Chai's the Limit: $130M To Build a CAD Suite for Molecules
Chai Discovery just closed a $130 million Series B to speed up R&D and push commercialization of its AI platform for de novo antibody design. The company's ambition is clear: create the equivalent of a computer-aided design suite for molecules, where you specify properties upfront and generate candidates that fit.
With this round, Chai's total funding climbs above $225 million and the company is now valued at $1.3 billion. For product teams, the signal is strong: the spec-first, simulation-heavy model that software enjoys is moving into therapeutics.
What's new: zero-shot antibody design that ships
- Chai-2 is a zero-shot generative platform producing de novo antibodies with double-digit success rates.
- Reported 16% success across ≤20 molecules per target on 52 diverse targets-over 100x better than prior computational methods. The design-to-wet-lab loop finished in under two weeks (bioRxiv).
- Updated data: >86% of full-length mAbs show strong developability on par with therapeutic antibodies; resolved structures align with in silico predictions, indicating atomistic accuracy.
- Scale: initial benchmark on 50+ targets; internal testing now on 100+ targets.
Why product teams should care
- Spec-first development: Define target class, affinity, specificity, and developability early. Generate candidates that meet those specs instead of endless tweak-and-test.
- Shorter build-measure-learn loops: From design to wet-lab validation in weeks, not quarters.
- Higher-fidelity simulation: Structural agreement between predictions and resolved designs raises confidence in pre-experiment decisions.
- Small teams, big throughput: Chai operates with ~25 people-a hint at what lean, model-centric orgs can ship.
How they're building it
Chai is led by co-founders Joshua Meier (CEO), Matthew McPartlon, Jacques Boitreaud, and Jack Dent. Meier's path runs through Harvard, OpenAI (GPT-1/2 era), Meta's generative biology group (co-led ESM1), and Absci (as chief AI officer) where McPartlon led de novo antibody modeling. Boitreaud previously scaled AI tools for small molecules at Aqemia; Dent built product and engineering teams at Stripe.
As Meier puts it, "If these models can understand natural language, why can't they understand DNA and proteins?" The bet: language-model scale and training methods translate to biology-and hit an inflection point for drug discovery.
Funding and market signal
The round was co-led by Oak HC/FT and General Catalyst. Participants include Thrive Capital, OpenAI, Dimension, Menlo Ventures, Lachy Groom, Yosemite, Neo, SV Angel, with new investors Emerson Collective and Glade Brook Capital Partners. Chai also added Mikael Dolsten, MD, PhD (former Pfizer CSO) to its board, plus Annie Lamont (Oak HC/FT) and Hemant Taneja (General Catalyst).
Context: VC funding in AI drug development hit $2.7B through the first three quarters of 2025, and AI-native biotechs see a near 100% valuation premium over conventional peers (PitchBook).
Product implications: what to do next
- Define the spec sheet early: Target, epitope strategy, affinity thresholds, developability, manufacturability, and safety flags. Treat it like a product PRD.
- Instrument the loop: Automate wet-lab queues, data capture, and feedback into model training. Latency and throughput are now product levers.
- Evaluate models with real KPIs: Success rate per target class, diversity of hits, off-target risk, developability metrics, and time-to-validation.
- Build vs. partner: If you lack high-quality assay data or ML infra, partner first. Own your data flywheel as you scale.
- Governance: Put in model risk, audit trails, and versioning. Designs will become regulated artifacts.
- Team design: Cross-functional pods: ML, protein engineering, automation, and product ops. Keep the team small, shipping weekly.
Risks to track
- False confidence: Good in silico fit doesn't guarantee in vivo success; keep orthogonal assays in the loop.
- Throughput mismatch: Model speed outpacing wet-lab capacity creates bottlenecks-plan capacity and prioritization.
- IP and data security: Training data provenance and design ownership need clear policies.
- Regulatory readiness: Traceability from prompt to protein is essential for filings.
What to watch over the next year
- Therapeutic full-length mAb case studies entering later-stage validation.
- Performance on harder target classes and combination modalities.
- Adoption inside large biopharma pipelines; APIs that plug into existing ELN/LIMS and automation stacks.
Why this matters
Chai's approach moves drug design closer to software: define specs, generate candidates, validate fast, and feed results back into the model. If that loop holds, discovery becomes a product engine, not a scavenger hunt.
If your team is skilling up on AI-native product workflows, explore role-based programs here: Complete AI Training - Courses by Job.
Your membership also unlocks:
