Q1 2026 was a turning point for artificial intelligence in pharma. But it was not about the number of deals, although that number was significant. Doubling to thirty-six major AI partnerships in Q1 2026 versus eighteen in Q1 2025.
It is what sits behind those deals that really matters.
We are seeing AI activity that spans multiple pharma domains
simultaneously: drug discovery, diagnostics, clinical development, and now
embedding itself into the deeper plumbing of manufacturing and commercial
functions. As well as breadth, we are seeing depth. AI integrating itself
profoundly into the pharma operating model rather than sitting alongside it as
an innovation experiment.
The deals announced between January and March were not pilots. They were
architectural decisions. Multi-year terms. Billion-dollar values. Third and
fourth extensions on relationships that were already mature. And in at least
one case, an acquisition to a recent partnership. Taken together they tell a
story about an industry that has stopped asking whether AI works and started
asking how to build around it permanently.
The Lilly Blueprint
In Q1 no company was a better exemplar than Eli Lilly. Eight
significant AI partnerships in a single quarter is extraordinary by any
measure, but the structure of those deals is more revealing than the volume.
Lilly literally bet all its chips on
AI-enabled drug discovery this quarter. The first company to deploy NVIDIA's
DGX B300 AI supercomputer, co-investing over $1 billion in a co-innovation lab
alongside it. Capital expenditure at this scale does not get unwound at the
next budget review.
The Insilico Medicine collaboration, worth up to $2.75 billion, is
not a discovery bet on a single programme. It is a platform commitment across
multiple targets and therapeutic areas with milestone-driven payments tied to
pipeline progress.
Then there is TuneLab. Lilly's federated AI model network, trained
on over a billion dollars of proprietary data, expanded across Schrödinger,
Revvity and BigHat in Q1. The logic is self-reinforcing: more
partners generate more diverse training data, which improves the models, which
attract more partners. Lilly is not just building internal capability.
It is positioning itself as the connective tissue of an industry-wide AI
discovery ecosystem, opening the doors to collaborators who can access Lilly
data across a federated interface.
Alongside this, partnerships with Chai Discovery for biologics, InduPro
for oncology membrane interactomics, and Fauna Bio for obesity target
discovery, inspired by how hibernating animals regulate weight and metabolism,
round out a portfolio that jointly spans biology, chemistry, compute and
platform infrastructure.
Bristol Myers Squibb: Four Deals, Three Domains, One
Quarter
Where Lilly went deep on discovery infrastructure, BMS went
broad. Four distinct AI partnerships spanning drug discovery, diagnostics and
clinical development in a single quarter is a portfolio strategy, not a series
of one-off decisions.
Immunai brought AI-driven immune analysis
into BMS clinical programmes, generating insights that inform patient
stratification and feed back into discovery decisions. Microsoft brought
AI radiology into lung cancer detection, embedding FDA-cleared algorithms into
clinical workflows already operating across most US hospitals. Evinova
brought agentic AI into trial design, enabling BMS to scenario-plan and
optimise protocols before studies begin, shifting from running more trials to
running better ones. And Faro brought AI upstream into protocol drafting
and benchmarking, turning trial design into structured, optimisable
infrastructure rather than a manual drafting exercise.
Each deal targets a different bottleneck. Together they suggest a
deliberate attempt to embed AI across the entire development cycle rather than
concentrate it in one area.
The Structure Signal
The rest of the map tells the same story from different angles.
Daiichi Sankyo ran two AI deals
with a specific strategic logic: as antibody-drug conjugates become more
competitive, patient selection becomes the differentiator. BostonGene
applied AI-powered digital twin models to refine which patients are most likely
to respond to Daiichi's ADCs. Tempus brought in its PRISM2
foundation model to strengthen biomarker discovery and stratification across
the same pipeline. Two deals, two tools, one objective: getting the right
patients into the right trials.
Merck ran three deals: a genomics platform
with Quotient Therapeutics for IBD target identification, a virtual cell
collaboration with Mayo Clinic running AI models inside Mayo's secure
environment on multimodal clinical data, and a Tempus partnership for
precision oncology. GSK partnered with Noetik on virtual cell
modelling for tumour biology and with Helix for population-scale
genomics. Servier signed two major generative AI deals in a single week,
Insilico and Iktos, with Pierre Fabre following Iktos
into a separate collaboration shortly after.
Bayer and Cradle committed to a
three-year protein engineering collaboration. Boehringer Ingelheim
partnered with Variant Bio on kidney disease target discovery and with Brainomix
on AI imaging as a co-primary endpoint in a Phase III pulmonary fibrosis trial.
Using AI output as regulatory evidence is a significant escalation from using
AI as an internal decision support tool, and a signal of how far that
particular partnership has matured. Sanofi extended its CytoReason
relationship for the third time and partnered with Earendil Labs on an
autoimmune bispecific programme in a deal that could reach $2.5 billion.
And then there is AstraZeneca and Modella AI. AZ's shotgun
wedding with its medical AI imaging partner, acquired outright just months
after initiating the partnership, is perhaps the best poster child of an
industry that wants to build AI capabilities into its operating system rather
than access them at arm's length. Partnership was no longer enough.
What This Quarter Actually Means
The pilot phase of pharma AI had a recognisable signature: short
engagements, narrow scope, single programmes, cautious language around proof of
concept. Q1 2026 had almost none of that signature.
What it had instead was duration, scale, specificity and repetition.
Companies returning to partners they already trusted and committing more.
Companies building compute infrastructure they will depreciate over years.
Companies embedding AI into clinical workflows as standard operating procedure.
Companies acquiring rather than partnering when the strategic fit was clear
enough.
The industry did not suddenly discover that AI works in Q1 2026. What
happened is that enough internal evidence accumulated, across enough
programmes, enough organisations, enough functions, that the case for permanent
infrastructure finally outweighed the case for continued experimentation.
Thirty-six deals. Double the prior year. But that number is still the
least interesting thing about this quarter.
