AI drug discovery is moving from experimental research projects into the pipelines of major pharmaceutical companies. Chai Discovery has partnered with Bristol Myers Squibb (BMS) to use AI models for therapeutic antibody discovery, giving the drugmaker access to Chai’s molecular prediction and design technology as BMS builds what it describes as a continuously learning, AI-powered discovery system.
The pharmaceutical industry’s AI race is entering a more demanding phase. Predicting the structure of a molecule is one challenge; designing a therapeutic molecule that actually performs as intended in a biological system is another.
Chai Discovery is trying to close that gap with AI models designed to predict and engineer molecular interactions. Its latest pharmaceutical collaboration, with Bristol Myers Squibb, puts those capabilities inside the R&D organization of one of the world’s largest drugmakers.
Under the collaboration, BMS will use Chai’s AI models and platform capabilities—including molecular folding and design models—to support the discovery of antibody candidates across its portfolio. The companies have not disclosed financial terms or specific therapeutic targets.
The partnership is significant less because another pharmaceutical company is experimenting with AI and more because of where Chai’s technology sits in the drug-discovery workflow.
Chai describes its platform as a computer-aided design suite for molecules. Its models learn patterns underlying biochemical structure and molecular interaction, allowing scientists to predict structures and generate new biomolecules according to specified design criteria.
That is a different proposition from using a general-purpose large language model to summarize research papers or automate administrative work. The models are being applied to the physical behavior of proteins and other biological molecules.
From prediction to molecular design
Chai’s technology first gained attention with Chai-1, a multimodal foundation model for predicting molecular structures. The company subsequently expanded into de novo protein and antibody design.
Its current platform supports antibody formats including monoclonal antibodies, VH-VL and VHH designs, while allowing researchers to specify targets, epitopes and other molecular constraints. Chai says its design system is intended to reduce dependence on high-throughput screening by generating candidates for experimental characterization.
That distinction matters.
Traditional drug discovery often involves generating or screening large numbers of molecules and experimentally determining which ones have desirable properties. AI can shift some of that search into the computational domain, potentially narrowing the number of candidates that need to reach the laboratory.
But “potentially” is doing important work here. AI-generated molecules still have to be experimentally tested, optimized and ultimately subjected to the extensive preclinical and clinical processes required for a medicine.
The value proposition is therefore not that AI eliminates biology. It is that better computational predictions could make the experimental cycles more selective.
Why antibodies are an important test case
Antibodies are particularly attractive for AI-driven design because their therapeutic activity depends heavily on molecular structure and interactions with specific targets.
An effective antibody must recognize the right target and often the right epitope, while also meeting other requirements related to developability, specificity and biological behavior.
Chai’s platform is designed to incorporate those types of constraints into molecular design. Its product materials say researchers can specify an epitope, antibody format and target structure, including challenging targets such as membrane proteins.
For BMS, the attraction is scale.
The pharmaceutical company already has substantial expertise in antibody discovery and development. Its AI strategy is broader than any single vendor relationship: BMS says it is applying AI and machine learning across R&D and pursuing a “predict first” approach to selecting targets and designing molecules.
The Chai collaboration adds another specialized model layer to that ecosystem.
BMS is building a broader AI discovery stack
The partnership should also be viewed alongside BMS’s other AI initiatives.
In May 2026, BMS announced a strategic agreement with Anthropic to deploy Claude Enterprise across research, clinical development, manufacturing, commercial and corporate functions. The company described the move as a transition toward agentic AI connecting employees, systems and institutional knowledge across the organization.
The Chai collaboration addresses a very different layer.
Anthropic’s models are general-purpose AI systems that can support knowledge work and workflows. Chai’s models are specialized around molecular structure, interaction and design. Together, these developments illustrate how pharmaceutical companies are assembling AI stacks rather than looking for a single model to handle every R&D problem.
BMS has explicitly described its approach as combining human expertise with computational methods. Its research strategy uses AI to help identify targets based on causal human biology, match therapeutic modalities to mechanisms and prioritize molecules.
That human-machine combination remains important because drug discovery is a high-stakes domain where computational predictions can inform decisions but cannot substitute for experimental evidence.
Chai’s growing pharmaceutical customer base
BMS is also joining an expanding group of major biopharmaceutical companies working with Chai.
The company has announced relationships with Pfizer, Eli Lilly, Novartis and argenx, among others. Chai says its models are being deployed by pharmaceutical companies to support drug-discovery programs.
The Pfizer agreement announced in June, for example, gave the company access to Chai’s then-latest model and a custom model incorporating Pfizer’s proprietary data and workflows.
That model of collaboration points toward a potentially important competitive advantage in AI drug discovery: proprietary biological data.
Foundation models may provide the general capabilities, but pharmaceutical companies possess enormous collections of experimental results, assay data and historical development information. Connecting those datasets to specialized AI models could allow companies to create systems tailored to their own research environments.
Chai’s platform is consequently competing not only on model performance but also on how effectively its technology fits into pharmaceutical R&D infrastructure.
The industry’s bigger problem remains unsolved
The commercial opportunity is enormous because drug discovery remains slow, expensive and uncertain.
BMS says only about 10% of investigative medicines ultimately become approved therapies, illustrating the attrition problem confronting pharmaceutical R&D.
AI could improve that equation by helping researchers identify stronger candidates earlier, but there is no guarantee that computational success translates into clinical success.
This is where enterprise adoption becomes more complicated than simply purchasing an AI platform.
Pharmaceutical R&D teams will need to assess model accuracy, experimental validation, reproducibility, data provenance, intellectual-property protections, integration with laboratory workflows and the ability to learn from new experimental results.
Chai’s pitch is ultimately about turning molecular discovery into something closer to engineering: specify desired properties, generate candidate designs, test them experimentally and feed the resulting evidence back into the discovery process.
BMS is now adding that approach to a broader AI strategy spanning target discovery, molecule design and development.
The more consequential story is not that AI has solved drug discovery. It has not.
It is that pharmaceutical companies are increasingly treating specialized AI models as core components of their scientific infrastructure. If those systems can shorten the cycle between biological hypothesis, molecular design and experimental validation, they could change where and how drug candidates are discovered.
For Chai Discovery, the BMS collaboration is another validation of that model. For the pharmaceutical industry, it is another sign that AI drug discovery is moving from a promising research category toward an enterprise technology layer.
Market Landscape
The AI drug discovery market is increasingly separating into several technology layers: foundation models for biological structure, generative models for molecule design, AI-powered target identification, laboratory automation and systems that connect computational predictions with experimental data.
Chai Discovery is focused primarily on molecular design and prediction. Competitors and adjacent platforms include Isomorphic Labs, Recursion, Insilico Medicine, Generate and AI capabilities developed internally by pharmaceutical companies.
The competitive question is shifting from whether AI can predict molecular structures to whether models can consistently generate candidates with experimentally useful properties.
That is a much harder benchmark.
Chai’s own platform reports measurable success rates for antibody designs, but those are company-reported results and should not be treated as equivalent to clinical validation.
Meanwhile, BMS’s strategy illustrates the likely enterprise model: combine specialized AI models with proprietary data, internal scientific expertise and laboratory validation rather than relying on an autonomous AI system to make drug-development decisions.
Top Insights
- Bristol Myers Squibb will use Chai Discovery’s molecular AI models to support antibody discovery, adding specialized generative design capabilities to its broader AI R&D strategy.
- Chai’s technology moves beyond molecular structure prediction toward de novo antibody design, allowing researchers to computationally generate candidates around targets, epitopes and biological constraints.
- BMS is building a continuously learning discovery system that combines proprietary data, AI models, experimental evidence and scientific expertise across its research organization.
- The partnership reflects a wider pharmaceutical shift toward specialized AI models, with Pfizer, Eli Lilly, Novartis and other drugmakers also adopting Chai technology.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
