How a TechBio Raised $100M in 4 Months to Build Data Power Plants for a Biological World Model

More than two decades ago a compound synthesized at Celera Genomics’ lab later became the blockbuster cancer drug Ibrutinib, illustrating how a modest chemical tool can evolve into a high‑value therapeutic.

After a series of financing rounds, Aureka (寻明生科) closed a $100 million Series B in just four months, funding three large‑scale “data power plants” in California that continuously feed experimental results back into its AI models.

Entering the AlphaFold era, the company shifted focus from small‑molecule screening to antibody design, arguing that the expanding design space of bispecific and multi‑specific antibodies demands AI‑driven large‑scale search rather than manual trial‑and‑error.

To close the gap between in‑silico predictions and real‑world efficacy, Aureka built a Lab‑in‑the‑Loop system that uses single‑cell micro‑fluidic droplets—100 000‑fold smaller than traditional 96‑well plates—enabling million‑to‑tens‑of‑million parallel assays and dramatically lowering per‑candidate cost.

The core model, AuraIDE, adopts a full‑atom tokenization that separates backbone and side‑chain atoms into distinct tokens, allowing the network to learn richer structural patterns. Trained on 100‑120 billion tokens with 700 million parameters (over twice the size of AlphaFold 3), OpenDDE (the open‑source version) achieved a DockQ success rate of 76.1 % on FoldBench v1, far surpassing AlphaFold 3’s 47.9 %.

Aureka’s commercial approach combines “selling the engine” – offering custom AI services, licensing, and deep‑tech collaborations – with “selling the car” – licensing validated drug assets. The company now has eight disclosed antibody candidates (including multi‑specific, agonist, and pH‑switch formats), two of which are IND‑ready for clinical trials in 2026, and has generated multi‑million‑dollar revenue from BD and NewCo partnerships.

Looking ahead, Aureka aims to move beyond static 3D structure prediction toward a “biological world model” that can ingest dynamic experimental feedback, discover unknown biological patterns, and illuminate new disease mechanisms, positioning AI as a discovery engine rather than merely a speed‑up tool.