly 300 million yuan (≈ US$41.7 million / HK$326 million, approximate). The company works on AI virtual cells (AI虚拟细胞), or AIVC, and says the round sets a record for a single financing in the field in China.

The round and the people in it

Legend Capital (君联资本) led. Following were Huaxia Lingtou, China Merchants China Fund, a fund linked to JD.com, CDH VGC and CMC Capital. Existing backers Monolith, Innovation Works, Daotong Investment and Yanyuan Ventures added more. Light Capital acted as sole financial adviser.

The speed is the striking part. In less than eight months the company has completed an angel round, an angel-plus round and this Pre-A, taking cumulative funding to nearly 400 million yuan (≈ US$55.6 million / HK$435 million, approximate). The two earlier rounds, announced in October, together were close to 100 million yuan (≈ US$13.9 million / HK$109 million, approximate).

That is an unusual pace for a company whose product is not yet a registered drug, or even a published model. It reflects how seriously investors are taking the idea that the next step in AI-driven drug discovery is not a better molecule but a better cell.

What an AI virtual cell is

Most AI drug companies work at the level of molecules. They predict protein structures or score how well a small compound might bind to a target. The problem is that a good binding score does not guarantee a useful effect inside a living cell, where signalling pathways, metabolism and context all intervene.

A virtual cell tries to model that intervening layer. Yuanshi says it wants a system that can predict how a cell responds to a disease state, an experimental condition or a drug intervention — effectively moving expensive trial-and-error out of the laboratory and into computation.

The company describes its architecture as a biological raw data engine plus a universal cell model (通用细胞大模型). The first part generates the evidence; the second learns from it. This is why the company talks about "second-generation AI drug discovery" (第二代AI制药): the first generation optimised molecules, the second simulates the system those molecules enter.

The loop that does the work

The central mechanism is a cycle the company calls a dry-wet loop (干湿闭环): the model predicts, a wet experiment checks the prediction, the result feeds back, and the model updates. The company describes this as recursive self-improvement (递归自我改进) applied to a biological foundation model.

The argument for this design is subtle. More data is not automatically better data. Measuring similar cells repeatedly may add thousands of rows without telling you anything about a rare state or an intervention that matters. What matters, in the company's framing, is evidence that links cell context, intervention, matched control and measured response.

So the model is supposed to identify where its own predictions are weakest and recommend the experiment with the highest information gain. The wet lab then runs it. The result both checks the prediction and becomes new training material. In principle, each round should be more useful than the last.

In practice, a loop like this lives or dies on throughput. A model that recommends clever experiments is only as fast as the lab that performs them. That is where the hardware partnership comes in.

The lab equipment angle

On November 6, during the ninth China International Import Expo, Yuanshi announced a strategic partnership with Yongsheng Intelligent, a subsidiary of MGI (华大智造), a Chinese sequencing and life-science equipment maker. The two plan to build automation for the virtual-cell workflow, covering sample handling, transport by automated guided vehicles and scheduling driven by AI.

MGI also released a single-cell product called C6, which the company says can capture more than a million valid cells in a single run. High-throughput measurement is the fuel for a model that claims to learn from experiments rather than from public datasets.

The CEO, Wu Jiaqi (吴佳奇), holds a doctorate in systems biology from Peking University and did three years of postdoctoral work at Rockefeller University. She was the second author on a 2021 Nature Communications paper on yeast cell fate regulation. The chief technology officer, Dong Mingze (董明泽), studied at Peking University's Yuanpei College, earned a doctorate in computational biology at Yale, did postdoctoral work at Stanford and worked at the Arc Institute on virtual-cell models including STATE and STACK before moving into AI safety at Anthropic. He joined full time on November 11.

What the money buys

The proceeds are earmarked for four things: continued accumulation of biological data, construction of an automated experimental platform, iteration of the universal cell model, and hiring. In other words, most of it goes into the loop rather than into a single asset.

That is the right allocation for this kind of bet, and also the riskiest. A virtual cell is valuable only if its predictions hold in real drug development. The test is not a benchmark score but whether a partner can reduce failed experiments. No such validation has been published.

Honest limitations

Funding amounts, the "single-round record," the zero-shot generalisation claim and the description of the dry-wet loop come from company announcements reported by Chinese business and technology media, including Tencent News, Zhitong She and National Business Daily. I found no audited filing, no peer-reviewed paper on the model and no independent confirmation of the record claim.

The academic credentials and past publications of the two founders are reported in the same coverage and appear consistent across sources, but I did not verify the papers directly in a journal database. The MGI partnership and the C6 single-cell figure are company statements from the same period.

No clinical, preclinical or regulatory result is described here because none was reported. A large early round is evidence of investor interest, not of a working model.

What readers can do now

  • If you work in drug discovery, ask a virtual-cell vendor for a prospective test, not a retrospective score. Predicting a result you already measured is not the same as guiding the next experiment.
  • If you follow AI funding, read the allocation. Money going into automated labs is a different signal from money going into marketing.
  • If you are a researcher, watch for a publication. A model that genuinely generalises across cell types and perturbations should eventually leave the investor deck and enter the literature.

Related coverage

Cite this article

NeuroAI (脑机网), "A Startup Eight Months Old Raised Nearly 300 Million Yuan to Simulate Whole Cells," 21 November 2026, https://neuroai.site/a/na-capital-virtual-cell-funding

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Capital & Markets” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories