GSK announced an expanded AI drug discovery collaboration with Chai Discovery on October 9, 2026, after the pharmaceutical giant’s own scientists independently validated the San Francisco-based AI biotech’s models in its laboratories rather than taking the startup’s published performance claims at face value. The expansion gives GSK broader access to Chai’s protein folding and design platform across its drug discovery pipeline, joining a growing list of pharmaceutical companies — including Bristol Myers Squibb, Eli Lilly, Novartis and Pfizer — that have struck deals with the AI-native biotech over the past year.
Why GSK tested before it bought
What distinguishes this deal from many AI-pharma partnerships announced in 2026 is the sequencing: GSK ran its own wet-lab evaluation of Chai’s models before expanding the relationship, rather than licensing access first and validating later. According to people familiar with the arrangement, GSK’s scientists generated designs zero-shot against a panel of GSK-selected protein targets and found that Chai’s models produced binders against every target tested, a result that reportedly gave GSK’s internal discovery teams confidence to widen the collaboration beyond an initial evaluation phase.
What zero-shot design actually means
Chai’s platform is built around generative models that can propose candidate binding molecules against a protein target without first training on examples specific to that target, a capability the company calls zero-shot design. An earlier preprint linked one of Chai’s models to a 16% hit rate in fully de novo antibody design across 52 different targets, a figure that, while well short of guaranteed success, is notable in a field where traditional wet-lab screening against novel targets can take months and frequently fails to produce any viable candidates at all.
Terms left undisclosed
Neither company has disclosed the financial terms of the expanded agreement, including the size of any upfront or milestone payments, which targets or disease areas are covered, or how many discovery programs will use Chai’s models under the new arrangement. That opacity is common across AI-pharma deals this year, where headline deal values — when disclosed at all — often represent the maximum potential total across years of milestones rather than money that has actually changed hands.
Chai’s expanding pharma roster
The GSK expansion comes on the heels of Chai Discovery closing a $400 million Series C in July 2026 at a $3.8 billion valuation, giving the company substantial capital to pursue additional pharma partnerships and build out its modeling infrastructure. In July, Novartis separately gained access to Chai’s latest models, including Chai-3, to support antibody discovery across several of its own therapeutic programs, suggesting Chai is positioning itself as an infrastructure layer serving multiple competing pharmaceutical companies simultaneously rather than an exclusive partner to any single one.
The validation gap that remains
GSK’s decision to run its own wet-lab confirmation before expanding the deal reflects a broader shift in how pharmaceutical companies are approaching AI drug discovery partnerships after an initial wave of deals struck largely on the basis of vendor-reported benchmarks. Industry observers have noted that this kind of independent, pre-expansion validation step — while still short of the clinical trial data that will ultimately determine whether any of these platforms produce approved drugs — is becoming a more common gatekeeping practice as pharma companies grow more cautious about AI vendors’ self-reported hit rates.
What comes next
For GSK, the expanded Chai partnership adds another computational design tool to a drug discovery operation already juggling multiple AI platforms and in-house capabilities. Whether the zero-shot binders Chai’s models generate during GSK’s expanded program advance into preclinical candidates, let alone clinical trials, will likely take one to two years to become clear — a timeline that will test whether GSK’s upfront wet-lab validation translates into a meaningfully higher success rate than deals struck on benchmark data alone. The broader pharmaceutical industry will also be watching how GSK structures its internal workflow around Chai’s outputs, since even a validated AI model only creates value if discovery teams actually integrate its suggestions into existing pipelines rather than treating them as a parallel, rarely-consulted data source — a cultural and operational adoption challenge that has proven just as difficult for some pharma companies as the underlying science. GSK has not said publicly how it plans to measure the expanded collaboration’s success internally, whether through a target number of validated binders, advancement of a specific candidate into preclinical testing, or some other milestone, leaving outside observers without a clear marker to track progress against until the company next updates on its AI-enabled discovery programs.
Photo: Lefteris Betsis / PEXELS via Pexels
