The race to embed AI into drug development just got interesting. Big Pharma companies are no longer experimenting with LLMs in isolation—they're locking into strategic partnerships with the three major frontier labs, and the market share numbers tell a clear story. As of May 2026, there are 27 confirmed deals between major pharmaceutical players and OpenAI, Anthropic, and Google Gemini. But not everyone is splitting their bets equally.

Anthropic's Dominance

Claude has captured 52% of these partnerships—14 out of 27 deals—with the biggest names in pharma. This isn't a fluke. When you're dealing with billion-dollar clinical programs where hallucinations could cost lives or sink regulatory submissions, you want an AI that prioritizes safety and reliability over raw capability theater. Anthropic built their reputation on constitutional AI principles, and apparently that resonates with people who have to explain bad outcomes to the FDA.

OpenAI Holds Steady

OpenAI comes in second with 11 deals, representing 41% of the tracked partnerships. Six companies are hedging their bets here, partnering with both OpenAI and Anthropic rather than going all-in on a single provider. But the most eye-popping number belongs to Merck's deal with Google Gemini—$1 billion disclosed value, making it the largest single partnership in this dataset despite Gemini only accounting for two total deals.

Google's Surprising Lag

Speaking of which: two deals? For the company that has arguably the deepest healthcare data footprint on the planet through Verily, DeepMind's AlphaFold breakthroughs, and Google Health? Gemini should be dominating this space. Instead it's playing catch-up while Anthropic cleans up. Whether this reflects actual model performance concerns or just slower enterprise sales cycles remains to be seen, but it's a gap that should concern everyone in Mountain View.

GSK Goes Rogue

The most interesting outlier is GSK, which has decided to tell the AI labs to kick rocks entirely and build their own system called JulesOS with internal AI/ML engineers. Their reasoning cuts straight to the heart of what makes Pharma different from other industries: clinical-risk hallucinations that could compromise trial integrity, safeguarding multi-billion dollar pre-clinical IP from third-party access, and unlocking proprietary functional genomics datasets that public LLMs literally cannot see. GSK becomes a natural experiment—an active control group against which we can measure whether build-vs-buy actually matters in drug discovery.

Where the Rubber Meets the Road

The functional distribution of these deals reveals where Pharma thinks LLMs are actually useful today versus vaporware. A stunning 82% touch Research and Discovery in some form, with Clinical Development as the next most common area. Manufacturing and CMC (Chemistry, Manufacturing, and Controls) barely registers despite years of industry chatter about AI-driven production optimization. Either that implementation is happening silently without public announcements, or Pharma's quality systems aren't ready to trust LLMs on the factory floor.

Key Takeaways

  • Anthropic/Claude dominates with 52% market share among tracked partnerships (14 deals)
  • OpenAI holds second place with 11 deals (41%), while Google Gemini surprisingly has only two despite its healthcare data advantages
  • Merck's $1B deal with Gemini is the largest disclosed value in the dataset, but an outlier
  • GSK is the lone wolf building JulesOS internally rather than partnering—serving as a live test case for build-vs-buy in Pharma AI
  • 82% of partnerships target Research and Discovery; Clinical Development comes second

The Bottom Line

The frontier AI race in pharma isn't about who has the flashiest demo or biggest benchmark numbers—it's about trust, IP protection, and regulatory tolerance. Anthropic gets this. OpenAI is competitive but not dominant. Google is sleeping while its healthcare data empire goes underutilized. And GSK's bet on JulesOS will either become a cautionary tale about reinventing wheels or proof that sometimes the best AI strategy is keeping your most valuable data far away from anyone else's servers.