GSK has widened its partnership with Chai Discovery, the AI-native biotech focused on protein and molecule design, after the pharmaceutical company’s own scientists independently tested Chai’s AI models before committing to the expanded arrangement. The sequencing is notable: rather than simply licensing access to Chai’s technology on the strength of its published results, GSK validated the models internally first — a more cautious approach that stands out amid a broader wave of AI drug discovery deals struck largely on the basis of a partner’s own reported performance claims.
Why internal validation matters
Much of the AI drug discovery dealmaking boom over the past two years has relied heavily on AI biotech companies’ own published benchmarks and case studies to demonstrate their models’ effectiveness, benchmarks that outside researchers have sometimes struggled to independently reproduce. GSK’s decision to have its own scientists test Chai’s models before expanding the relationship offers a more rigorous form of validation than many of the headline-grabbing, deal-value-first partnerships that have characterized the sector, and could set a precedent other large pharmaceutical companies follow before committing to new AI partnerships.
What Chai Discovery actually builds
Chai Discovery has positioned itself at the intersection of protein structure prediction and generative molecule design, building AI models aimed at predicting how proteins fold and interact, and then designing drug candidates optimized for those interactions. The approach sits adjacent to, but distinct from, the foundational work done by DeepMind’s AlphaFold in protein structure prediction, with Chai focused more specifically on translating structural prediction into actionable drug design.
The broader pattern of pharma skepticism
GSK’s validate-first approach reflects a maturing phase in the pharma industry’s relationship with AI drug discovery vendors, following an earlier period where many large pharmaceutical companies entered AI partnerships somewhat speculatively, drawn by the scale of potential upside rather than rigorous internal diligence. As the sector has grown — AI drug discovery licensing deals reached $47.1 billion in the first half of 2026 alone — some pharmaceutical companies have grown more disciplined about distinguishing AI partners with genuinely validated technology from those primarily differentiated by marketing and headline deal values.
Pfizer’s parallel move
GSK is not alone in partnering with Chai Discovery — Pfizer separately signed a licensing agreement with the company for AI drug discovery work, suggesting Chai has emerged as one of the more sought-after AI biotech partners among large pharmaceutical companies specifically because of its structural biology and protein design capabilities. Having multiple major pharma partners simultaneously also gives outside observers more opportunity to compare notes, informally, on how well Chai’s technology performs across different disease areas and target types.
The case for caution anyway
Even with GSK’s internal validation, drug discovery experts caution that successfully predicting protein structure and designing plausible candidate molecules is a meaningfully different and earlier-stage achievement than demonstrating clinical safety and efficacy in humans — the stage where the vast majority of drug candidates, AI-discovered or otherwise, ultimately fail. Internal validation of the AI models themselves doesn’t eliminate that much larger, later-stage risk inherent to all drug development.
What it means next
GSK’s expanded, validation-first partnership with Chai Discovery may offer a more durable model for pharma-AI collaboration than deals struck primarily on reported benchmarks, and its progress will likely be watched by other large pharmaceutical companies as a test case for how much internal scientific diligence is worth investing before expanding AI drug discovery partnerships. Whether that more rigorous approach translates into better eventual clinical trial outcomes for the resulting candidates remains, like the rest of the AI drug discovery field, an open question still years from resolution.
Photo: Mikhail Nilov / PEXELS via Pexels
