Two UK Shops, One German Buyer: Merck's $1.3B AI Discovery Bet Tests the Phase III Clock

Pharma's AI discovery narrative has oscillated between "40% faster preclinical timelines" and a harder question: can AI-origin molecules survive Phase III? On August 4, 2026, Merck KGaA answered with capital, not slogans—signing parallel AI drug discovery deals with UK's Exscientia and BenevolentAI worth up to a combined $1.3 billion if milestones hit.

The agreements are discovery-stage alliances, not approvals. They nonetheless mark one of the largest single-week AI biotech partnership bursts from a major pharma incumbent—and they arrive four days after the EU AI Act's Article 50 transparency obligations took effect on August 2.

Deal structure: upfront cash, discovery milestones, oncology-neuro-immuno focus

Exscientia (Oxford): $20 million upfront; up to $113 million in discovery-phase milestones; total deal value up to $674 million across three initial projects in oncology, neurology, and immunology.

BenevolentAI (London): "Low double-digit million dollar" upfront (pharmaphorum's reporting); up to $594 million including discovery, development, and commercial milestones across three initial programmes in the same therapeutic areas.

In both deals, the AI companies deliver small-molecule candidates; Merck handles further preclinical and clinical development. Additional targets can be nominated later.

Merck chief medical officer Danny Bar-Zohar said the alliances create an "end-to-end" AI platform to "fast-track the development of new and truly innovative candidates," and signaled more internal and external investment is coming.

Exscientia CEO Prof Andrew Hopkins said the partnership targets "some of the hardest drug design challenges in cancer and immunology" while leveraging Merck's oncology and neuroinflammation expertise.

BenevolentAI CEO Joanna Shields highlighted an "extensively validated approach of combining AI, molecular biology, medicinal chemistry, and in vivo pharmacology" to discover candidates "in fewer cycles."

Merck's in-house stack: Addison and SYNTHIA

These are not Merck's first AI bets. The company already runs Addison, an in-house engine using generative methods, virtual screening, and hit-to-lead optimization tools. A separate platform, SYNTHIA, predicts synthetic feasibility for in silico designs—addressing the classic AI discovery failure mode of molecules that look brilliant and cannot be made.

The Exscientia and BenevolentAI deals layer external discovery capacity onto that internal stack rather than replacing it—a pattern large pharmas increasingly use: build core infrastructure, buy specialized velocity where pipelines thin.

Context: validation beats acceleration in 2026

Merck's deals land in a year industry analysts describe as a "prove it" moment for AI drug discovery.

Drug Target Review published predictions in early 2026 noting that no AI-originated drug had yet received regulatory approval, while multiple AI-designed candidates approached Phase III readouts. Recursion and Genentech separately announced August 6 that the first validated neuroscience target from their AI phenomics map entered co-development—a discovery milestone, not a marketed therapy.

The EU AI Act's high-risk system obligations for many drug-development AI tools were deferred by the July 2026 AI Omnibus to December 2027 (Annex III) and August 2028 (product-embedded medical AI). Article 50 transparency duties, however, are live now—relevant for patient-facing or content-generating tools pharma deploys in EU markets.

Merck's Monday deals are classic early-discovery economics: modest upfronts, large optionality. They do not by themselves answer whether AI compression of hit identification translates into higher Phase III success rates.

What to watch next

Milestone triggers: Exscientia and BenevolentAI earn discovery payments only if nominated targets and candidates meet contractual gates—watch SEC filings and partner earnings for program names.

Clinical readouts: Industry observers continue to frame 2026–2027 Phase III results from earlier AI discovery alliances as the definitive test of the sector's thesis.

Regulatory mapping: Pharma legal teams must separate Article 50 transparency compliance (now) from deferred high-risk conformity work (2027–2028)—a timeline split Goodwin and other firms flagged in August 2026 client alerts.

Merck's twin deals are best read as supply-chain diversification for molecular novelty: buy AI throughput where internal Addison/SYNTHIA pipelines want more shots on goal in oncology, neurology, and immunology. The science case remains open until a Merck-sponsored AI-origin candidate produces a registrational dataset worth citing.

Until then, $1.3 billion in headline potential is a bet that discovery acceleration still has economic value—even in a year defined by clinical validation anxiety.

### Sources

• pharmaphorum — Merck puts $1.3bn into AI via Exscientia, BenevolentAI deals (August 4, 2026)
• Drug Target Review — AI in drug discovery: predictions for 2026 (2026)
• GEN — Recursion Partners with Genentech to Advance First Validated Neuro Target (August 6, 2026)