Lila Sciences Raises $235M to Scale AI Science Factories for Materials and Drug Discovery

Lila Sciences Raises $235M to Build AI-Driven Science Factories

Lila Sciences has secured $235 million at a roughly $1.23 billion valuation to speed up lab discovery with AI and automated experimentation. The Massachusetts-based startup emerged from stealth in March after a $200 million seed round and is now scaling what it calls "AI science factories."

The company's platform trains models on academic literature across materials, chemistry, and life sciences, then validates model-generated hypotheses in automated labs. Results are fed back into the models to improve the next round of designs.

How the model works

Traditional research cycles-form hypothesis, gather data, run experiments, iterate-take months or years. Lila is compressing that loop by combining software with automated lab execution and continuous feedback.

"If your training input is entirely publicly available data, then one hits a ceiling," said co-founder and CEO Geoffrey von Maltzahn. The feedback loop is intended to push past that ceiling by generating proprietary data and testing it at scale.

What's new in this round

• Expansion of AI-enabled facilities where humans and software run parallel research programs.
• Opening parts of the platform to outside partners by year-end, enabling companies to use Lila's models and labs.
• Head-to-head trials pitting internal research teams against Lila's AI to test "scientific superintelligence," similar in spirit to chess matches between humans and IBM's Deep Blue.

Early output, but no products yet

Since its 2023 founding, Lila says it has generated and tested thousands of novel proteins, nucleic acids, chemistries, and materials. None are commercialized yet, and the company has not disclosed specific partners.

Investors and why they're interested

The round was led by Braidwell and Collective Global, with participation from ARK Venture Fund, General Catalyst, and Flagship Pioneering. Collective Global co-founder and co-CEO Daniel Adamson called Lila "an IP factory par excellence," highlighting the appeal of a faster path to patents across materials and therapeutics.

Implications for research leaders

• Shorter experimental cycles: closed-loop design-make-test-learn pipelines can cut weeks or months from iteration timelines.
• Data advantage: proprietary experimental data becomes a compounding asset that improves model accuracy and lowers search costs.
• Cross-domain discovery: the same infrastructure can explore targets in materials (e.g., carbon capture) and drug discovery.
• IP strategy: higher-throughput hypothesis testing can produce broader, earlier filings-expect more crowded patent spaces.

What to watch next (6-12 months)

• Evidence of generalization: peer-reviewed studies, preprints, or benchmarks showing consistent gains across domains.
• Partner announcements: named collaborations and early licensing deals for materials or therapeutics.
• Throughput metrics: cycle time per experiment, success rates, and the share of wins originating from AI vs. human teams.
• Regulatory paths: IND-enabling work for therapeutics or validation standards for new materials.

Risks and open questions

• Validation at scale: reproducibility and external replication for AI-proposed candidates.
• Data provenance: managing biases from literature-trained models and ensuring clean, well-annotated lab feedback.
• Cost and safety: robotics, reagents, biosafety, and compute budgets that can keep pace with model appetite.
• Measuring "scientific superintelligence": clear, domain-relevant tests beyond headline competitions.

Industry context

AI-first labs are drawing interest across materials and pharma. Players like Orbital Materials and Isomorphic Labs are pursuing similar goals with different stacks, signaling a broader shift toward automated discovery.

Practical next steps for your lab

• Audit your design-make-test-learn loop and identify bottlenecks for automation or model assistance.
• Stand up a clean data layer: consistent schemas, assay metadata, and versioned protocols to enable trustworthy feedback.
• Start narrow: pick one high-value screen or assay, instrument it end-to-end, and quantify cycle-time and hit-rate gains.
• Revise IP playbooks: plan for faster filings and cross-functional review as candidate volume increases.

If you're building skills and workflows for AI-assisted research, explore curated training paths by job role at Complete AI Training.