Eli Lilly and Company has entered into a collaboration with Chai Discovery to deploy the biotech startup’s frontier AI platform for biologics drug design, including the development of a custom model trained on Lilly’s proprietary data. The partnership aims to accelerate the discovery of novel therapeutic candidates across multiple targets using zero-shot antibody design technology, marking a high-profile validation of Chai’s generative molecular modeling suite.

What this signals about AI’s next frontier in biologics: customized model training with internal data

While AI has made visible inroads across small molecule discovery and protein structure prediction, the biologics space has posed steeper challenges, particularly in generating functionally viable antibodies and other large molecules with therapeutic properties. Chai Discovery is attempting to shift that paradigm by offering what it claims is the first AI antibody design platform capable of achieving double-digit hit rates in experimental validation—without requiring fine-tuning on the target class.

The core differentiator in this collaboration is the creation of a bespoke AI model that will be trained on Eli Lilly and Company’s own preclinical and clinical data. Industry observers note that this is a significant move beyond licensing access to a generic AI toolkit. It reflects a growing recognition that enterprise-grade deployment of AI in biologics will require model architectures tailored to an organization’s unique assay formats, therapeutic areas, and sequence-function relationships.

By integrating its data assets with Chai’s model-building infrastructure, Eli Lilly and Company is attempting to improve both the specificity and developability of candidate molecules, and to do so earlier in the discovery process. If successful, the collaboration could reduce early-stage R&D cycle times and improve candidate selection fidelity before human testing.

Why AI-native antibody platforms are gaining traction after initial hesitancy

The antibody discovery market has traditionally been dominated by platform players such as AbCellera Biologics, Distributed Bio (acquired by Charles River Laboratories), and Adimab, all of which rely on high-throughput wet-lab screening. AI-first platforms like Chai Discovery are positioning themselves as alternatives or complements, using deep learning to search vast design spaces more efficiently than wet-lab methods alone.

However, the field has faced skepticism over generalizability, manufacturability of in silico designs, and real-world hit rates. According to clinicians tracking the space, only a handful of AI-designed antibodies have progressed past in vitro evaluation to functional studies, and fewer still have reached clinical development.

Chai-2, the company’s current platform iteration, reportedly achieves hit rates high enough to warrant head-to-head comparisons with traditional discovery workflows. The collaboration with Eli Lilly and Company will test these claims at scale, offering a real-world testbed across multiple programs with internal validation endpoints.

What differentiates Chai Discovery’s approach from generic protein language models

Unlike open-source protein models such as ESMFold or general-purpose LLMs applied to sequences, Chai Discovery’s platform appears to be designed from the ground up to handle antibody structure-function optimization, epitope specificity, and developability constraints. The startup claims it can produce molecules with drug-like properties, not just binding affinity.

In-house capabilities include zero-shot generation (producing viable candidates without training on similar targets), integrated filtering for manufacturability and biophysical properties, and loop modeling techniques to improve structural realism. Industry analysts point out that these elements are critical for translation into therapeutic candidates, especially when targeting difficult epitopes or avoiding immunogenicity.

The creation of a custom model trained on Eli Lilly and Company’s datasets further deepens this differentiation. Rather than relying solely on transfer learning from public protein databases, the Chai–Lilly collaboration will build a closed-loop system informed by Lilly’s own successes and failures across multiple modalities.

What this reveals about pharma’s evolving AI procurement strategy

This agreement also reflects a broader shift in how pharmaceutical companies are approaching AI partnerships. Rather than acquiring or building in-house AI capabilities for every function, major players like Eli Lilly and Company are increasingly opting for modular alliances—embedding frontier platforms like Chai Discovery within existing R&D architectures but customizing them to proprietary workflows.

Unlike some prior biotech–big pharma AI deals, this collaboration is not structured as an equity investment or acquisition. Instead, it is a platform-access and customization agreement, allowing Chai Discovery to remain independent and continue building its generalizable suite while tailoring specific models to Lilly’s needs. This structure suggests growing comfort with outsourcing AI innovation without ceding core IP control.

Regulatory watchers note that this strategy could align well with emerging frameworks on explainability and model documentation, particularly in biologics IND submissions. If Chai’s platform enables traceable design decisions and testable hypotheses, it may reduce regulatory risk compared to “black box” approaches.

Where adoption friction still looms: validation, scalability, and integration hurdles

Despite the enthusiasm, key hurdles remain. AI-generated molecules still require extensive wet-lab validation, and the field lacks a clear benchmark for what constitutes an acceptable hit rate or developability success. Manufacturing concerns, particularly for unusual or de novo sequences, could delay translation from discovery to clinical-grade production.

Moreover, integration of AI design outputs into large pharmaceutical R&D pipelines is non-trivial. Bioinformatics infrastructure, assay compatibility, and workflow harmonization will all influence how quickly Eli Lilly and Company can deploy Chai’s platform across programs.

As reimbursement and payer scrutiny tighten across biologics, success will hinge not only on speed but also on producing candidates with competitive cost of goods, IP defensibility, and clinical differentiation. Clinicians following the space believe that AI platforms must ultimately demonstrate not just discovery efficiency but downstream value.

What industry observers will watch next as Chai’s valuation climbs

Chai Discovery’s $1.3 billion valuation following its Series B round in December 2025 puts it in rarefied air among early-stage AI biotech companies. The firm has now raised close to $230 million, and counts OpenAI, Thrive Capital, and Menlo Ventures among its backers. But with valuation expectations high, investor scrutiny will turn toward proof of value delivery.

For Chai, the Lilly partnership is both a strategic win and a proving ground. The depth of the collaboration, including custom model development, will serve as a case study for whether AI-native biologics platforms can truly outperform traditional discovery methods in real-world pharmaceutical settings.

If the company can demonstrate rapid candidate identification, improved quality, and meaningful reductions in early R&D timelines, it could set a new standard for biologics innovation workflows. Conversely, delays, failures to translate in silico designs, or limited platform extensibility could dampen broader enthusiasm.

As the field of AI-driven biologics design matures, this collaboration will be a bellwether—not just for Chai, but for the credibility of generative biology as a foundation for 21st-century drug development.

  AI drug discovery, antibody design, biologics, Chai Discovery, Chai-2 platform, custom AI models, Eli Lilly and Company, frontier AI, generative biology, IND submissions