Aureka raises $100M to build a biological world model for drug discovery

Why it matters

Aureka is betting that proprietary wet-lab feedback will matter as much as model scale in AI drug discovery. The $100 million round funds a direct test of that thesis.

Weian Zhao, the founder and CEO of Aureka Biotechnologies, closed a $100 million Series B on August 10th to train larger biological foundation models and connect them more tightly to the experiments that determine whether their molecular designs work.

The financing announcement says Granite Asia funded the first tranche exclusively. An unnamed strategic investor led a later tranche, joined by HighLight Capital, MPCi and NRL Capital. Aureka says it has raised nearly $200 million since its 2023 founding. Aureka did not disclose its valuation or identify the strategic investor.

Zhao came to the problem through chemistry, translational medicine and prior biotechnology work. A tenured pharmaceutical sciences professor at the University of California, Irvine, he studied chemistry at Shandong University and McMaster University before completing postdoctoral work across Harvard Medical School, Brigham and Women's Hospital and the Harvard-MIT Health Sciences and Technology program. His UC Irvine research has covered diagnostics, biosensors, microfluidics and cell therapies.

That career also exposed the gap between a promising laboratory result and a medicine that can reach a patient. Zhao has said that families contacted him after reading his scientific papers, hoping his work could immediately help sick relatives. He repeatedly had to explain that the research was years away from clinical use, according to an interview with VCBeat Health. Aureka grew from his attempt to replace some of drug discovery's slow trial-and-error work with higher-throughput design and testing.

The laboratory is Aureka's data strategy

Aureka will spend the Series B primarily on research and large-scale training for models covering molecular design, biological structure and function prediction. Aureka also plans to expand Lab-in-the-Loop, its system for passing model-generated molecules and hypotheses through single-cell functional screening and high-throughput experimental validation.

The feedback loop is central to Zhao's pitch. Models propose molecules, Aureka tests them, and the experimental results become training material for the next round. That gives Aureka a potential source of proprietary functional data instead of leaving its models dependent on public datasets that competitors can also access.

Aureka laid out an earlier version of that strategy in a November 2024 investor presentation. At the time, Aureka reported more than 20 employees and partnerships with five pharmaceutical groups and five biotechnology companies. Those figures were presented by Aureka and were not independently audited. The deck described an ambition to learn protein design rules from large volumes of sequence, binding and function data generated by high-throughput experiments.

The new "biological world model" language extends that thesis. Zhao is positioning Aureka's models as systems that could eventually reason about interactions among molecules, predict the consequences of biological interventions and direct AI agents through successive design and testing cycles. Current capabilities remain narrower. The near-term work centers on biomolecular structure prediction, antibody design and experimental feedback.

OpenDDE gives outsiders something to test

Aureka published OpenDDE-preview on GitHub, an open-source, all-atom biomolecular model for drug discovery. The repository's preview status means its interfaces and checkpoints remain subject to change.

Aureka describes OpenDDE as the open-source version of AuraIDE, its proprietary biological foundation model trained partly on protein co-evolution data. The repository uses co-folding to model interactions among proteins, DNA, RNA and small-molecule ligands.

OpenDDE has received useful external validation. Tamarind Bio's evaluation examined OpenDDE on FoldBench v1, an antibody-antigen structure-prediction benchmark. A benchmark result does not establish that Aureka can produce a safe, effective drug. Drug candidates still have to survive laboratory replication, animal testing, manufacturing, clinical trials and regulatory review. Aureka has not tied the benchmark result to a clinical-stage candidate or human efficacy data.

Capital is concentrating around integrated drug-design engines

Aureka is raising into a market where investors increasingly favor platforms that combine models, proprietary data and experimental execution. Isomorphic Labs raised $2.1 billion in May to scale its drug-design engine and therapeutic pipeline. Chai Discovery announced a $130 million Series B in December 2025 after developing models for molecular and antibody design.

Aureka's distinction rests on Zhao's attempt to build the computational model and experimental feedback machinery as one system, with an emphasis on antibodies and single-cell functional screening. The Series B syndicate also reflects Aureka's cross-border structure: Aureka is based in Laguna Hills, California, and operates research and development in China, while its new backers bring substantial experience in Asian technology and life-sciences investing.

Aureka says pharmaceutical collaborations generated tens of millions of dollars over the past two years. Aureka has not named those customers in the Series B announcement or provided audited revenue, contract duration or the portion tied to milestone payments. The number indicates commercial activity, while offering limited visibility into repeatable revenue or drug-program progress.

The $100 million gives Zhao room to test the premise that experimental feedback can compound into a durable model advantage. OpenDDE has established a credible technical starting point and made part of Aureka's work inspectable. Zhao's larger goal now depends on whether the loop produces molecules that keep working as they move from a benchmark to Aureka's laboratory and eventually into patients.